{"id":85453,"date":"2026-07-27T15:41:17","date_gmt":"2026-07-27T07:41:17","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/85453.html"},"modified":"2026-07-27T15:41:17","modified_gmt":"2026-07-27T07:41:17","slug":"pytorch%e3%80%90%e7%ac%ac%e4%ba%8c%e9%83%a8%e5%88%86%e3%80%91","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/85453.html","title":{"rendered":"pytorch\u3010\u7b2c\u4e8c\u90e8\u5206\u3011"},"content":{"rendered":"

\u7b2c\u4e09\u7ae0<\/h2>\n

\u60a8\u53ef\u4ee5\u5728\u6b64\u5904\u67e5\u770b\u524d\u4e8c\u7ae0\u7684\u5185\u5bb9<\/p>\n

\u6570\u636e\u7ba1\u9053<\/h3>\n

\u6211\u4eec\u4f7f\u7528MNIST\u7684\u6570\u636e\u96c6，\u8fd9\u662f\u6807\u51c6\u7684\u6df1\u5ea6\u5b66\u4e60\u5165\u95e8\u65b9\u5f0f\u3002\u53ef\u662f\u5982\u679c\u6211\u4eec\u5e0c\u671b\u6784\u5efa\u7279\u5b9a\u529f\u80fd\u7684\u6a21\u578b，\u6bd4\u5982\u4e0d\u53ea\u662f\u8bc6\u522b\u6570\u5b57，\u800c\u662f\u8bc6\u522b\u82b1\u5349\u3001\u8bc6\u522b\u884c\u4eba\u3002\u6211\u4eec\u6216\u8bb8\u5c31\u8981\u7528\u5230\u81ea\u5df1\u51c6\u5907\u7684\u8bad\u7ec3\u96c6\u4e86，\u8fd9\u4e2a\u65f6\u5019，\u8bad\u7ec3\u96c6\u53ef\u80fd\u5e76\u4e0d\u6807\u51c6，\u6216\u8bb8\u5305\u542b\u7a7a\u503c\u6216\u8005\u6ca1\u6709\u6ca1\u6709\u610f\u4e49\u7684\u503c，\u800c\u8fd9\u4e9b\u503c\u4e00\u65e6\u4ea4\u7ed9\u6a21\u578b\u8bad\u7ec3，\u5c31\u4f1a\u5bfc\u81f4\u6a21\u578b\u65e0\u6cd5\u8bfb\u53d6\u6709\u6548\u6570\u636e\u800c\u9020\u6210\u6700\u7ec8\u7684\u8bad\u7ec3\u6548\u679c\u4e0d\u5c3d\u4eba\u610f\u7684\u60c5\u51b5，\u8fd9\u65f6\u5019\u5bf9\u4e8e\u6570\u636e\u7684\u5904\u7406\u5c31\u81f3\u5173\u91cd\u8981\u4e86<\/p>\n

\u5f53\u60a8\u62ff\u5230\u8bad\u7ec3\u96c6\u540e，\u60a8\u4e00\u822c\u9700\u8981\u505a\u4ee5\u4e0b\u7684\u9884\u5907\u5de5\u4f5c<\/p>\n

  • \u5c06\u56fe\u7247\u4e0e\u6807\u7b7e\u5173\u8054\u8d77\u6765\u3010\u56fe\u7247\u548c\u6807\u7b7e\u7684\u539f\u672c\u5173\u7cfb\u662f\u4ec0\u4e48\u6837\u7684\u5462？\u6807\u7b7e\u5c31\u85cf\u5728\u56fe\u7247\u7684\u540d\u79f0\u5f53\u4e2d？\u8fd8\u662f\u6807\u7b7e\u4e0e\u56fe\u7247\u7684\u5173\u7cfb\u5b58\u50a8\u5728\u4e00\u4e2a\u6587\u4ef6\u5f53\u4e2d？\u60a8\u9700\u8981\u5904\u7406\u5404\u79cd\u4e0d\u540c\u7684\u60c5\u51b5，\u5e76\u8f6c\u6362\u4e3a\u4e0a\u8ff0\u7684(data,labels)\u7684\u6807\u51c6\u683c\u5f0f\u8bad\u7ec3\u3011<\/li>\n
  • \u5c06\u56fe\u7247\u8f6c\u6362\u4e3a\u6b63\u786e\u7684\u683c\u5f0f（\u6b63\u786e\u7684\u5927\u5c0f，\u6b63\u786e\u7684\u683c\u5f0f，\u6b63\u786e\u7684\u6570\u636e\u7ed3\u6784），\u4ee5\u4fdd\u8bc1\u8bad\u7ec3\u5668\u53ef\u4ee5\u8bc6\u522b\u5e76\u6839\u636e\u8fd9\u4e9b\u6570\u636e\u8bad\u7ec3\u6a21\u578b<\/li>\n
  • \u60a8\u9700\u8981\u9ad8\u6548\u5730\u6279\u91cf\u52a0\u8f7d\u6240\u6709\u6570\u636e，\u63a7\u5236\u6bcf\u6b21\u52a0\u8f7d\u7684\u5927\u5c0f\u3010\u592a\u5c0f\u6a21\u578b\u8bad\u7ec3\u7684\u901f\u5ea6\u4f1a\u8f83\u6162（\u56e0\u4e3a\u6279\u6b21\u5f88\u591a，\u6a21\u578b\u4f1a\u82b1\u65f6\u95f4\u505a\u5927\u91cf\u91cd\u590d\u7684\u5de5\u4f5c），\u592a\u5927\u53c8\u4f1a\u5bfc\u81f4\u5185\u5b58\u7684\u5360\u7528\u8fc7\u8f7d（\u5927\u91cf\u8d44\u6e90\u52a0\u8f7d\u800c\u5bfc\u81f4\u7a7a\u95f4\u4e0d\u8db3）\u3011<\/li>\n

    \u6211\u4eec\u53ef\u4ee5\u5c06\u8fd9\u4e9b\u95ee\u9898\u5927\u81f4\u5206\u4e3a\u8bbf\u95ee\u95ee\u9898（\u6570\u636e\u52a0\u8f7d\u5668\u53ef\u5408\u6cd5\u8bbf\u95ee，\u5efa\u7acb\u56fe\u7247\u4e0e\u6807\u7b7e\u7684\u5173\u8054），\u8d28\u91cf\u95ee\u9898（\u8bad\u7ec3\u96c6\u7684\u8d28\u91cf，\u4f7f\u7528\u6b63\u786e\u7edf\u4e00\u7684\u683c\u5f0f）\u548c\u6548\u7387\u95ee\u9898（\u9ad8\u6548\u8fd0\u4f5c，\u9009\u62e9\u5408\u9002\u7684\u6279\u91cf\u52a0\u8f7d\u5927\u5c0f）<\/p>\n

    \u5728\u672c\u7ae0\u8282\u4e2d，\u6211\u4eec\u5c06\u4f7f\u7528oxFord\u82b1\u5349\u6570\u636e\u96c6\u8fdb\u884c\u5b66\u4e60<\/p>\n

    \u56fe\u50cf\u4e0b\u8f7d\u5730\u5740：https:\/\/www.robots.ox.ac.uk\/~vgg\/data\/flowers\/102\/102flowers.tgz<\/p>\n

    \u6807\u7b7e\u4e0b\u8f7d\u5730\u5740：https:\/\/www.robots.ox.ac.uk\/~vgg\/data\/flowers\/102\/imagelabels.mat<\/p>\n

    \u89e3\u51b3\u8bbf\u95ee\u95ee\u9898<\/h3>\n

    \u8d44\u6e90\u4e0b\u8f7d<\/h4>\n

    \u5728oxFord\u6570\u636e\u96c6\u4e2d，\u6807\u7b7e\u5b58\u50a8\u5728.mat\u6587\u4ef6\u4e2d，\u6240\u4ee5\u6211\u4eec\u9700\u8981\u624b\u52a8\u6784\u5efa\u8fd9\u6837\u7684\u5bf9\u5e94\u5173\u7cfb，\u786e\u4fdd\u6a21\u578b\u53ef\u4ee5\u6b63\u786e\u7684\u83b7\u53d6\u6211\u4eec\u7684\u56fe\u50cf\u2014\u2014\u6807\u7b7e\u6620\u5c04<\/p>\n

    \u5728\u6b64\u4e4b\u524d，\u6211\u4eec\u9700\u8981\u4e0b\u8f7d\u6570\u636e\u96c6\u548c\u6807\u7b7e，urllib\u5e93\u4e3a\u6211\u4eec\u63d0\u4f9b\u4e86\u8fd9\u6837\u7684\u652f\u6301<\/p>\n

    import<\/span> urllib
    \n# ……<\/span>
    \nurllib.<\/span>request.<\/span>urlretrieve(<\/span>"download_url"<\/span>,<\/span>"target_path"<\/span>)<\/span><\/p>\n

    \u6211\u4eec\u4f7f\u7528urllib\u4e2drequest\u6a21\u5757\u7684urlretrieve\u65b9\u6cd5\u4ece\u5bf9\u5e94url\u4e0b\u8f7d\u6587\u4ef6\u5230\u672c\u5730，\u7b2c\u4e00\u4e2a\u53c2\u6570\u662f\u4e0b\u8f7d\u8def\u5f84，\u7b2c\u4e8c\u4e2a\u53c2\u6570\u662f\u76ee\u6807\u8def\u5f84\u3010\u672c\u5730\u3011，\u683c\u5f0f\u4e3a\u6587\u4ef6\u6240\u5728\u76ee\u5f55\/\u6587\u4ef6\u540d\u79f0，\u8fd9\u91cc\u7684\u6587\u4ef6\u540d\u79f0\u6211\u4eec\u53ef\u4ee5\u76f4\u63a5\u83b7\u53d6\u4e0b\u8f7d\u6587\u4ef6\u7684\u540d\u79f0，\u800c\u76ee\u5f55\u5219\u9700\u8981\u6211\u4eec\u81ea\u5df1\u6307\u5b9a，\u5728\u8bfe\u7a0b\u5f53\u4e2d，\u6211\u4eec\u4f7f\u7528\u8fd9\u6837\u7684\u76ee\u5f55\u3010\u60a8\u4e5f\u53ef\u4ee5\u6784\u5efa\u81ea\u5df1\u7684\u76ee\u5f55\u89c4\u5219\u3011<\/p>\n

    — flowers
    \n |— img
    \n |— **imgfiles\u3010need download\u3011**
    \n |— **labelsfile\u3010need download\u3011**<\/p>\n

    \u6211\u4eec\u4f7f\u7528flowers\u4f5c\u4e3a\u6211\u4eec\u7684\u6570\u636e\u96c6\u6839\u76ee\u5f55，\u5c06\u6211\u4eec\u7684\u6807\u7b7e\u6587\u4ef6\u5b58\u50a8\u5728\u6839\u76ee\u5f55\u4e0b，\u540c\u65f6\u5728\u6570\u636e\u5e93\u6839\u76ee\u5f55\u4e0b\u65b0\u5efa\u4e00\u4e2aimg\u76ee\u5f55，\u6765\u5b58\u50a8\u4e0b\u8f7d\u7684\u56fe\u50cf\u8d44\u6e90<\/p>\n

    \u636e\u6b64，\u6211\u4eec\u53ef\u4ee5\u6784\u5efa\u56fe\u50cf\u6587\u4ef6\u7684\u8def\u5f84\u548cLabel\u6587\u4ef6\u7684\u8def\u5f84，os.path.join\u662f\u4e00\u4e2a\u5904\u7406\u8def\u5f84\u62fc\u63a5\u7684\u65b9\u6cd5，\u5b83\u80fd\u5e2e\u52a9\u6211\u4eec\u4f18\u96c5\u5730\u89e3\u51b3\u8fd9\u4e9b\u95ee\u9898<\/p>\n

    image_url =<\/span> "https:\/\/www.robots.ox.ac.uk\/~vgg\/data\/flowers\/102\/102flowers.tgz"<\/span>
    \nlabels_url =<\/span> "https:\/\/www.robots.ox.ac.uk\/~vgg\/data\/flowers\/102\/imagelabels.mat"<\/span>
    \nos.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span>"img"<\/span>,<\/span>os.<\/span>path.<\/span>basename(<\/span>image_url)<\/span>)<\/span> # \u62fc\u63a5\u56fe\u50cf\u8d44\u6e90\u76ee\u6807\u8def\u5f84<\/span>
    \nos.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span>os.<\/span>path.<\/span>basename(<\/span>image_url)<\/span>)<\/span> # \u62fc\u63a5label\u8d44\u6e90\u76ee\u6807\u8def\u5f84<\/span><\/p>\n

    \u6211\u4eec\u6709\u4e86\u8d44\u6e90\u4e0b\u8f7d\u8def\u5f84\u548c\u76ee\u6807\u8def\u5f84，\u63a5\u4e0b\u6765\u5c31\u53ef\u4ee5\u4f7f\u7528urlretrieve\u8fdb\u884c\u4e0b\u8f7d\u4e86<\/p>\n

    urllib.<\/span>request.<\/span>urlretrieve(<\/span>image_url,<\/span>os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span>"img"<\/span>,<\/span>os.<\/span>path.<\/span>basename(<\/span>image_url)<\/span>)<\/span>)<\/span>
    \nurllib.<\/span>request.<\/span>urlretrieve(<\/span>labels_url,<\/span>os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span>os.<\/span>path.<\/span>basename(<\/span>labels_url)<\/span>)<\/span>)<\/span><\/p>\n

    \u6211\u4eec\u6ce8\u610f\u5230image_url\u4e2d\u6587\u4ef6\u5e76\u4e0d\u662f\u76f4\u63a5\u4ee5\u56fe\u50cf\u6587\u4ef6\u7684\u683c\u5f0f\u5b58\u5728\u7684，\u4ed6\u662f\u4e00\u4e2atgz\u538b\u7f29\u5305，\u6211\u4eec\u9700\u8981\u4f7f\u7528\u89e3\u538b\u5de5\u5177\u89e3\u538b\u5230\u56fe\u7247\u8d44\u6e90\u6240\u5728\u8def\u5f84，\u8fd9\u6837\u540e\u7eed\u6a21\u578b\u624d\u80fd\u8bbf\u95ee\u6211\u4eec\u7684\u56fe\u7247<\/p>\n

    \u5728\u4e0b\u8f7d\u597d\u6587\u4ef6\u4e4b\u540e，\u6211\u4eec\u9700\u8981\u4ece\u76ee\u6807\u8def\u5f84\u521b\u5efa\u6587\u4ef6\u6d41\u3010\u4f7f\u7528tarfile.open\u3011，\u7136\u540e\u5bf9\u6587\u4ef6\u6d41\u7ba1\u63a7\u7684\u6587\u4ef6\u8fdb\u884c\u89e3\u538b\u64cd\u4f5c\u3010\u60a8\u4e0d\u7528\u624b\u52a8\u5904\u7406\u8fd9\u4e9b\u6b65\u9aa4，tg(your file stream).extractall(\u201cyour target path\u201d)\u65b9\u6cd5\u53ef\u4ee5\u5e2e\u60a8\u89e3\u51b3\u89e3\u538b\u7f29\u7684\u95ee\u9898\u3011<\/p>\n

    import<\/span> tarfile
    \n# ……<\/span>
    \nwith<\/span> tarfile.<\/span>open<\/span>(<\/span>os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span> "img"<\/span>,<\/span> "102flowers.tgz"<\/span>)<\/span>)<\/span> as<\/span> tg:<\/span>
    \n tg.<\/span>extractall(<\/span>os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span> "img"<\/span>)<\/span>)<\/span><\/p>\n

    \u60a8\u53ef\u4ee5\u7528\u51fd\u6570\u5c01\u88c5\u8fd9\u4e9b\u6b65\u9aa4，\u4ee5\u5728\u4ee3\u7801\u8fd0\u884c\u7684\u65f6\u5019，\u53ef\u4ee5\u76f4\u63a5\u4eceurl\u4e0b\u8f7d\u6587\u4ef6\u5e76\u8fd0\u884c\u60a8\u6240\u505a\u7684\u8fd9\u4e9b\u9884\u5904\u7406\u64cd\u4f5c\u8ba9\u5b83\u6210\u4e3a\u6570\u636e\u52a0\u8f7d\u5668\u53ef\u4ee5\u76f4\u63a5\u8bbf\u95ee\u7684\u6587\u4ef6\u3002<\/p>\n

    def<\/span> download_dataset<\/span>(<\/span>)<\/span> –<\/span>><\/span> None<\/span>:<\/span>
    \n image_url =<\/span> "https:\/\/www.robots.ox.ac.uk\/~vgg\/data\/flowers\/102\/102flowers.tgz"<\/span>
    \n labels_url =<\/span> "https:\/\/www.robots.ox.ac.uk\/~vgg\/data\/flowers\/102\/imagelabels.mat"<\/span><\/p>\n

    os.<\/span>makedirs(<\/span>"flowers"<\/span>,<\/span>exist_ok=<\/span>True<\/span>)<\/span>
    \n # \u9632\u6b62\u91cd\u590d\u4e0b\u8f7d<\/span>
    \n if<\/span> not<\/span> os.<\/span>path.<\/span>exists(<\/span>os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span>"img"<\/span>,<\/span>"102flowers.tgz"<\/span>)<\/span>)<\/span>:<\/span>
    \n urllib.<\/span>request.<\/span>urlretrieve(<\/span>image_url,<\/span> os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span>"img"<\/span>,<\/span>os.<\/span>path.<\/span>basename(<\/span>image_url)<\/span>)<\/span>)<\/span>
    \n with<\/span> tarfile.<\/span>open<\/span>(<\/span>os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span> "img"<\/span>,<\/span> "102flowers.tgz"<\/span>)<\/span>)<\/span> as<\/span> tg:<\/span>
    \n tg.<\/span>extractall(<\/span>os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span> "img"<\/span>)<\/span>)<\/span>
    \n # \u9632\u6b62\u91cd\u590d\u4e0b\u8f7d<\/span>
    \n if<\/span> not<\/span> os.<\/span>path.<\/span>exists(<\/span>os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span>"imagelabels.mat"<\/span>)<\/span>)<\/span>:<\/span>
    \n urllib.<\/span>request.<\/span>urlretrieve(<\/span>labels_url,<\/span> os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span>os.<\/span>path.<\/span>basename(<\/span>labels_url)<\/span>)<\/span>)<\/span><\/p>\n

    \u521b\u5efaDataSet\u7c7b<\/h4>\n

    \u6211\u4eec\u77e5\u9053\u6570\u636e\u96c6\u548c\u6807\u7b7e\u5b58\u653e\u7684\u4f4d\u7f6e，\u53ef\u662f\u6a21\u578b\u8be5\u5982\u4f55\u53bb\u5bfb\u627e\u5b83\u5462，\u60a8\u8fd8\u8bb0\u5f97\u6211\u4eec\u4e4b\u524d\u63d0\u5230\u7684SomeDataSet\u5417，\u4ed6\u5e76\u4e0d\u662f\u4e00\u4e2a\u771f\u5b9e\u7684\u7c7b，\u800c\u662f\u6211\u4eec\u9700\u8981\u53bb\u5b9e\u73b0\u7684\u7c7b，\u5c31\u50cf\u6211\u4eec\u4e4b\u524d\u5728\u5206\u7c7bMNIST\u6570\u636e\u96c6\u7684\u65f6\u5019，MNISI\u5e93\u672c\u8eab\u5b9e\u73b0\u4e86\u81ea\u5df1\u7684DataSet<\/p>\n

    train_dataset=<\/span>torchvision.<\/span>datasets.<\/span>MNIST(<\/span>'.\/data'<\/span>,<\/span>train=<\/span>True<\/span>,<\/span>download=<\/span>True<\/span>,<\/span>transform=<\/span>transform)<\/span><\/p>\n

    \u7ee7\u627fDataSet\u7684\u7c7b\u9700\u8981\u5b9e\u73b0\u4ee5\u4e0b\u4e24\u4e2a\u65b9\u6cd5，\u4ee5\u4fbf\u540e\u7eedDataLoader\u52a0\u8f7d\u7684\u65f6\u5019\u80fd\u591f\u6b63\u786e\u83b7\u53d6\u4fe1\u606f<\/p>\n

    __len__\u8fd4\u56de\u6570\u636e\u96c6\u7684\u6570\u91cf\u2014\u2014\u6a21\u578b\u4e00\u4e2aepoch\u8be5\u83b7\u53d6\u591a\u5c11\u8d44\u6e90<\/p>\n

    __getitem__\u8fd4\u56de\u7d22\u5f15\u5bf9\u5e94\u7684\u8d44\u6e90（image，label）-> \u6a21\u578b\u901a\u8fc7\u7d22\u5f15\u4ece\u8d44\u6e90\u96c6\u5408\u4e2d\u83b7\u53d6\u5355\u7ec4（image，label）\u5177\u4f53\u7684\u8d44\u6e90<\/p>\n

    _init_<\/h5>\n

    \u65e0\u8bba\u662f\u8fd4\u56de\u8d44\u6e90\u6570\u91cf\u8fd8\u662f\u8fd4\u56de\u5177\u4f53\u8d44\u6e90，\u6211\u4eec\u90fd\u9700\u8981\u5728\u8fd9\u4e9b\u7c7b\u4e2d\u63d0\u4f9b\u8fd9\u4e9b\u4fe1\u606f，\u56e0\u4e3a\u4e0a\u8ff0\u4e24\u4e2a\u65b9\u6cd5\u53ea\u662f\u83b7\u53d6\u7684\u64cd\u4f5c，\u6211\u4eec\u53ef\u4ee5\u5728\u7c7b\u7684\u6784\u9020\u51fd\u6570\u4e2d__init__\u52a0\u8f7d\u8fd9\u4e9b\u8d44\u6e90（\u4e3a\u4e86\u5bf9\u5185\u5b58\u66f4\u53cb\u597d，\u6211\u4eec\u53ef\u4ee5\u91c7\u7528\u61d2\u52a0\u8f7d\u7684\u5f62\u5f0f），\u5373\u7c7b\u672c\u8eab\u53ea\u4fdd\u5b58\u8fd9\u4e9b\u8d44\u6e90\u7684\u83b7\u53d6\u4f4d\u7f6e，\u7b49\u5230\u6a21\u578b\u771f\u6b63\u8c03\u7528__getitem__\u518d\u53bb\u83b7\u53d6\u5b83<\/p>\n

    def<\/span> __init__<\/span>(<\/span>self,<\/span> root,<\/span> transform,<\/span> target_transform=<\/span>None<\/span>)<\/span>:<\/span>
    \n # \u6570\u636e\u96c6\u6839\u8def\u5f84\u2014\u2014\u5916\u90e8\u8c03\u7528\u7684\u65f6\u5019\u5c06\u4f20\u5165flowers<\/span>
    \n self.<\/span>root =<\/span> root
    \n self.<\/span>img_dir =<\/span> os.<\/span>path.<\/span>join(<\/span>self.<\/span>root,<\/span>"img"<\/span>,<\/span>"jpg"<\/span>)<\/span>
    \n self.<\/span>transform =<\/span> transform<\/p>\n

    labels_mat =<\/span> scipy.<\/span>io.<\/span>loadmat(<\/span>os.<\/span>path.<\/span>join(<\/span>self.<\/span>root,<\/span>"imagelabels.mat"<\/span>)<\/span>)<\/span>
    \n self.<\/span>labels =<\/span> labels_mat[<\/span>"labels"<\/span>]<\/span>[<\/span>0<\/span>]<\/span> –<\/span> 1<\/span><\/p>\n

  • \n

    \u60a8\u4f1a\u53d1\u73b0\u4e0a\u4e00\u6b65\u6211\u4eec\u89e3\u538b\u56fe\u50cftgz\u538b\u7f29\u6587\u4ef6\u7684\u65f6\u5019，\u6211\u4eec\u7684img\u6587\u4ef6\u5939\u591a\u51fa\u4e86\u4e00\u4e2ajpg\u6587\u4ef6\u5939，\u5f53\u6253\u5f00\u8fd9\u4e2a\u6587\u4ef6\u5939\u540e，\u60a8\u4f1a\u53d1\u73b0\u90a3\u91cc\u624d\u662f\u5b58\u50a8\u771f\u5b9e\u56fe\u7247\u7684\u5730\u65b9，\u6240\u4ee5\u6211\u4eec\u7684img_dir\u4f1a\u5728\u62fc\u63a5img\u6587\u4ef6\u5939\u7684\u57fa\u7840\u4e0a\u518d\u62fc\u63a5\u4e00\u4e2ajpg\u6587\u4ef6\u5939<\/p>\n<\/li>\n

  • \n

    \u6211\u4eec\u5728\u4fdd\u5b58loabels\u7684\u65f6\u5019\u5e76\u6ca1\u6709\u4f7f\u7528\u61d2\u52a0\u8f7d，\u56e0\u4e3a\u8fd9\u53ea\u662f\u4e00\u4e2a\u6587\u4ef6，\u5bf9\u5185\u5b58\u7684\u538b\u529b\u5fae\u4e4e\u5176\u5fae\u3010\u901a\u8fc7\u67e5\u770b\u53d1\u73b0\u5b83\u4ec5\u6709502\u5b57\u8282（\u53ef\u80fd\u7ecf\u8fc7\u4e86\u538b\u7f29\u7b97\u6cd5）\u3011，\u8fd9\u4e2a\u65f6\u5019\u653e\u5728__getitem__\u9891\u7e41\u7684\u52a0\u8f7d\u53cd\u800c\u4f1a\u56e0\u4e3a\u62fc\u63a5\u7684\u51fd\u6570\u8c03\u7528\u5f00\u9500\u800c\u5bfc\u81f4\u8bad\u7ec3\u901f\u5ea6\u53d8\u6162<\/p>\n<\/li>\n

  • \n

    transform\u6211\u60f3\u7b49\u5230\u8bb2\u5b8c\u6574\u4e2adataSet\u540e\u518d\u8bb2<\/p>\n<\/li>\n

    _len_<\/h5>\n

    def<\/span> __len__<\/span>(<\/span>self)<\/span>:<\/span>
    \n return<\/span> len<\/span>(<\/span>self.<\/span>labels)<\/span><\/p>\n

    \u8fd4\u56de\u6570\u636e\u96c6\u7684\u5927\u5c0f，\u8fd9\u91cc\u6807\u7b7e\u4e0e\u56fe\u50cf\u4e00\u4e00\u5bf9\u5e94，\u6240\u4ee5\u6807\u7b7e\u7684\u5927\u5c0f\u5c31\u662f\u6574\u4e2a\u6570\u636e\u96c6\u7684\u5927\u5c0f<\/p>\n

    _getitem_<\/h5>\n

    def<\/span> __getitem__<\/span>(<\/span>self,<\/span> index)<\/span>:<\/span>
    \n img_name =<\/span> f'image_<\/span>{<\/span>index+<\/span>1<\/span>:<\/span>05d<\/span>}<\/span><\/span>.jpg'<\/span><\/span>
    \n img_path =<\/span> os.<\/span>path.<\/span>join(<\/span>self.<\/span>img_dir,<\/span>img_name)<\/span><\/p>\n

    img =<\/span> Image.<\/span>open<\/span>(<\/span>img_path)<\/span>
    \n if<\/span> self.<\/span>transform is<\/span> not<\/span> None<\/span>:<\/span>
    \n img =<\/span> self.<\/span>transform(<\/span>img)<\/span>
    \n labels =<\/span> self.<\/span>labels[<\/span>index]<\/span>
    \n return<\/span> img,<\/span> labels<\/p>\n

    \u5728\u83b7\u53d6\u5177\u4f53\u6570\u636e\u7684\u65f6\u5019，\u6a21\u578b\u4f7f\u7528\u53c2\u6570idx\u8c03\u7528\u51fd\u6570，\u8fd9\u610f\u5473\u7740\u60a8\u9700\u8981\u8fd4\u56de\u7b2cidx\u4e2a\u8d44\u6e90（\u7b2cidx\u4e2a\u56fe\u50cf\u6570\u636e\u4ee5\u53ca\u7b2cidx\u4e2a\u56fe\u50cf\u6570\u636e\u6240\u5bf9\u5e94\u7684\u6807\u7b7e），\u6807\u7b7e\u5f88\u5bb9\u6613\u5bfb\u627e，labels\u6570\u7ec4\u5f53\u4e2d\u7684\u7b2cidx\u4e2a\u6570\u636e\u5c31\u662f\u7b2cidx\u56fe\u7247\u6240\u5bf9\u5e94\u7684\u6807\u7b7e，\u5bf9\u4e8e\u6587\u4ef6\u7ed3\u6784\u6765\u8bf4，\u5b83\u662f\u4e00\u4e00\u5bf9\u5e94\u7684，\u4f46\u662f\u6211\u4eec\u5982\u4f55\u4ece\u56fe\u7247\u6587\u4ef6\u5939\u4e2d\u83b7\u53d6\u7b2cidx\u4e2a\u56fe\u50cf\u5462？<\/p>\n

    \u53ef\u4ee5\u50cf\u5bfb\u627e\u6807\u7b7e\u90a3\u6837，\u4fdd\u5b58\u6240\u6709\u6587\u4ef6\u5939\u7684\u56fe\u7247\u8def\u5f84\u5230\u4e00\u4e2a\u6570\u7ec4，\u7136\u540e\u901a\u8fc7\u627eidx\u4f4d\u627e\u5230\u7b2cidx\u4e2a\u56fe\u50cf\u8def\u5f84，\u7136\u540e\u6839\u636e\u56fe\u50cf\u8def\u5f84\u8bbf\u95ee\u5417？\u6216\u8bb8\u8fd9\u662f\u53ef\u884c\u7684，\u4f46\u662f\u8fd9\u6837\u5e76\u4e0d\u7a33\u5b9a，\u60a8\u5f97\u4fdd\u8bc1\u6570\u7ec4\u4e2d\u7684\u8d44\u6e90\u662f\u6709\u5e8f\u7684\u3002\u6807\u7b7e\u6587\u4ef6\u662f\u9010\u5b57\u8282\u6709\u5e8f\u8bfb\u53d6\u7684，\u6240\u4ee5\u4ed6\u80fd\u4fdd\u8bc1\u8bfb\u53d6\u7684\u6587\u4ef6\u548c\u6e90\u6587\u4ef6\u6570\u636e\u4e00\u81f4，\u4f46\u662f\u5bf9\u4e8e\u6587\u4ef6\u5939\u6765\u8bf4，\u4ed6\u53ef\u80fd\u5e76\u6ca1\u6709\u4e00\u4e2a\u89c4\u8303\u7684\u7ed3\u6784\u5c06\u6587\u4ef6\u4e0e\u4e00\u4e2a\u4f4d\u7f6e\u4e00\u4e00\u5bf9\u5e94，\u5c31\u50cf\u60a8\u5728\u4fdd\u5b58\u6587\u4ef6\u7684\u65f6\u5019\u4e0d\u4f1a\u6307\u5b9a\u4e00\u53f7\u6587\u4ef6\u653e\u5728\u6587\u4ef6\u5939\u7684\u7b2c\u4e00\u4e2a\u4f4d\u7f6e，\u6211\u4eec\u5f88\u5c11\u542c\u8bf4\u6587\u4ef6\u5939\u6709\u7b2c\u4e00\u4e2a\u4f4d\u7f6e\u8fd9\u79cd\u8bf4\u6cd5，\u8bfb\u53d6\u65b9\u5f0f\u6216\u8bb8\u4e5f\u662f\u968f\u673a\u7684，\u79cd\u79cd\u56e0\u7d20\u53ef\u80fd\u4f1a\u5bfc\u81f4\u6570\u7ec4\u4e2d\u7684\u6587\u4ef6\u5e76\u4e0d\u662f\u60a8\u60f3\u8981\u7684\u987a\u5e8f\u7ed3\u6784\u3010\u9664\u975e\u60a8\u5728\u8bfb\u53d6\u4e4b\u540e\u5bf9\u6570\u7ec4\u6309\u7167\u540d\u79f0\u8fdb\u884c\u6392\u5e8f\u3011\u4f46\u662f\u6211\u4eec\u7684oxford\u6570\u636e\u96c6\u7684\u786e\u662f\u6709\u7ed3\u6784\u7684，\u8fd9\u6837\u7684\u7ed3\u6784\u85cf\u5728\u6587\u4ef6\u547d\u540d\u4e2d<\/p>\n

    \u6211\u4eec\u4f1a\u53d1\u73b0，\u6587\u4ef6\u4ee5image__00001.jpg,image_00002.jpg\u8fd9\u6837\u7684\u65b9\u5f0f\u547d\u540d，\u5b9e\u9645\u4e0a\u8fd9\u5c31\u662f\u6587\u4ef6\u987a\u5e8f<\/p>\n

    00001\u5bf9\u5e94\u7b2c1\u4e2a\u6587\u4ef6，00002\u5bf9\u5e94\u7b2c\u4e8c\u4e2a\u6587\u4ef6，\u4ee5\u6b64\u7c7b\u63a8f'image_{index+1:05d}.jpg'\u5bf9\u5e94\u7b2cindex+1\u4e2a\u6587\u4ef6<\/p>\n

    05d\u4ee3\u88685\u4f4d\u6574\u6570，< 5\u4e3a\u6574\u6570\u75280\u586b\u5145，\u4e3a\u4ec0\u4e48\u8981\u7528index + 1\u800c\u4e0d\u662findex\u5462？<\/p>\n

    \u5bf9\u4e8e\u56fe\u7247\u6765\u8bf4，\u662f\u4ece00001\u5f00\u59cb\u547d\u540d\u7684，\u800c\u6a21\u578b\u4f20\u53c2\u4ece0\u5f00\u59cb，\u4ed6\u4ee3\u8868\u83b7\u53d6\u7b2c\u4e00\u5f20\u56fe\u7247<\/p>\n\n\u6a21\u578b（index）\u83b7\u53d6\u8d44\u6e90<\/tr>\n\n\n\n\n
    0<\/td>\n00001<\/td>\n<\/tr>\n
    1<\/td>\n00002<\/td>\n<\/tr>\n
    index<\/td>\nindex+1:05d<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u901a\u8fc7\u6587\u4ef6\u540d\u4e0e\u5b9e\u9645\u6587\u4ef6\u7684\u5bf9\u5e94，\u6211\u4eec\u5c31\u80fd\u627e\u5230\u7b2cidx+1\u5f20\u56fe\u7247\u548c\u7b2cidx+1\u4e2a\u6807\u7b7e\u5e76\u8fd4\u56de\u7ed9\u6a21\u578b<\/p>\n

    if<\/span> self.<\/span>transform is<\/span> not<\/span> None<\/span>:<\/span>
    \n img =<\/span> self.<\/span>transform(<\/span>img)<\/span><\/p>\n

    \u8fd9\u884c\u4ee3\u7801\u518d\u6b21\u51fa\u73b0\u4e86transform，\u8bed\u4e49\u662f\u5982\u679c\u5bf9\u8c61\u5305\u542btrannsform\u5c5e\u6027 \u540c\u65f6 transform\u5c5e\u6027\u4e0d\u662fnone\u5c31\u5e94\u7528transform\u53d8\u6362<\/p>\n

    \u89e3\u51b3\u8d28\u91cf\u95ee\u9898<\/h3>\n

    transforms\u7684\u51fa\u73b0，\u6b63\u662f\u4e3a\u89e3\u51b3\u6211\u4eec\u63d0\u51fa\u7684\u8d28\u91cf\u95ee\u9898\u3002<\/p>\n

    \u6a21\u578b\u671f\u671b\u8f93\u5165\u7684\u6570\u636e\u662f\u6807\u51c6\u4e14\u89c4\u6574\u7684，\u5373\u5927\u5c0f\u4e00\u81f4，\u8272\u5f69\u8868\u793a\u8303\u56f4\u4e00\u81f4，\u6570\u636e\u7ed3\u6784\u4e00\u81f4，\u6211\u4eec\u53ef\u4ee5\u7528transforms\u8fdb\u884c\u7edf\u4e00\u5904\u7406<\/p>\n

    # \u6ce8\u610ftransform\u662f\u60a8\u81ea\u5b9a\u4e49\u7684\u53d8\u91cf，transforms\u662f\u6a21\u5757<\/span>
    \ntransform =<\/span> transforms.<\/span>Compose(<\/span>[<\/span>
    \n transforms.<\/span>Resize(<\/span>256<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>CenterCrop(<\/span>224<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>ToTensor(<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>Normalize(<\/span>mean=<\/span>[<\/span>0.485<\/span>,<\/span> 0.456<\/span>,<\/span> 0.406<\/span>]<\/span>,<\/span> std=<\/span>[<\/span>0.229<\/span>,<\/span> 0.224<\/span>,<\/span> 0.225<\/span>]<\/span>)<\/span>,<\/span>
    \n]<\/span>)<\/span><\/p>\n

    \u8fd9\u91cc\u6211\u4eec\u7684\u5904\u7406\u6d41\u7a0b\u662f\u8fd9\u6837\u7684<\/p>\n

  • \u91cd\u65b0\u4fee\u6539\u56fe\u7247\u5c3a\u5bf8\u2014\u2014\u8fd9\u91cc\u628a\u77ed\u8fb9\u7f29\u653e\u4e3a256\u50cf\u7d20，\u957f\u8fb9\u81ea\u52a8\u7b49\u6bd4\u4f8b\u7f29\u653e<\/li>\n

    <\/p>\n

    o<\/p>\n

    r<\/p>\n

    S<\/p>\n

    h<\/p>\n

    o<\/p>\n

    r<\/p>\n

    t<\/p>\n

    o<\/p>\n

    r<\/p>\n

    L<\/p>\n

    e<\/p>\n

    n<\/p>\n

    g<\/p>\n

    t<\/p>\n

    h<\/p>\n

    \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    a<\/p>\n

    f<\/p>\n

    S<\/p>\n

    h<\/p>\n

    o<\/p>\n

    r<\/p>\n

    t<\/p>\n

    :<\/p>\n

    256<\/p>\n

    a<\/p>\n

    f<\/p>\n

    L<\/p>\n

    e<\/p>\n

    n<\/p>\n

    g<\/p>\n

    t<\/p>\n

    h<\/p>\n

    o<\/p>\n

    r<\/p>\n

    L<\/p>\n

    e<\/p>\n

    n<\/p>\n

    g<\/p>\n

    t<\/p>\n

    h
    \n \u2005\u200a<\/p>\n

    \u2217
    \n \u2005\u200a<\/p>\n

    a<\/p>\n

    f<\/p>\n

    S<\/p>\n

    h<\/p>\n

    o<\/p>\n

    r<\/p>\n

    t<\/p>\n

    :<\/p>\n

    256
    \n \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    o<\/p>\n

    r<\/p>\n

    S<\/p>\n

    h<\/p>\n

    o<\/p>\n

    r<\/p>\n

    t
    \n \u2005\u200a<\/p>\n

    \u2217
    \n \u2005\u200a<\/p>\n

    a<\/p>\n

    f<\/p>\n

    L<\/p>\n

    e<\/p>\n

    n<\/p>\n

    g<\/p>\n

    t<\/p>\n

    h<\/p>\n

    a<\/p>\n

    f<\/p>\n

    L<\/p>\n

    e<\/p>\n

    n<\/p>\n

    g<\/p>\n

    t<\/p>\n

    h
    \n \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    a<\/p>\n

    f<\/p>\n

    S<\/p>\n

    h<\/p>\n

    o<\/p>\n

    r<\/p>\n

    t<\/p>\n

    :<\/p>\n

    256
    \n \u2005\u200a<\/p>\n

    \u2217
    \n \u2005\u200a<\/p>\n

    o<\/p>\n

    r<\/p>\n

    L<\/p>\n

    e<\/p>\n

    n<\/p>\n

    g<\/p>\n

    t<\/p>\n

    h<\/p>\n

    o<\/p>\n

    r<\/p>\n

    S<\/p>\n

    h<\/p>\n

    o<\/p>\n

    r<\/p>\n

    t<\/p>\n

    \\\\begin{align} &\\\\dfrac{orShort}{orLength}\\\\;=\\\\;\\\\dfrac{afShort:256}{afLength} \\\\\\\\ &orLength\\\\;*\\\\;afShort:256\\\\;=\\\\;orShort\\\\;*\\\\;afLength \\\\\\\\ &afLength\\\\;=\\\\; \\\\dfrac{afShort:256\\\\;*\\\\;orLength}{orShort} \\\\end{align} <\/p>\n

    <\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>or<\/span>L<\/span>e<\/span>n<\/span>g<\/span>t<\/span>h<\/span><\/span><\/span><\/span><\/span><\/span><\/span>or<\/span>S<\/span>h<\/span>or<\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span>a<\/span>f<\/span>L<\/span>e<\/span>n<\/span>g<\/span>t<\/span>h<\/span><\/span><\/span><\/span><\/span><\/span><\/span>a<\/span>f<\/span>S<\/span>h<\/span>or<\/span>t<\/span><\/span>:<\/span><\/span>256<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>or<\/span>L<\/span>e<\/span>n<\/span>g<\/span>t<\/span>h<\/span><\/span><\/span>\u2217<\/span><\/span><\/span>a<\/span>f<\/span>S<\/span>h<\/span>or<\/span>t<\/span><\/span>:<\/span><\/span>256<\/span><\/span><\/span>=<\/span><\/span><\/span>or<\/span>S<\/span>h<\/span>or<\/span>t<\/span><\/span><\/span>\u2217<\/span><\/span><\/span>a<\/span>f<\/span>L<\/span>e<\/span>n<\/span>g<\/span>t<\/span>h<\/span><\/span><\/span><\/span><\/span>a<\/span>f<\/span>L<\/span>e<\/span>n<\/span>g<\/span>t<\/span>h<\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span>or<\/span>S<\/span>h<\/span>or<\/span>t<\/span><\/span><\/span><\/span><\/span><\/span><\/span>a<\/span>f<\/span>S<\/span>h<\/span>or<\/span>t<\/span><\/span>:<\/span><\/span>256<\/span><\/span><\/span>\u2217<\/span><\/span><\/span>or<\/span>L<\/span>e<\/span>n<\/span>g<\/span>t<\/span>h<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n

  • \u53d6\u6837\u6709\u6548\u533a\u57df\u2014\u2014\u53d6\u4e2d\u5fc3\u533a\u57df\u957f\u5bbd\u4e3a224\u50cf\u7d20\u7684\u5c3a\u5bf8，\u8fd9\u6837\u6240\u6709\u7ecf\u8fc7\u5904\u7406\u7684\u5355\u5f20\u56fe\u7247\u5c31\u90fd\u7edf\u4e00\u6807\u51c6\u5316\u6210\u4e86\u30103，224，224\u3011<\/li>\n
  • \n

    \u4e3a\u4ec0\u4e48\u4e0d\u5728\u7b2c\u4e00\u6b65\u5c31Resize(224)\u5462？<\/p>\n

    \u7b2c\u4e00\u6b65\u5c06\u77ed\u8fb9\u7f29\u653e\u4e3a224\u50cf\u7d20，\u518d\u4ece\u4e2d\u5fc3\u622a\u53d6，\u786e\u5b9e\u80fd\u591f\u6784\u5efa\u51fa\u30103，224，224\u3011\u5c3a\u5bf8\u7684\u56fe\u7247，\u4f46\u662f\u6211\u4eec\u8fd8\u8003\u8651\u5230\u4e86\u566a\u58f0\u7684\u5904\u7406，\u591a\u6570\u65f6\u5019，\u56fe\u7247\u7684\u4e00\u4e9b\u8fb9\u7f18\u4fe1\u606f\u662f\u4e0d\u91cd\u8981\u7684，\u4ed6\u548c\u6574\u4f53\u8bc6\u522b\u65e0\u5173，\u6216\u8bb8\u662f\u73af\u5883\u6216\u8bb8\u662f\u6c34\u5370，\u5982\u679c\u6211\u4eec\u5c06\u5e26\u6709\u8fd9\u4e9b\u5370\u8bb0\u7684\u4fe1\u606f\u6dfb\u52a0\u8fdb\u53bb，\u4e0d\u4ec5\u65e0\u6548，\u751a\u81f3\u53ef\u80fd\u4f1a\u5e72\u6270\u6a21\u578b\u7684\u8bad\u7ec3，\u6240\u4ee5\u6211\u4eec\u505a\u4e86\u4e00\u4e2a\u7b80\u5355\u7684\u566a\u58f0\u5904\u7406，\u9ed8\u8ba4\u8fb9\u7f18\u4e3a\u65e0\u5173\u4fe1\u606f，\u81f3\u5c11\u4f1a\u5728\u77ed\u8fb9\u88c1\u526a\u51fa\u5de6\u53f3\u540416\u50cf\u7d20\u7684\u566a\u58f0<\/p>\n<\/li>\n

  • \n

    \u4e3a\u4ec0\u4e48\u91c7\u7528\u4e2d\u5fc3\u88c1\u526a，\u800c\u4e0d\u662f\u76f4\u63a5\u5c06\u56fe\u7247\u62c9\u4f38\u4e3a（3，224，224）<\/p>\n

    \u62c9\u4f38\u540e\u7684\u56fe\u7247\u53d8\u5f62，\u4f1a\u8ba9\u4f9d\u8d56\u4e8e\u76f8\u5bf9\u4f4d\u7f6e\u5173\u7cfb\u7684\u7279\u5f81\u88ab\u4fee\u6539，\u4ea7\u751f\u9519\u8bef\u7684\u7279\u5f81，\u4ece\u800c\u5e72\u6270\u6a21\u578b\u7684\u8bad\u7ec3<\/p>\n<\/li>\n

  • \u8f6c\u6362\u4e3a\u6a21\u578b\u80fd\u591f\u8bad\u7ec3\u7684\u6570\u636e\u7c7b\u578b\u2014\u2014（RGB\u7a7a\u95f4->\u5f20\u91cf）<\/li>\n
  • \u5f52\u4e00\u5316（\u30100~255\u3011\/ 255 ->\u30100 ~ 1\u3011)，\u6a21\u578b\u5bf9\u4e8e\u6570\u503c\u8f83\u5c0f\u7684\u6570\u636e，\u5728\u8bad\u7ec3\u7684\u8fc7\u7a0b\u4e2d\u66f4\u6709\u4f18\u52bf，\u800c\u4e14\u5c0f\u6570\u636e\u53ef\u4ee5\u9632\u6b62\u6570\u636e\u8fc7\u5927\u800c\u5bfc\u81f4\u7684\u6ea2\u51fa\u95ee\u9898<\/li>\n

    \u6700\u540e\u5c31\u53ef\u4ee5\u901a\u8fc7\u6784\u9020\u51fd\u6570\u4f20\u5165transform，\u4f7f\u5f97\u6a21\u578b\u5728\u8c03\u7528__getitem__\u83b7\u53d6\u6570\u636e\u7684\u65f6\u5019\u81ea\u52a8\u5e94\u7528transform<\/p>\n

    transform =<\/span> transforms.<\/span>Compose(<\/span>[<\/span>
    \n transforms.<\/span>Resize(<\/span>256<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>CenterCrop(<\/span>224<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>ToTensor(<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>Normalize(<\/span>mean=<\/span>[<\/span>0.485<\/span>,<\/span> 0.456<\/span>,<\/span> 0.406<\/span>]<\/span>,<\/span> std=<\/span>[<\/span>0.229<\/span>,<\/span> 0.224<\/span>,<\/span> 0.225<\/span>]<\/span>)<\/span>,<\/span>
    \n]<\/span>)<\/span>
    \ndata_prepare.<\/span>download_dataset(<\/span>)<\/span>
    \ndataset =<\/span> OxfordFlowersDataSet(<\/span>'flowers'<\/span>,<\/span>transform=<\/span>transform)<\/span><\/p>\n

    \u89e3\u51b3\u6548\u7387\u95ee\u9898<\/h3>\n

    \u6a21\u578b\u8bad\u7ec3\u4e0d\u4ec5\u5305\u542b\u8bad\u7ec3\u7684\u8fc7\u7a0b，\u8fd8\u5305\u542b\u9a8c\u8bc1\u3001\u6d4b\u8bd5，\u5177\u4f53\u53ef\u89c1pyTorch\u6784\u5efa\u6a21\u5757\u5316\u7684\u65b9\u5f0f<\/p>\n

    \u4ec0\u4e48\u662f\u9a8c\u8bc1\u96c6<\/p>\n

    \u60a8\u53ef\u4ee5\u5c06\u5b83\u7406\u89e3\u4e3a\u8bad\u7ec3\u9636\u6bb5\u7684\u6d4b\u8bd5\u96c6\u3002\u8fd9\u91cc\u7684\u8bad\u7ec3\u9636\u6bb5\u4f1a\u6bd4\u6a21\u5757\u5316\u7684\u6211\u8bad\u7ec3\u66f4\u5e7f\u6cdb，\u53ea\u8981\u60a8\u8fd8\u51c6\u5907\u8c03\u53c2\u4f18\u5316\u6216\u8005\u8c03\u6574\u4ee3\u7801，\u4e0d\u6253\u7b97\u6b63\u5f0f\u90e8\u7f72，\u90fd\u53ef\u4ee5\u88ab\u89c4\u5212\u4e3a\u8bad\u7ec3\u9636\u6bb5，\u60a8\u5728\u8bad\u7ec3\u51e0\u4e2aepoch\u540e，\u4f1a\u7528\u9a8c\u8bc1\u96c6\u8fdb\u884c\u9a8c\u8bc1，\u5e76\u6839\u636e\u7ed3\u679c\u8c03\u53c2\u4f18\u5316，\u800c\u6d4b\u8bd5\u96c6\u5c31\u662f\u6b63\u5f0f\u90e8\u7f72\u4e4b\u524d\u7684\u6700\u540e\u6d4b\u8bd5，\u60a8\u53ef\u4ee5\u7528\u4e0d\u540c\u7684\u6d4b\u8bd5\u96c6，\u4f46\u662f\u7ed3\u679c\u4e0d\u80fd\u518d\u4f5c\u4e3a\u8c03\u53c2\u7684\u4f9d\u636e，<\/p>\n

    \u6211\u4eec\u53ef\u4ee5\u5bf9\u6574\u4e2a\u6570\u636e\u96c6\u6309\u5927\u5c0f\u968f\u673a\u5212\u5206，\u4ece\u800c\u5212\u5206\u51fa\u4e09\u4e2a\u4e0d\u540c\u7684\u6570\u636e\u96c6，\u6240\u6709\u96c6\u5408\u7684\u5927\u5c0f\u4e00\u822c\u662f（\u8bad\u7ec3\u96c6：0.7，\u6d4b\u8bd5\u96c6：0.15，\u9a8c\u8bc1\u96c6：0.15）<\/p>\n

    \u4e3a\u4ec0\u4e48\u8981\u968f\u673a\u5212\u5206<\/p>\n

    \u5982\u679c\u6570\u636e\u96c6\u672c\u8eab\u662f\u6709\u5e8f\u7684，\u90a3\u4e48\u5f88\u6709\u53ef\u80fd\u4f1a\u51fa\u73b0\u4e09\u4e2a\u96c6\u5408；\u843d\u5728\u67d0\u4e00\u4e2a\u6216\u8005\u67d0\u51e0\u4e2a\u7c7b\u522b\u4e0a，\u4ece\u800c\u65e0\u6cd5\u8986\u76d6\u5927\u591a\u6570\u6d4b\u8bd5\u60c5\u51b5，\u5bf9\u4e8e\u8bad\u7ec3\u96c6\u6211\u4eec\u8fd8\u9700\u8981\u5728\u6bcf\u6b21\u52a0\u8f7d\u7684\u65f6\u5019\u8fdb\u884c\u968f\u673a\u6d17\u724c，\u4ee5\u907f\u514d\u5728\u591a\u6b21\u7684epoch\u4e2d\u5b66\u4e60\u5230\u4e0e\u6279\u6b21\u76f8\u5173\u7684\u7279\u5f81\u800c\u975e\u4e0e\u56fe\u7247\u672c\u8eab，\u5982\u679c\u6bcf\u6b21\u7684epoch\u91c7\u7528\u76f8\u540c\u7684\u5173\u7cfb，\u6a21\u578b\u53ef\u80fd\u4f1a\u8fd9\u6837\u8ba4\u4e3a，\u7b2c\u4e00\u6b21\u662f\u73ab\u7470\u3001\u6708\u5b63\u3001\u7261\u4e39\u2026\u2026，\u7b2c\u4e8c\u6b21\u4e5f\u662f\u73ab\u7470\u3001\u6708\u5b63\u3001\u7261\u4e39\u3001\u7b2c\u4e09\u6b21\u4e5f\u662f\u2026\u2026，\u7b2cn\u6b21，\u7136\u540e\u5b83\u4f1a\u8ba4\u4e3a\u7279\u5f81\u5e94\u8be5\u5c31\u662f\u5982\u6b64，\u7528\u6237\u7b2c\u4e00\u6b21\u5582\u7ed9\u6211\u7684\u6570\u636e\u662f\u73ab\u7470，\u7b2c\u4e8c\u6b21\u5582\u7ed9\u6211\u7684\u6570\u636e\u662f\u6708\u5b63\u2026\u2026，\u5f53\u6211\u4eecDataLoader\u968f\u673a\u6d17\u724c\u540e，\u6bcf\u4e2a\u6279\u6b21\u7684\u6570\u636e，\u6279\u6b21\u4e0e\u6279\u6b21\u7684\u6570\u636e\u90fd\u53d8\u5f97\u4e0d\u5b8c\u5168\u4e00\u81f4，\u8fd9\u6709\u52a9\u4e8e\u9632\u6b62\u6a21\u578b\u8fc7\u62df\u5408，\u63d0\u9ad8\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b<\/p>\n

    Knuth\u6d17\u724c\u7b97\u6cd5（\u62d3\u5c55）<\/h5>\n

    \u6d17\u724c\u7b97\u6cd5\u6709\u4e00\u4e2a\u6781\u5176\u4f18\u96c5\u7684\u5b9e\u73b0\u2014\u2014\u4ed6\u80fd\u4fdd\u8bc1\u516c\u5e73\u7684\u6d17\u724c，\u5728\u67d0\u4e2a\u4f4d\u7f6e\u4e0a，\u6bcf\u4e2a\u5143\u7d20\u51fa\u73b0\u7684\u6982\u7387\u90fd\u662f 1 \/ n<\/p>\n

    import<\/span> secrets
    \nk_list =<\/span> [<\/span>1<\/span>,<\/span>2<\/span>,<\/span>3<\/span>,<\/span>4<\/span>,<\/span>5<\/span>]<\/span>
    \nfor<\/span> i in<\/span> range<\/span>(<\/span>len<\/span>(<\/span>k_list)<\/span> –<\/span> 1<\/span>,<\/span>–<\/span>1<\/span>,<\/span>–<\/span>1<\/span>)<\/span>:<\/span>
    \n rd =<\/span> secrets.<\/span>randbelow(<\/span>i+<\/span>1<\/span>)<\/span>
    \n k_list[<\/span>i]<\/span>,<\/span>k_list[<\/span>rd]<\/span> =<\/span> k_list[<\/span>rd]<\/span>,<\/span>k_list[<\/span>i]<\/span>
    \nprint<\/span>(<\/span>k_list)<\/span><\/p>\n

    $$ \\\\begin{align}<\/p>\n

    &index_{n-1}:\\\\dfrac{1}{n}\\\\ &index_{n-2}:\\\\dfrac{1}{n-1};;\\\\dfrac{n-1}{n};=;\\\\dfrac{1}{n}\\\\ &index_{n-3}:\\\\dfrac{1}{n-2};;\\\\dfrac{n-2}{n-1};;\\\\dfrac{n-1}{n};=;\\\\dfrac{1}{n}\\\\ &\u2026\\\\ &index_{0}:\\\\dfrac{1}{\\\\cancel{1}};;\\\\dfrac{\\\\cancel{1}}{\\\\cancel{2}};;\\\\dfrac{\\\\cancel{2}}{\\\\cancel{3}};;\u2026;;\\\\dfrac{\\\\cancel{n-2}}{\\\\cancel{n-1}};*;\\\\dfrac{\\\\cancel{n-1}}{n};=\\\\dfrac{1}{n} \\\\end{align} $$<\/p>\n

    \u5728\u7b2cn\u4e2a\u4f4d\u7f6e\u7684\u65f6\u5019，\u6bcf\u4e2a\u5143\u7d20\u53d6\u5230\u7684\u6982\u7387\u76f8\u540c\u3010secrets.randbelow(i)\u662f\u771f\u968f\u673a\u751f\u6210[0,i)\u7684\u968f\u673a\u6574\u6570\u3011，\u6982\u7387<\/p>\n

    p
    \n \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    1<\/p>\n

    n<\/p>\n

    {p\\\\;=\\\\;\\\\dfrac{1}{n}}<\/p>\n

    <\/span><\/span>p<\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span>n<\/span><\/span><\/span><\/span><\/span><\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u7136\u540e\u8be5\u4f4d\u7f6e\u5df2\u7ecf\u662f\u968f\u673a，\u4e0d\u9700\u8981\u518d\u7eb3\u5165\u968f\u673a\u8303\u56f4\u4e86，\u6b64\u65f6\u968f\u673a\u8303\u56f4\u4e3a[0,i-1)<\/p>\n

    \u5728\u7b2cn-1\u4e2a\u4f4d\u7f6e\u65f6，\u6bcf\u4e2a\u5143\u7d20\u53d6\u5230\u7684\u6982\u7387\u662f<\/p>\n

    p
    \n \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    1<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    1<\/p>\n

    p\\\\;=\\\\;{\\\\dfrac{1}{n-1}}<\/p>\n

    <\/span><\/span>p<\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>1<\/span><\/span><\/span><\/span><\/span><\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u4f46\u60a8\u9700\u8981\u786e\u4fdd\u4e0a\u4e00\u6b21\u7684\u968f\u673a\u4e0d\u4f1a\u53d6\u5230\u8be5\u5143\u7d20，\u4e0a\u4e00\u6b21\u53d6\u5230p\u7684\u6982\u7387\u662f<\/p>\n

    1<\/p>\n

    n<\/p>\n

    {\\\\dfrac{1}{n}}<\/p>\n

    <\/span><\/span><\/span><\/span>n<\/span><\/span><\/span><\/span><\/span><\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u5bf9\u4e8e\u4e24\u79cd\u60c5\u51b5\u7684\u5426\u547d\u9898\u3010\u53d6\u548c\u4e0d\u53d6\u3011\u6765\u8bf4，\u4e0d\u53d6\u7684\u6982\u7387\u662f <\/p>\n

    \u00ac<\/p>\n

    P<\/p>\n

    \u53d6<\/p>\n

    {\\\\neg{P_\u53d6}}<\/p>\n

    <\/span><\/span>\u00ac<\/span>P<\/span><\/span>\u53d6<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u6982\u7387\u662f <\/p>\n

    1
    \n \u2005\u200a<\/p>\n

    \u2212
    \n \u2005\u200a<\/p>\n

    P<\/p>\n

    \u53d6<\/p>\n

    \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    1<\/p>\n

    n<\/p>\n

    {1\\\\;-\\\\;P_\u53d6\\\\;=\\\\;\\\\dfrac{n-1}{n}}<\/p>\n

    <\/span><\/span>1<\/span><\/span><\/span>\u2212<\/span><\/span><\/span>P<\/span><\/span>\u53d6<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span>n<\/span><\/span><\/span><\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u6240\u4ee5\u6982\u7387\u662f\u5728\u4e0a\u4e00\u8f6e\u4e0d\u4f1a\u53d6\u5230\u8be5\u5143\u7d20\u7684\u60c5\u51b5\u4e0b\u8fd9\u4e00\u8f6e\u53d6\u5230\u8be5\u5143\u7d20<\/p>\n

    p<\/p>\n

    =<\/p>\n

    1<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    1<\/p>\n

    \u2005\u200a<\/p>\n

    \u2217
    \n \u2005\u200a<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    1<\/p>\n

    n<\/p>\n

    \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    1<\/p>\n

    n<\/p>\n

    {p = \\\\dfrac{1}{n-1}\\\\;*\\\\;\\\\dfrac{n-1}{n}\\\\;=\\\\;\\\\dfrac{1}{n}}<\/p>\n

    <\/span><\/span>p<\/span><\/span>=<\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>1<\/span><\/span><\/span><\/span><\/span><\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span><\/span><\/span>n<\/span><\/span><\/span><\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span>n<\/span><\/span><\/span><\/span><\/span><\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u7136\u540e\u8be5\u4f4d\u7f6e\u5df2\u7ecf\u662f\u968f\u673a，\u4e0d\u9700\u8981\u518d\u7eb3\u5165\u968f\u673a\u8303\u56f4\u4e86，\u6b64\u65f6\u968f\u673a\u8303\u56f4\u4e3a[0,i-2)<\/p>\n

    \u540c\u7406，\u5728n-2\u4e2a\u4f4d\u7f6e\u65f6，\u6bcf\u4e2a\u5143\u7d20\u53d6\u5230\u7684\u6982\u7387\u662f <\/p>\n

    p
    \n \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    1<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    2<\/p>\n

    p\\\\;=\\\\;{\\\\dfrac{1}{n-2}}<\/p>\n

    <\/span><\/span>p<\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u5728\u524d\u9762\u4e24\u6b21\u968f\u673a\u90fd\u4e0d\u53d6\u5f97\u60c5\u51b5\u4e0b\u3010\u7b2c\u4e00\u6b21\u662f<\/p>\n

    1
    \n \u2005\u200a<\/p>\n

    \u2212
    \n \u2005\u200a<\/p>\n

    P<\/p>\n

    \u53d6<\/p>\n

    1<\/p>\n

    \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    1<\/p>\n

    n<\/p>\n

    {1\\\\;-\\\\;P_{\u53d61}\\\\;=\\\\;\\\\dfrac{n-1}{n}}<\/p>\n

    <\/span><\/span>1<\/span><\/span><\/span>\u2212<\/span><\/span><\/span>P<\/span><\/span>\u53d6<\/span>1<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span>n<\/span><\/span><\/span><\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>，\u7b2c\u4e8c\u6b21\u662f<\/p>\n

    1
    \n \u2005\u200a<\/p>\n

    \u2212
    \n \u2005\u200a<\/p>\n

    P<\/p>\n

    \u53d6<\/p>\n

    2<\/p>\n

    \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    1
    \n \u2005\u200a<\/p>\n

    \u2212
    \n \u2005\u200a<\/p>\n

    1<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    1<\/p>\n

    \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    2<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    1<\/p>\n

    {1\\\\;-\\\\;P_{\u53d62}\\\\;=\\\\;1\\\\;-\\\\;\\\\dfrac{1}{n-1}\\\\;=\\\\;\\\\dfrac{n-2}{n-1}}<\/p>\n

    <\/span><\/span>1<\/span><\/span><\/span>\u2212<\/span><\/span><\/span>P<\/span><\/span>\u53d6<\/span>2<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span>1<\/span><\/span><\/span>\u2212<\/span><\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>1<\/span><\/span><\/span><\/span><\/span><\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>1<\/span><\/span><\/span><\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>2<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u3011,<\/p>\n

    p
    \n \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    1<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    2<\/p>\n

    \u2005\u200a<\/p>\n

    \u2217
    \n \u2005\u200a<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    2<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    1<\/p>\n

    \u2005\u200a<\/p>\n

    \u2217
    \n \u2005\u200a<\/p>\n

    n<\/p>\n

    \u2212<\/p>\n

    1<\/p>\n

    n<\/p>\n

    \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    1<\/p>\n

    n<\/p>\n

    {p\\\\;=\\\\;\\\\dfrac{1}{n-2}\\\\;*\\\\;\\\\dfrac{n-2}{n-1}\\\\;*\\\\;\\\\dfrac{n-1}{n}\\\\;=\\\\;\\\\dfrac{1}{n}}<\/p>\n

    <\/span><\/span>p<\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>2<\/span><\/span><\/span><\/span><\/span><\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>1<\/span><\/span><\/span><\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>2<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span><\/span><\/span>n<\/span><\/span><\/span><\/span><\/span><\/span><\/span>n<\/span><\/span>\u2212<\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span>n<\/span><\/span><\/span><\/span><\/span><\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n

    \u4ee5\u6b64\u7c7b\u63a8\u5f53n = 0\u65f6，p\u4e5f\u7b49\u4e8e<\/p>\n

    1<\/p>\n

    n<\/p>\n

    {\\\\dfrac{1}{n}}<\/p>\n

    <\/span><\/span><\/span><\/span>n<\/span><\/span><\/span><\/span><\/span><\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n

    batch\u8bbe\u7f6e<\/h4>\n

    \u8ba9\u6211\u4eec\u56de\u5230pytorch\u672c\u8eab，\u5bf9\u4e8eDataLoader\u7684batchSize，\u5efa\u8bae\u914d\u7f6e\u4e3a32，\u5982\u679c\u60a8\u7684\u5185\u5b58 \/ \u663e\u5b58\u65e0\u6cd5\u652f\u6301\u4e00\u6b21\u6027\u52a0\u8f7d\u8fd9\u4e9b\u8d44\u6e90，\u5c1d\u8bd5\u5c06\u5b83\u7f29\u51cf，\u7b49\u5230\u53ef\u4ee5\u5b58\u50a8\u7684\u65f6\u5019\u518d\u653e\u5927，\u76f4\u5230\u65e0\u6cd5\u5b58\u50a8\u53c8\u7f29\u51cf，\u7136\u540e\u53ef\u4ee5\u53c8\u653e\u5927\u2014\u2014\u8fd9\u6837\u5faa\u73af\u5f80\u590d\u540e\u53ef\u4ee5\u627e\u5230\u4e34\u754c\u503c（\u5982\u679c\u60a8\u4e86\u89e3\u4e8c\u5206\u601d\u60f3\u7684\u8bdd，\u53ef\u4ee5\u7528\u4e8c\u5206\u6765\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898），\u5982\u679c\u60a8\u60f3\u8c03\u6574\u6bd432\u66f4\u9ad8\u7684 \u52a0\u8f7d\u6570\u636e \/ batch，\u60a8\u4e5f\u53ef\u4ee5\u901a\u8fc7\u4e8c\u5206\u601d\u60f3\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898<\/p>\n

    \u5173\u4e8e batch \u548c epoch \u7684\u533a\u522b<\/p>\n

    \u60a8\u4f1a\u5728batch\u91cc\u9762\u89c4\u5b9a\u6bcf\u4e00\u8f6e\u6b21\u52a0\u8f7d\u591a\u5c11\u5f20\u56fe\u7247，\u800c\u4e3a\u4ec0\u4e48epoch\u4e0d\u8981\u89c4\u5b9a\u5462，\u4e8b\u5b9e\u4e0a，\u60a8\u5df2\u7ecf\u89c4\u5b9a\u4e86，\u60a8\u8fd8\u8bb0\u5f97\u60a8\u7f16\u5199\u7684__len__\u65b9\u6cd5\u5417，\u5b83\u4f1a\u8fd4\u56de\u6240\u6709\u6570\u636e\u96c6\u7684\u603b\u91cf，\u4e8b\u5b9e\u4e0a，\u8fd9\u5c31\u662f\u4e00\u4e2aepoch，\u4e00\u4e2aepoch\u8986\u76d6\u4e86\u6240\u6709\u7684\u56fe\u50cf\u3010\u81f3\u5c11\u8986\u76d6\u4e86\u6574\u4e2a\u6570\u636e\u96c6\u7684\u8bad\u7ec3\u96c6\u90e8\u5206\u3011，\u800cbatch\u5c31\u662f\u4e3a\u5b9e\u73b0\u4e00\u4e2aepoch\u6240\u505a\u7684\u52a0\u8f7d\u7b56\u7565，\u6bcf\u6b21\u53d632\u4e2abatch，\u60a8\u4f1a\u5728(\u4e00\u4e2aepoch\u9700\u8981\u5904\u7406\u7684\u603b\u56fe\u50cf\/ 32 + (\u4e00\u4e2aepoch\u9700\u8981\u5904\u7406\u7684\u603b\u56fe\u50cf%32 > 0 ? 1 : 0) )\u4e2abatch\u4e2d\u5b8c\u6210\u4e00\u8f6eepoch\u7684\u8bad\u7ec3，\u60a8\u53ef\u4ee5\u9009\u62e9\u8bad\u7ec3\u591a\u4e2aepoch，\u5b83\u4f1a\u7528\u60a8\u7684\u8bad\u7ec3\u96c6\u518d\u6b21\u8bad\u7ec3，\u4e0d\u8fc7\u7531\u4e8eshuffle，\u8fd9\u6b21\u8bad\u7ec3\u96c6\u7684\u52a0\u8f7d\u987a\u5e8f\u4e0e\u4e0a\u4e00\u6b21\u7684\u4f1a\u4e0d\u540c<\/p>\n

    code-clip<\/h5>\n

    train_data_size =<\/span> int<\/span>(<\/span>len<\/span>(<\/span>dataset)<\/span> *<\/span> 0.7<\/span>)<\/span>
    \nvalid_data_size =<\/span> int<\/span>(<\/span>len<\/span>(<\/span>dataset)<\/span> *<\/span> 0.15<\/span>)<\/span>
    \ntest_data_size =<\/span> len<\/span>(<\/span>dataset)<\/span> –<\/span> train_data_size –<\/span> valid_data_size<\/p>\n

    train_data,<\/span>valid_data,<\/span>test_data =<\/span> random_split(<\/span>dataset,<\/span>[<\/span>train_data_size,<\/span>valid_data_size,<\/span>test_data_size]<\/span>)<\/span><\/p>\n

    data_batch =<\/span> DataLoader(<\/span>dataset=<\/span>train_data,<\/span>batch_size=<\/span>32<\/span>,<\/span>shuffle=<\/span>True<\/span>)<\/span>
    \nvalid_batch =<\/span> DataLoader(<\/span>dataset=<\/span>valid_data,<\/span>batch_size=<\/span>32<\/span>,<\/span>shuffle=<\/span>False<\/span>)<\/span>
    \ntest_batch =<\/span> DataLoader(<\/span>dataset=<\/span>test_data,<\/span>batch_size=<\/span>32<\/span>,<\/span>shuffle=<\/span>False<\/span>)<\/span><\/p>\n

    print<\/span>(<\/span>f'Train Batch size: <\/span>{<\/span>len<\/span>(<\/span>train_data)<\/span>}<\/span><\/span>'<\/span><\/span>)<\/span>
    \nprint<\/span>(<\/span>f'Validation Batch size: <\/span>{<\/span>len<\/span>(<\/span>valid_data)<\/span>}<\/span><\/span>'<\/span><\/span>)<\/span>
    \nprint<\/span>(<\/span>f'Test Batch size: <\/span>{<\/span>len<\/span>(<\/span>test_data)<\/span>}<\/span><\/span>'<\/span><\/span>)<\/span>
    \nfor<\/span> index,<\/span>(<\/span>images,<\/span>labels)<\/span> in<\/span> enumerate<\/span>(<\/span>data_batch)<\/span>:<\/span>
    \n if<\/span> index ><\/span> 175<\/span>:<\/span>
    \n print<\/span>(<\/span>f'the <\/span>{<\/span>index}<\/span><\/span>th batch size is <\/span>{<\/span>images.<\/span>shape}<\/span><\/span>'<\/span><\/span>)<\/span><\/p>\n

    \u9632\u6f0f\u6d1e\u90e8\u7f72\u548c\u6570\u636e\u589e\u5f3a<\/h3>\n

    \u6570\u636e\u589e\u5f3a<\/h4>\n

    \u5982\u679c\u60a8\u6765\u81ea\u7b2c\u56db\u7ae0\u2014\u2014\u6a21\u5757\u5316\u67b6\u6784，\u8bf7\u70b9\u51fb\u6b64\u5904\u8fd4\u56de，\u5426\u5219\u8bf7\u5ffd\u7565<\/p>\n

    \u60a8\u662f\u5426\u6709\u8fc7\u8fd9\u6837\u7684\u7ecf\u5386\u5462，\u5f53\u60a8\u5728\u4eba\u8138\u8bc6\u522b\u7684\u65f6\u5019，\u4e0d\u7ba1\u662f\u5728\u660e\u4eae\u7684\u5730\u65b9\u8fd8\u662f\u5728\u7a0d\u5fae\u660f\u6697\u7684\u5730\u65b9，\u8bbe\u5907\u603b\u80fd\u7cbe\u786e\u8bc6\u522b\u5230\u60a8\u7684\u9762\u90e8，\u4ece\u800c\u4e3a\u60a8\u5f00\u542f\u6743\u9650，\u8fd9\u5728\u6a21\u578b\u8bad\u7ec3\u4e0a\u7684\u4e00\u4e2a\u542f\u53d1\u5c31\u662f，\u6a21\u578b\u80fd\u4e0d\u80fd\u6709\u4e00\u5b9a\u7684\u6297\u5e72\u6270\u56e0\u7d20，\u4e3a\u4e86\u505a\u5230\u8fd9\u4e00\u70b9，\u6211\u4eec\u9700\u8981\u5728\u6a21\u578b\u8bad\u7ec3\u7684\u8fc7\u7a0b\u4e2d，\u4e3a\u8bad\u7ec3\u96c6\u505a\u4e00\u4e9b\u6570\u636e\u589e\u5f3a（\u7ffb\u8f6c，\u4e8c\u503c\u5316\u2026\u2026）\u7684\u64cd\u4f5c，\u671f\u671b\u6a21\u578b\u5728\u6570\u636e\u589e\u5f3a\u540e\u4f9d\u7136\u80fd\u591f\u8bc6\u522b\u56fe\u7247<\/p>\n

    transform =<\/span> transforms.<\/span>Compose(<\/span>[<\/span>
    \n # \u6570\u636e\u589e\u5f3a<\/span>
    \n transforms.<\/span>RandomHorizontalFlip(<\/span>p=<\/span>0.5<\/span>)<\/span>,<\/span> # \u968f\u673a\u6c34\u5e73\u7ffb\u8f6c<\/span>
    \n transforms.<\/span>RandomRotation(<\/span>degrees=<\/span>10<\/span>)<\/span>,<\/span> # \u968f\u673a\u65cb\u8f6c<\/span>
    \n transforms.<\/span>ColorJitter(<\/span>brightness=<\/span>0.2<\/span>)<\/span>,<\/span> # \u968f\u673a\u8c03\u6574\u660e\u6697\u5ea6<\/span>
    \n # \u6570\u636e\u6807\u51c6\u5316<\/span>
    \n transforms.<\/span>Resize(<\/span>256<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>CenterCrop(<\/span>224<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>ToTensor(<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>Normalize(<\/span>mean=<\/span>[<\/span>0.485<\/span>,<\/span> 0.456<\/span>,<\/span> 0.406<\/span>]<\/span>,<\/span> std=<\/span>[<\/span>0.229<\/span>,<\/span> 0.224<\/span>,<\/span> 0.225<\/span>]<\/span>)<\/span>,<\/span>
    \n]<\/span>)<\/span><\/p>\n

    \u6211\u4eec\u5728\u5e94\u7528\u6570\u636e\u53d8\u6362\u7684\u65f6\u5019\u4f1a\u968f\u673a\u5bf9\u6570\u636e\u8fdb\u884c\u6c34\u5e73\u7ffb\u8f6c\u3001\u65cb\u8f6c\u3001\u8c03\u6574\u4eae\u5ea6\u7b49\u64cd\u4f5c，\u4ee5\u6a21\u62df\u540c\u4e00\u4e2a\u82b1\u5349\u5728\u4e0d\u540c\u7684\u73af\u5883\u548c\u62cd\u6444\u89d2\u5ea6\u4e0b\u7684\u89c6\u89c9\u5448\u73b0，\u5c31\u50cf\u4eba\u8138\u5728\u660e\u4eae\u548c\u660f\u6697\u7684\u60c5\u51b5\u4e0b，\u6a21\u578b\u662f\u5426\u80fd\u6b63\u786e\u8bc6\u522b\u8fd9\u662f\u540c\u4e00\u4eba\u3002<\/p>\n

    \u7136\u800c\u6211\u4eec\u53ea\u5e0c\u671b\u8fd9\u6837\u7684\u53d8\u6362\u5e94\u7528\u5728\u8bad\u7ec3\u96c6\u4e2d，\u800c\u5bf9\u6d4b\u8bd5\u96c6\u548c\u9a8c\u8bc1\u96c6\u65e0\u6548，\u56e0\u4e3a\u5982\u679c\u5bf9\u6d4b\u8bd5\u96c6\u548c\u9a8c\u8bc1\u96c6\u5e94\u7528\u76f8\u540c\u7684\u53d8\u6362，\u5c31\u65e0\u6cd5\u786e\u5b9a\u6bcf\u4e00\u6b21epoch，\u6a21\u578b\u7684\u8bad\u7ec3\u7ed3\u679c\u7684\u53d8\u5316\u4ec5\u4ec5\u662f\u56e0\u4e3a\u8f93\u5165\u4e0d\u540c\u8fd8\u662f\u56e0\u4e3a\u6a21\u578b\u672c\u8eab\u7684\u6027\u80fd\u5f97\u5230\u6539\u53d8\u3002\u5728\u6211\u4eec\u73b0\u6709\u7684\u67b6\u6784\u4e2d，\u65e0\u6cd5\u505a\u5230\u5bf9\u4e0d\u540c\u7684\u6570\u636e\u96c6\u4f7f\u7528\u4e0d\u540c\u7684transform，\u56e0\u4e3a\u6211\u4eec\u7684transform\u53ea\u5728OxfordFlowersDataSet\u7c7b\u88ab\u4f20\u5165，\u968f\u673a\u5206\u914d\u4ea7\u751f\u76843\u4e2a\u6570\u636e\u96c6，\u672c\u8eab\u90fd\u662fdataset\u7684\u5b50\u96c6，\u5e94\u7528\u548cdataset\u4e00\u6837\u7684\u53d8\u6362，\u53ef\u662f\u5982\u679c\u6211\u4eec\u5728\u5f97\u5230\u5b50\u96c6\u4e4b\u540e，\u518d\u6b21\u5e94\u7528\u4e0d\u540c\u7684\u53d8\u6362\u5462<\/p>\n

    \u8fd9\u662f\u6211\u4eec\u539f\u6765\u7684\u67b6\u6784<\/p>\n

    #mermaid-svg-CcmAAKhOvDn78Qyo{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-CcmAAKhOvDn78Qyo .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-CcmAAKhOvDn78Qyo .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-CcmAAKhOvDn78Qyo .error-icon{fill:#552222;}#mermaid-svg-CcmAAKhOvDn78Qyo .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-CcmAAKhOvDn78Qyo .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-CcmAAKhOvDn78Qyo .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-CcmAAKhOvDn78Qyo .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-CcmAAKhOvDn78Qyo .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-CcmAAKhOvDn78Qyo .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-CcmAAKhOvDn78Qyo .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-CcmAAKhOvDn78Qyo .marker{fill:#333333;stroke:#333333;}#mermaid-svg-CcmAAKhOvDn78Qyo .marker.cross{stroke:#333333;}#mermaid-svg-CcmAAKhOvDn78Qyo svg{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-CcmAAKhOvDn78Qyo p{margin:0;}#mermaid-svg-CcmAAKhOvDn78Qyo .label{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-CcmAAKhOvDn78Qyo .cluster-label text{fill:#333;}#mermaid-svg-CcmAAKhOvDn78Qyo .cluster-label span{color:#333;}#mermaid-svg-CcmAAKhOvDn78Qyo .cluster-label span p{background-color:transparent;}#mermaid-svg-CcmAAKhOvDn78Qyo .label text,#mermaid-svg-CcmAAKhOvDn78Qyo span{fill:#333;color:#333;}#mermaid-svg-CcmAAKhOvDn78Qyo .node rect,#mermaid-svg-CcmAAKhOvDn78Qyo .node circle,#mermaid-svg-CcmAAKhOvDn78Qyo .node ellipse,#mermaid-svg-CcmAAKhOvDn78Qyo .node polygon,#mermaid-svg-CcmAAKhOvDn78Qyo .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-CcmAAKhOvDn78Qyo .rough-node .label text,#mermaid-svg-CcmAAKhOvDn78Qyo .node .label text,#mermaid-svg-CcmAAKhOvDn78Qyo .image-shape .label,#mermaid-svg-CcmAAKhOvDn78Qyo .icon-shape .label{text-anchor:middle;}#mermaid-svg-CcmAAKhOvDn78Qyo .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-CcmAAKhOvDn78Qyo .rough-node .label,#mermaid-svg-CcmAAKhOvDn78Qyo .node .label,#mermaid-svg-CcmAAKhOvDn78Qyo .image-shape .label,#mermaid-svg-CcmAAKhOvDn78Qyo .icon-shape .label{text-align:center;}#mermaid-svg-CcmAAKhOvDn78Qyo .node.clickable{cursor:pointer;}#mermaid-svg-CcmAAKhOvDn78Qyo .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-CcmAAKhOvDn78Qyo .arrowheadPath{fill:#333333;}#mermaid-svg-CcmAAKhOvDn78Qyo .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-CcmAAKhOvDn78Qyo .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-CcmAAKhOvDn78Qyo .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-CcmAAKhOvDn78Qyo .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-CcmAAKhOvDn78Qyo .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-CcmAAKhOvDn78Qyo .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-CcmAAKhOvDn78Qyo .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-CcmAAKhOvDn78Qyo .cluster text{fill:#333;}#mermaid-svg-CcmAAKhOvDn78Qyo .cluster span{color:#333;}#mermaid-svg-CcmAAKhOvDn78Qyo div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-CcmAAKhOvDn78Qyo .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-CcmAAKhOvDn78Qyo rect.text{fill:none;stroke-width:0;}#mermaid-svg-CcmAAKhOvDn78Qyo .icon-shape,#mermaid-svg-CcmAAKhOvDn78Qyo .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-CcmAAKhOvDn78Qyo .icon-shape p,#mermaid-svg-CcmAAKhOvDn78Qyo .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-CcmAAKhOvDn78Qyo .icon-shape .label rect,#mermaid-svg-CcmAAKhOvDn78Qyo .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-CcmAAKhOvDn78Qyo .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-CcmAAKhOvDn78Qyo .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-CcmAAKhOvDn78Qyo :root{–mermaid-font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;}<\/p>\n

    <\/span><\/p>\n

    <\/span><\/p>\n

    <\/span><\/p>\n

    <\/span><\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    transform<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    root<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    OxfordFlowersDataSet<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    train_set<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    valid_set<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    test_set<\/p>\n

    <\/span><\/p>\n

    \u8fd9\u662f\u6211\u4eec\u53ef\u4ee5\u5c1d\u8bd5\u7684\u65b0\u67b6\u6784<\/p>\n

    #mermaid-svg-aiDkG02SXvWH92YW{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-aiDkG02SXvWH92YW .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-aiDkG02SXvWH92YW .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-aiDkG02SXvWH92YW .error-icon{fill:#552222;}#mermaid-svg-aiDkG02SXvWH92YW .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-aiDkG02SXvWH92YW .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-aiDkG02SXvWH92YW .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-aiDkG02SXvWH92YW .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-aiDkG02SXvWH92YW .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-aiDkG02SXvWH92YW .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-aiDkG02SXvWH92YW .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-aiDkG02SXvWH92YW .marker{fill:#333333;stroke:#333333;}#mermaid-svg-aiDkG02SXvWH92YW .marker.cross{stroke:#333333;}#mermaid-svg-aiDkG02SXvWH92YW svg{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-aiDkG02SXvWH92YW p{margin:0;}#mermaid-svg-aiDkG02SXvWH92YW .label{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-aiDkG02SXvWH92YW .cluster-label text{fill:#333;}#mermaid-svg-aiDkG02SXvWH92YW .cluster-label span{color:#333;}#mermaid-svg-aiDkG02SXvWH92YW .cluster-label span p{background-color:transparent;}#mermaid-svg-aiDkG02SXvWH92YW .label text,#mermaid-svg-aiDkG02SXvWH92YW span{fill:#333;color:#333;}#mermaid-svg-aiDkG02SXvWH92YW .node rect,#mermaid-svg-aiDkG02SXvWH92YW .node circle,#mermaid-svg-aiDkG02SXvWH92YW .node ellipse,#mermaid-svg-aiDkG02SXvWH92YW .node polygon,#mermaid-svg-aiDkG02SXvWH92YW .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-aiDkG02SXvWH92YW .rough-node .label text,#mermaid-svg-aiDkG02SXvWH92YW .node .label text,#mermaid-svg-aiDkG02SXvWH92YW .image-shape .label,#mermaid-svg-aiDkG02SXvWH92YW .icon-shape .label{text-anchor:middle;}#mermaid-svg-aiDkG02SXvWH92YW .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-aiDkG02SXvWH92YW .rough-node .label,#mermaid-svg-aiDkG02SXvWH92YW .node .label,#mermaid-svg-aiDkG02SXvWH92YW .image-shape .label,#mermaid-svg-aiDkG02SXvWH92YW .icon-shape .label{text-align:center;}#mermaid-svg-aiDkG02SXvWH92YW .node.clickable{cursor:pointer;}#mermaid-svg-aiDkG02SXvWH92YW .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-aiDkG02SXvWH92YW .arrowheadPath{fill:#333333;}#mermaid-svg-aiDkG02SXvWH92YW .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-aiDkG02SXvWH92YW .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-aiDkG02SXvWH92YW .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-aiDkG02SXvWH92YW .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-aiDkG02SXvWH92YW .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-aiDkG02SXvWH92YW .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-aiDkG02SXvWH92YW .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-aiDkG02SXvWH92YW .cluster text{fill:#333;}#mermaid-svg-aiDkG02SXvWH92YW .cluster span{color:#333;}#mermaid-svg-aiDkG02SXvWH92YW div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-aiDkG02SXvWH92YW .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-aiDkG02SXvWH92YW rect.text{fill:none;stroke-width:0;}#mermaid-svg-aiDkG02SXvWH92YW .icon-shape,#mermaid-svg-aiDkG02SXvWH92YW .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-aiDkG02SXvWH92YW .icon-shape p,#mermaid-svg-aiDkG02SXvWH92YW .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-aiDkG02SXvWH92YW .icon-shape .label rect,#mermaid-svg-aiDkG02SXvWH92YW .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-aiDkG02SXvWH92YW .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-aiDkG02SXvWH92YW .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-aiDkG02SXvWH92YW :root{–mermaid-font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;}<\/p>\n

    <\/span><\/p>\n

    <\/span><\/p>\n

    <\/span><\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    transform_strengthen<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    transform_base<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    transform_base<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    transform_strengthen<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    transform_base<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    transform_base<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    root<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    OxfordFlowersDataSet<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    SubSet<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    train_set<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    valid_set<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    test_set<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    train_set<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    valid_set<\/p>\n

    <\/span><\/p>\n

    <\/p>\n

    test_set<\/p>\n

    <\/span><\/p>\n

    \u6211\u4eec\u5c06\u83b7\u53d6\u4e0e\u5e94\u7528\u5206\u79bb，OxfordFlowersDataSet\u7c7b\u4ec5\u63d0\u4f9b\u4ece\u539f\u6587\u4ef6\u5939\u83b7\u53d6\u8d44\u6e90（\u4f20\u5165root\u6587\u4ef6\u5939\u8def\u5f84\u6784\u9020），\u800cSubSet\u7c7b\u4e13\u6ce8\u4e8e\u8d44\u6e90\u7684\u52a0\u5de5（\u4f20\u5165OxfordFlowersDataSet\u7c7b\u5206\u79bb\u51fa\u6765\u7684\u6570\u636e\u96c6\u4f5c\u4e3a\u52a0\u5de5\u5bf9\u8c61，\u4f20\u5165transform\u4f5c\u4e3a\u5e94\u7528\u53d8\u6362），\u8fd9\u6837\u5c31\u53ef\u4ee5\u4e3a\u4e09\u4e2a\u6570\u636e\u96c6\u5e94\u7528\u4e0d\u540c\u7684transform<\/p>\n

    \u8ba9\u6211\u4eec\u6765\u8bd5\u8bd5<\/p>\n

    class<\/span> OxfordFlowersDataSet<\/span>(<\/span>Dataset)<\/span>:<\/span>
    \n def<\/span> __init__<\/span>(<\/span>self,<\/span> root,<\/span> target_transform=<\/span>None<\/span>)<\/span>:<\/span>
    \n self.<\/span>root =<\/span> root
    \n self.<\/span>img_dir =<\/span> os.<\/span>path.<\/span>join(<\/span>self.<\/span>root,<\/span>"img"<\/span>,<\/span>"jpg"<\/span>)<\/span><\/p>\n

    labels_mat =<\/span> scipy.<\/span>io.<\/span>loadmat(<\/span>os.<\/span>path.<\/span>join(<\/span>self.<\/span>root,<\/span>"imagelabels.mat"<\/span>)<\/span>)<\/span><\/p>\n

    self.<\/span>labels =<\/span> labels_mat[<\/span>"labels"<\/span>]<\/span>[<\/span>0<\/span>]<\/span> –<\/span> 1<\/span>
    \n def<\/span> __len__<\/span>(<\/span>self)<\/span>:<\/span>
    \n return<\/span> len<\/span>(<\/span>self.<\/span>labels)<\/span>
    \n def<\/span> __getitem__<\/span>(<\/span>self,<\/span> index)<\/span>:<\/span>
    \n img_name =<\/span> f'image_<\/span>{<\/span>index+<\/span>1<\/span>:<\/span>05d<\/span>}<\/span><\/span>.jpg'<\/span><\/span>
    \n img_path =<\/span> os.<\/span>path.<\/span>join(<\/span>self.<\/span>img_dir,<\/span>img_name)<\/span><\/p>\n

    img =<\/span> Image.<\/span>open<\/span>(<\/span>img_path)<\/span>
    \n labels =<\/span> self.<\/span>labels[<\/span>index]<\/span><\/p>\n

    return<\/span> img,<\/span> labels
    \n def<\/span> info<\/span>(<\/span>self)<\/span>:<\/span>
    \n print<\/span>(<\/span>len<\/span>(<\/span>self.<\/span>labels)<\/span>)<\/span>
    \n print<\/span>(<\/span>f"Max label: <\/span>{<\/span>max<\/span>(<\/span>self.<\/span>labels)<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>f"Min label: <\/span>{<\/span>min<\/span>(<\/span>self.<\/span>labels)<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span><\/p>\n

    class<\/span> SubSet<\/span>(<\/span>Dataset)<\/span>:<\/span>
    \n def<\/span> __init__<\/span>(<\/span>self,<\/span> dataset,<\/span>transform =<\/span> None<\/span>)<\/span>:<\/span>
    \n self.<\/span>dataset =<\/span> dataset
    \n self.<\/span>transform =<\/span> transform<\/p>\n

    def<\/span> __len__<\/span>(<\/span>self)<\/span>:<\/span>
    \n return<\/span> len<\/span>(<\/span>self.<\/span>dataset)<\/span><\/p>\n

    def<\/span> __getitem__<\/span>(<\/span>self,<\/span> index)<\/span>:<\/span>
    \n img,<\/span>label =<\/span> self.<\/span>dataset[<\/span>index]<\/span>
    \n if<\/span> self.<\/span>transform is<\/span> not<\/span> None<\/span>:<\/span>
    \n img =<\/span> self.<\/span>transform(<\/span>img)<\/span>
    \n return<\/span> img,<\/span>label<\/p>\n

    from<\/span> sympy import<\/span> false
    \nfrom<\/span> torch.<\/span>utils.<\/span>data import<\/span> DataLoader
    \nfrom<\/span> torchvision import<\/span> transforms
    \nimport<\/span> data_prepare
    \nfrom<\/span> data_prepare import<\/span> OxfordFlowersDataSet
    \nfrom<\/span> data_prepare import<\/span> SubSet
    \nfrom<\/span> torch.<\/span>utils.<\/span>data import<\/span> random_split<\/p>\n

    transform_strengthen =<\/span> transforms.<\/span>Compose(<\/span>[<\/span>
    \n transforms.<\/span>RandomHorizontalFlip(<\/span>p=<\/span>0.5<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>RandomRotation(<\/span>degrees=<\/span>10<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>ColorJitter(<\/span>brightness=<\/span>0.2<\/span>)<\/span>,<\/span><\/p>\n

    transforms.<\/span>Resize(<\/span>256<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>CenterCrop(<\/span>224<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>ToTensor(<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>Normalize(<\/span>mean=<\/span>[<\/span>0.485<\/span>,<\/span> 0.456<\/span>,<\/span> 0.406<\/span>]<\/span>,<\/span> std=<\/span>[<\/span>0.229<\/span>,<\/span> 0.224<\/span>,<\/span> 0.225<\/span>]<\/span>)<\/span>,<\/span>
    \n]<\/span>)<\/span><\/p>\n

    transform_base =<\/span> transforms.<\/span>Compose(<\/span>[<\/span>
    \n transforms.<\/span>Resize(<\/span>256<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>CenterCrop(<\/span>224<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>ToTensor(<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>Normalize(<\/span>mean=<\/span>[<\/span>0.485<\/span>,<\/span> 0.456<\/span>,<\/span> 0.406<\/span>]<\/span>,<\/span> std=<\/span>[<\/span>0.229<\/span>,<\/span> 0.224<\/span>,<\/span> 0.225<\/span>]<\/span>)<\/span>,<\/span>
    \n]<\/span>)<\/span><\/p>\n

    data_prepare.<\/span>download_dataset(<\/span>)<\/span>
    \ndataset =<\/span> OxfordFlowersDataSet(<\/span>'flowers'<\/span>)<\/span><\/p>\n

    train_data_size =<\/span> int<\/span>(<\/span>len<\/span>(<\/span>dataset)<\/span> *<\/span> 0.7<\/span>)<\/span>
    \nvalid_data_size =<\/span> int<\/span>(<\/span>len<\/span>(<\/span>dataset)<\/span> *<\/span> 0.15<\/span>)<\/span>
    \ntest_data_size =<\/span> len<\/span>(<\/span>dataset)<\/span> –<\/span> train_data_size –<\/span> valid_data_size<\/p>\n

    train_data,<\/span>valid_data,<\/span>test_data =<\/span> random_split(<\/span>dataset,<\/span>[<\/span>train_data_size,<\/span>valid_data_size,<\/span>test_data_size]<\/span>)<\/span><\/p>\n

    train_set =<\/span> SubSet(<\/span>train_data,<\/span>transform_strengthen)<\/span>
    \nvalid_set =<\/span> SubSet(<\/span>valid_data,<\/span>transform_base)<\/span>
    \ntest_set =<\/span> SubSet(<\/span>test_data,<\/span>transform_base)<\/span><\/p>\n

    data_batch =<\/span> DataLoader(<\/span>dataset=<\/span>train_set,<\/span>batch_size=<\/span>32<\/span>,<\/span>shuffle=<\/span>True<\/span>)<\/span>
    \nvalid_batch =<\/span> DataLoader(<\/span>dataset=<\/span>valid_set,<\/span>batch_size=<\/span>32<\/span>,<\/span>shuffle=<\/span>False<\/span>)<\/span>
    \ntest_batch =<\/span> DataLoader(<\/span>dataset=<\/span>test_set,<\/span>batch_size=<\/span>32<\/span>,<\/span>shuffle=<\/span>False<\/span>)<\/span><\/p>\n

    print<\/span>(<\/span>f'Train Batch size: <\/span>{<\/span>len<\/span>(<\/span>train_data)<\/span>}<\/span><\/span>'<\/span><\/span>)<\/span>
    \nprint<\/span>(<\/span>f'Validation Batch size: <\/span>{<\/span>len<\/span>(<\/span>valid_data)<\/span>}<\/span><\/span>'<\/span><\/span>)<\/span>
    \nprint<\/span>(<\/span>f'Test Batch size: <\/span>{<\/span>len<\/span>(<\/span>test_data)<\/span>}<\/span><\/span>'<\/span><\/span>)<\/span>
    \nfor<\/span> index,<\/span>(<\/span>images,<\/span>labels)<\/span> in<\/span> enumerate<\/span>(<\/span>data_batch)<\/span>:<\/span>
    \n if<\/span> index ><\/span> 175<\/span>:<\/span>
    \n print<\/span>(<\/span>f'the <\/span>{<\/span>index}<\/span><\/span>th batch size is <\/span>{<\/span>images.<\/span>shape}<\/span><\/span>'<\/span><\/span>)<\/span><\/p>\n

    E:\\\\Users\\\\ASUS\\\\anaconda3\\\\envs\\\\aiAlgorithm\\\\python.<\/span>exe E:\\\\pythonProject\\\\PythonProject\\\\data_handle.<\/span>py
    \nTrain Batch size: 5732
    \nValidation Batch size: 1228
    \nTest Batch size: 1229
    \nthe 176th batch size is torch.<\/span>Size(<\/span>[<\/span>32,<\/span> 3,<\/span> 224,<\/span> 224]<\/span>)<\/span>
    \nthe 177th batch size is torch.<\/span>Size(<\/span>[<\/span>32,<\/span> 3,<\/span> 224,<\/span> 224]<\/span>)<\/span>
    \nthe 178th batch size is torch.<\/span>Size(<\/span>[<\/span>32,<\/span> 3,<\/span> 224,<\/span> 224]<\/span>)<\/span>
    \nthe 179th batch size is torch.<\/span>Size(<\/span>[<\/span>4,<\/span> 3,<\/span> 224,<\/span> 224]<\/span>)<\/span><\/p>\n

    \u7ed3\u679c\u8868\u793a\u8fd9\u786e\u5b9e\u53ef\u4ee5\u88ab\u5e94\u7528，\u4f46\u662f\u95ee\u9898\u65f6，\u6211\u4eec\u4ecd\u7136\u4e0d\u77e5\u9053transform.Compose([\u2026])\u5de5\u4f5c\u6d41\u5185\u90e8\u5230\u5e95\u53d1\u751f\u4e86\u4ec0\u4e48，\u4ed6\u4eec\u771f\u7684\u88ab\u5e94\u7528\u589e\u5f3a\u4e86\u5417？\u6700\u597d\u7684\u65b9\u6cd5\u662f\u5199\u4e00\u4e2a\u53ef\u89c6\u5316\u51fd\u6570\u4eb2\u81ea\u770b\u770b\u56fe\u50cf\u88ab\u589e\u5f3a\u540e\u7684\u6548\u679c<\/p>\n

    \u53ef\u89c6\u5316\u56fe\u50cf\u9a8c\u8bc1<\/h4>\n

    \u6211\u4eec\u5c06\u5e94\u7528\u7684\u56fe\u50cf\u8f6c\u6362\u6210\u4e86\u5f20\u91cf，\u5e76\u8fdb\u884c\u4e86\u5f52\u4e00\u5316，\u5982\u679c\u6211\u4eec\u4f7f\u7528\u8f6c\u6362\u540e\u7684\u56fe\u50cf\u4f5c\u4e3aimshow\u7684\u53c2\u6570\u663e\u793a，\u8fd9\u4f1a\u9020\u6210\u8272\u5f69\u5931\u771f，\u56e0\u4e3aimshow\u662f\u6839\u636e\u503c\u6e32\u67d3\u56fe\u50cf\u7684，\u6211\u4eec\u4e4b\u524d\u63d0\u5230，\u6570\u503c\u672c\u8eab\u5bf9\u4e8e\u6a21\u578b\u5e76\u6ca1\u6709\u610f\u4e49，\u800c\u662f\u6211\u4eec\u8d4b\u4e88\u7684\u610f\u4e49，\u4ed6\u6240\u505a\u7684\u5de5\u4f5c\u662f\u4ece\u6570\u503c\u7684\u5173\u7cfb\u4e2d\u627e\u5230\u7279\u5f81，\u7136\u540e\u8bc6\u522b，imshow\u9700\u8981\u5c55\u793a\u4eba\u53ef\u4ee5\u8fa8\u8ba4\u7684\u8272\u5f69，\u8fd9\u4e2a\u65f6\u5019\u503c\u5bf9\u4e8e\u6211\u4eec\u6765\u8bf4\u662f\u6709\u610f\u4e49\u7684\u3010\u6216\u8005\u8981\u7ed9\u4ed6\u7f29\u653e\u5728\u4e00\u4e2a\u4eba\u4e3a\u89c4\u5b9a\u7684\u6807\u51c6\u4e0b\u3011，\u6240\u4ee5\u6211\u4eec\u9700\u8981\u5bf9\u539f\u672c\u7684\u56fe\u50cf\u53cd\u5f52\u4e00\u5316<\/p>\n

    \u60a8\u53ef\u4ee5\u67e5\u9605\u6211\u5728\u7b2c\u4e8c\u7ae0\u2014\u2014\u6570\u636e\u6444\u53d6 & \u6570\u636e\u51c6\u5907\u5bf9\u4e8e\u5f52\u4e00\u5316\u7684\u89e3\u91ca，\u5728\u8fd9\u91cc\u5c06\u5f52\u4e00\u5316\u7684\u516c\u5f0f\u505a\u4e00\u4e2a\u9006\u5411\u63a8\u5bfc，\u5c31\u662f\u6211\u4eec\u5c06\u4f1a\u4f7f\u7528\u5230\u7684\u53cd\u5f52\u4e00\u5316 <\/p>\n

    n<\/p>\n

    o<\/p>\n

    r<\/p>\n

    m<\/p>\n

    a<\/p>\n

    l<\/p>\n

    i<\/p>\n

    z<\/p>\n

    e
    \n \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    o<\/p>\n

    r<\/p>\n

    i<\/p>\n

    g<\/p>\n

    i<\/p>\n

    n
    \n \u2005\u200a<\/p>\n

    \u2212
    \n \u2005\u200a<\/p>\n

    m<\/p>\n

    e<\/p>\n

    a<\/p>\n

    n<\/p>\n

    s<\/p>\n

    t<\/p>\n

    d<\/p>\n

    o<\/p>\n

    r<\/p>\n

    i<\/p>\n

    g<\/p>\n

    i<\/p>\n

    n
    \n \u2005\u200a<\/p>\n

    =
    \n \u2005\u200a<\/p>\n

    (<\/p>\n

    s<\/p>\n

    t<\/p>\n

    d
    \n \u2005\u200a<\/p>\n

    \u2217
    \n \u2005\u200a<\/p>\n

    n<\/p>\n

    o<\/p>\n

    r<\/p>\n

    m<\/p>\n

    a<\/p>\n

    l<\/p>\n

    i<\/p>\n

    z<\/p>\n

    e<\/p>\n

    )
    \n \u2005\u200a<\/p>\n

    +
    \n \u2005\u200a<\/p>\n

    m<\/p>\n

    e<\/p>\n

    a<\/p>\n

    n<\/p>\n

    normalize\\\\;=\\\\;\\\\dfrac{origin\\\\;-\\\\;mean}{std}\\\\\\\\ origin\\\\;=\\\\;(std\\\\;*\\\\;normalize)\\\\;+\\\\;mean <\/p>\n

    <\/span><\/span>n<\/span>or<\/span>ma<\/span>l<\/span>i<\/span>z<\/span>e<\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span><\/span><\/span>s<\/span>t<\/span>d<\/span><\/span><\/span><\/span><\/span><\/span><\/span>or<\/span>i<\/span>g<\/span>in<\/span><\/span><\/span>\u2212<\/span><\/span><\/span>m<\/span>e<\/span>an<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>or<\/span>i<\/span>g<\/span>in<\/span><\/span><\/span>=<\/span><\/span><\/span><\/span><\/span>(<\/span>s<\/span>t<\/span>d<\/span><\/span><\/span>\u2217<\/span><\/span><\/span><\/span><\/span>n<\/span>or<\/span>ma<\/span>l<\/span>i<\/span>z<\/span>e<\/span>)<\/span><\/span><\/span>+<\/span><\/span><\/span><\/span><\/span>m<\/span>e<\/span>an<\/span><\/span><\/span><\/span><\/span><\/span> \u5229\u7528\u7b2c\u4e8c\u4e2a\u9006\u63a8\u5bfc，\u6211\u4eec\u53ef\u4ee5\u5c06\u5f52\u4e00\u5316\u540e\u7684\u56fe\u50cf\u8f6c\u6362\u56de\u539f\u56fe\u50cf，\u4f46\u662f\u5728\u6b64\u4e4b\u524d，\u6211\u4eec\u8fd8\u9700\u8981\u505a\u4e00\u4e9b\u9884\u5904\u7406<\/p>\n

    \u5bf9\u4e8emean\u548cstd\u6765\u8bf4，\u5b83\u4eec\u662f\u5217\u8868\u7c7b\u578b\u65e0\u6cd5\u4e0eimg:Tensor\u7684\u7c7b\u578b\u505a\u70b9\u4e58，\u6211\u4eec\u9700\u8981\u628a\u5b83\u4eec\u7684\u7c7b\u578b\u4e5f\u53d8\u6362\u4e3aTensor，tensor()\u63d0\u4f9b\u4e86\u8fd9\u6837\u7684\u8f6c\u6362，\u5c31\u50cfint(double)\u53ef\u4ee5\u8f6c\u6362\u4e3adouble\u4e3aint\u90a3\u6837，\u6211\u4eec\u53ef\u4ee5\u5c1d\u8bd5\u6253\u5370\u8f6c\u6362\u540e\u7684shape，\u5b83\u4f1a\u663e\u793atorch.Size([3])，\u8fd9\u4ecd\u7136\u65e0\u6cd5\u4e0etorch,Size([3,224,224])\u5927\u5c0f\u7684\u5f20\u91cf\u8fdb\u884c\u70b9\u4e58 \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

    \u7136\u800c\u6211\u4eec\u53ef\u4ee5\u5c06mean\u548cstd\u8f6c\u6362\u4e3a[3,1,1]\u7684\u5927\u5c0f，\u6211\u4eec\u4f7f\u7528Tensor.view()\u65b9\u6cd5，\u7b2c\u4e00\u4e2a\u53c2\u6570\u4e3a -1 \u8868\u793a\u81ea\u52a8\u8ba1\u7b97\u5927\u5c0f，\u7b2c\u4e8c\u4e2a\u53c2\u6570\u548c\u7b2c\u4e09\u4e2a\u53c2\u6570\u4e3a1，\u8868\u793a\u8f6c\u6362\u540e\u5f20\u91cf\u7684\u7b2c\u4e8c\u5c42（\u884c）\u548c\u7b2c\u4e09\u5c42（\u5217）\u7684\u5927\u5c0f\u5404\u4f4d1，3 = 1 * 1 * 3，\u7ecf\u8ba1\u7b97\u540e，\u5f20\u91cf\u7684\u4e00\u5c42\u5927\u5c0f\u4e3a3（channel），\u8fd9\u770b\u6765\u4e5f\u65e0\u6cd5\u8fdb\u884c\u70b9\u4e58<\/p>\n

    \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

    \u4f46\u662f\u60a8\u8fd8\u8bb0\u5f97\u5f20\u91cf\u7684\u5e7f\u64ad\u673a\u5236\u561b，\u5f20\u91cf\u5728\u8fdb\u884c\u8fd0\u7b97\u7684\u65f6\u5019\u5982\u679c\u9047\u5230\u884c\u5217\u4e0d\u5339\u914d\u7684\u60c5\u51b5\u4f1a\u5c1d\u8bd5\u5411\u884c\u5217\u8fdb\u884c\u5e7f\u64ad，\u7b2c\u4e8c\u5c42\u4e3a1\u4f1a\u5e7f\u64ad\u4e3aimg\u7684\u7b2c\u4e8c\u5c42\u5927\u5c0f，\u7b2c\u4e09\u5c42\u4e3a1\u4f1a\u5e7f\u64ad\u4e3aimg\u7684\u7b2c\u4e09\u5c42\u7684\u5927\u5c0f，\u5177\u4f53\u53ef\u89c1\u5f20\u91cf\u7684\u5e7f\u64ad\u673a\u5236，\u6240\u4ee5\u5728\u8ba1\u7b97\u65f6，mean\u548cstd\u90fd\u4f1a\u88ab\u5e7f\u64ad\u6210[3,224,224]\u5927\u5c0f \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

    from<\/span> matplotlib import<\/span> pyplot as<\/span> plt
    \nimport<\/span> torch
    \nfrom<\/span> torch import<\/span> tensor<\/p>\n

    from<\/span> data_prepare import<\/span> *<\/span>
    \nfrom<\/span> torch.<\/span>utils.<\/span>data import<\/span> random_split
    \nfrom<\/span> data_handle import<\/span> transform_strengthen
    \nfrom<\/span> torchvision import<\/span> transforms<\/p>\n

    def<\/span> denormalize<\/span>(<\/span>img,<\/span>mean,<\/span>std)<\/span>:<\/span>
    \n img =<\/span> img.<\/span>clone(<\/span>)<\/span>.<\/span>detach(<\/span>)<\/span>.<\/span>cpu(<\/span>)<\/span>
    \n mean =<\/span> tensor(<\/span>mean)<\/span>.<\/span>view(<\/span>–<\/span>1<\/span>,<\/span>1<\/span>,<\/span>1<\/span>)<\/span>
    \n std =<\/span> tensor(<\/span>std)<\/span>.<\/span>view(<\/span>–<\/span>1<\/span>,<\/span>1<\/span>,<\/span>1<\/span>)<\/span>
    \n img =<\/span> img*<\/span>std +<\/span> mean
    \n return<\/span> img<\/p>\n

    def<\/span> vision_strengthen<\/span>(<\/span>img_set,<\/span>strengthen_set,<\/span>title,<\/span>count=<\/span>8<\/span>)<\/span>:<\/span>
    \n img_set =<\/span> img_set<\/p>\n

    fig,<\/span>axes =<\/span> plt.<\/span>subplots(<\/span>nrows=<\/span>2<\/span>,<\/span>ncols=<\/span>count,<\/span>figsize=<\/span>(<\/span>12<\/span>,<\/span>6<\/span>)<\/span>)<\/span>
    \n axes =<\/span> axes.<\/span>flatten(<\/span>)<\/span><\/p>\n

    for<\/span> index in<\/span> range<\/span>(<\/span>count)<\/span>:<\/span>
    \n img,<\/span>label =<\/span> img_set[<\/span>index]<\/span>
    \n img_trans,<\/span>label_trans =<\/span> strengthen_set[<\/span>index]<\/span>
    \n img =<\/span> transforms.<\/span>ToTensor(<\/span>)<\/span>(<\/span>img)<\/span>
    \n img_trans =<\/span> denormalize(<\/span>img_trans,<\/span>mean=<\/span>[<\/span>0.485<\/span>,<\/span> 0.456<\/span>,<\/span> 0.406<\/span>]<\/span>,<\/span>std=<\/span>[<\/span>0.229<\/span>,<\/span> 0.224<\/span>,<\/span> 0.225<\/span>]<\/span>)<\/span>
    \n axes[<\/span>index]<\/span>.<\/span>imshow(<\/span>img.<\/span>permute(<\/span>1<\/span>,<\/span>2<\/span>,<\/span>0<\/span>)<\/span>)<\/span>
    \n axes[<\/span>count +<\/span> index]<\/span>.<\/span>imshow(<\/span>img_trans.<\/span>permute(<\/span>1<\/span>,<\/span>2<\/span>,<\/span>0<\/span>)<\/span>)<\/span><\/p>\n

    axes[<\/span>index]<\/span>.<\/span>axis(<\/span>'off'<\/span>)<\/span>
    \n axes[<\/span>index]<\/span>.<\/span>set_title(<\/span>str<\/span>(<\/span>label)<\/span>)<\/span>
    \n axes[<\/span>count +<\/span> index]<\/span>.<\/span>axis(<\/span>'off'<\/span>)<\/span>
    \n axes[<\/span>count +<\/span> index]<\/span>.<\/span>set_title(<\/span>str<\/span>(<\/span>label_trans)<\/span>)<\/span>
    \n plt.<\/span>suptitle(<\/span>title)<\/span>
    \n plt.<\/span>tight_layout(<\/span>)<\/span>
    \n plt.<\/span>show(<\/span>)<\/span><\/p>\n

    originSet =<\/span> OxfordFlowersDataSet(<\/span>"flowers"<\/span>)<\/span><\/p>\n

    trainSet_size =<\/span> int<\/span>(<\/span>len<\/span>(<\/span>originSet)<\/span> *<\/span> 0.7<\/span>)<\/span>
    \nvalidSet_size =<\/span> int<\/span>(<\/span>len<\/span>(<\/span>originSet)<\/span> *<\/span> 0.15<\/span>)<\/span>
    \ntestSet_size =<\/span> len<\/span>(<\/span>originSet)<\/span> –<\/span> trainSet_size –<\/span> validSet_size<\/p>\n

    originTrain,<\/span>originValid,<\/span>originTest =<\/span> random_split(<\/span>originSet,<\/span>[<\/span>trainSet_size,<\/span>validSet_size,<\/span>testSet_size]<\/span>)<\/span>
    \ntransformTrain =<\/span> SubSet(<\/span>originTrain,<\/span>transform_strengthen)<\/span><\/p>\n

    vision_strengthen(<\/span>originTrain,<\/span>transformTrain,<\/span>"train-origin"<\/span>)<\/span><\/p>\n

    \u4ece\u56fe\u50cf\u7684\u7ed3\u679c\u6765\u770b，\u6211\u4eec\u5728\u8bad\u7ec3\u96c6\u4e0a\u7684\u5e94\u7528\u589e\u5f3a\u7684\u786e\u6709\u6548<\/p>\n

    \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

    \u9632\u6f0f\u6d1e\u90e8\u7f72<\/h4>\n

    \u5982\u679c\u5728\u6a21\u578b\u8bad\u7ec3\u7684\u8fc7\u7a0b\u4e2d\u53d1\u751f\u56fe\u7247\u635f\u574f\u6216\u8005\u683c\u5f0f\u95ee\u9898\u800c\u5f15\u53d1\u7684\u5f02\u5e38\u5bfc\u81f4\u7a0b\u5e8f\u7ec8\u6b62，\u8fd9\u662f\u4ee4\u4eba\u5d29\u6e83\u7684\u95ee\u9898，\u6a21\u578b\u8bad\u7ec3\u662f\u4e00\u4e2a\u6301\u4e45\u7684\u8fc7\u7a0b，\u8fd9\u610f\u5473\u4fee\u590d\u9519\u8bef，\u4ece\u5934\u5f00\u59cb\u8bad\u7ec3，\u9632\u6f0f\u6d1e\u90e8\u7f72\u6b63\u662f\u4e3a\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\u800c\u5f15\u51fa\u7684\u65b9\u6848，\u6211\u4eec\u4f1a\u60f3\u8981\u5982\u4f55\u6392\u67e5\u9519\u8bef\u5462？\u7528if else\u5c06\u6bcf\u4e2a\u53ef\u80fd\u51fa\u9519\u7684\u60c5\u51b5\u5217\u51fa\u6765\u5904\u7406\u561b，\u5bf9\u4e8e\u5e38\u89c1\u7684\u9519\u8bef\u8fd9\u662f\u6709\u6548\u7684，\u4f46\u662f\u5982\u679c\u9519\u8bef\u5b9e\u5728\u975e\u5e38\u9690\u853d\u8fd8\u662f\u4f1a\u56e0\u4e3a\u6ca1\u6709\u8003\u8651\u5230\u6240\u6709\u60c5\u51b5\u800c\u5bfc\u81f4\u5f02\u5e38\u629b\u51fa，\u6211\u4eec\u53ef\u4ee5\u91c7\u7528\u4e00\u79cd\u8bb0\u5f55\u4f46\u4e0d\u5904\u7406\u7684\u7b56\u7565<\/p>\n

    try-except \u5f02\u5e38\u6355\u6349\u673a\u5236\u80fd\u591f\u5e2e\u52a9\u6211\u4eec\u5904\u7406\u6240\u6709\u7a0b\u5e8f\u8fd0\u884c\u9519\u8bef\u800c\u5bfc\u81f4\u7684\u5f02\u5e38<\/p>\n

    class<\/span> OxfordFlowersDataSet<\/span>(<\/span>Dataset)<\/span>:<\/span>
    \n def<\/span> __init__<\/span>(<\/span>self,<\/span> root,<\/span> target_transform=<\/span>None<\/span>)<\/span>:<\/span>
    \n self.<\/span>root =<\/span> root
    \n self.<\/span>img_dir =<\/span> os.<\/span>path.<\/span>join(<\/span>self.<\/span>root,<\/span>"img"<\/span>,<\/span>"jpg"<\/span>)<\/span><\/p>\n

    labels_mat =<\/span> scipy.<\/span>io.<\/span>loadmat(<\/span>os.<\/span>path.<\/span>join(<\/span>self.<\/span>root,<\/span>"imagelabels.mat"<\/span>)<\/span>)<\/span><\/p>\n

    self.<\/span>labels =<\/span> labels_mat[<\/span>"labels"<\/span>]<\/span>[<\/span>0<\/span>]<\/span> –<\/span> 1<\/span>
    \n def<\/span> __len__<\/span>(<\/span>self)<\/span>:<\/span>
    \n return<\/span> len<\/span>(<\/span>self.<\/span>labels)<\/span>
    \n def<\/span> __getitem__<\/span>(<\/span>self,<\/span> index)<\/span>:<\/span>
    \n try<\/span>:<\/span>
    \n img_name =<\/span> f'image_<\/span>{<\/span>index+<\/span>1<\/span>:<\/span>05d<\/span>}<\/span><\/span>.jpg'<\/span><\/span>
    \n img_path =<\/span> os.<\/span>path.<\/span>join(<\/span>self.<\/span>img_dir,<\/span>img_name)<\/span><\/p>\n

    img =<\/span> Image.<\/span>open<\/span>(<\/span>img_path)<\/span>
    \n labels =<\/span> self.<\/span>labels[<\/span>index]<\/span><\/p>\n

    return<\/span> img,<\/span> labels
    \n except<\/span> Exception as<\/span> e:<\/span>
    \n throw_error(<\/span>e.<\/span>__str__(<\/span>)<\/span>)<\/span>
    \n def<\/span> getImgDir<\/span>(<\/span>self,<\/span>index)<\/span>:<\/span>
    \n return<\/span> os.<\/span>path.<\/span>join(<\/span>self.<\/span>img_dir,<\/span>f'image_<\/span>{<\/span>index+<\/span>1<\/span>:<\/span>05d<\/span>}<\/span><\/span>.jpg'<\/span><\/span>)<\/span>
    \n def<\/span> info<\/span>(<\/span>self)<\/span>:<\/span>
    \n print<\/span>(<\/span>len<\/span>(<\/span>self.<\/span>labels)<\/span>)<\/span>
    \n print<\/span>(<\/span>f"Max label: <\/span>{<\/span>max<\/span>(<\/span>self.<\/span>labels)<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>f"Min label: <\/span>{<\/span>min<\/span>(<\/span>self.<\/span>labels)<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span><\/p>\n

    class<\/span> SubSet<\/span>(<\/span>Dataset)<\/span>:<\/span>
    \n def<\/span> __init__<\/span>(<\/span>self,<\/span> dataset,<\/span>transform =<\/span> None<\/span>)<\/span>:<\/span>
    \n self.<\/span>dataset =<\/span> dataset
    \n self.<\/span>transform =<\/span> transform
    \n self.<\/span>errLog =<\/span> [<\/span>]<\/span><\/p>\n

    def<\/span> __len__<\/span>(<\/span>self)<\/span>:<\/span>
    \n return<\/span> len<\/span>(<\/span>self.<\/span>dataset)<\/span><\/p>\n

    def<\/span> __getitem__<\/span>(<\/span>self,<\/span> index)<\/span>:<\/span>
    \n try<\/span>:<\/span>
    \n img,<\/span>label =<\/span> self.<\/span>dataset[<\/span>index]<\/span>
    \n if<\/span> self.<\/span>transform is<\/span> not<\/span> None<\/span>:<\/span>
    \n img =<\/span> self.<\/span>transform(<\/span>img)<\/span>
    \n return<\/span> img,<\/span>label
    \n except<\/span> Exception as<\/span> e:<\/span>
    \n self.<\/span>errLog.<\/span>append(<\/span>{<\/span>
    \n 'index'<\/span>:<\/span>index,<\/span>
    \n 'error'<\/span>:<\/span>str<\/span>(<\/span>e)<\/span>,<\/span>
    \n 'path'<\/span>:<\/span>"unknow"<\/span> if<\/span> self.<\/span>dataset.<\/span>getImgDir(<\/span>index)<\/span> is<\/span> None<\/span> else<\/span> self.<\/span>dataset.<\/span>getImgDir(<\/span>index)<\/span>,<\/span>
    \n }<\/span>)<\/span><\/p>\n

    # \u9012\u5f52\u8fd4\u56de\u4e0b\u4e00\u6b21\u6570\u636e<\/span>
    \n return<\/span> self.<\/span>__getitem__(<\/span> (<\/span>index+<\/span>1<\/span>)<\/span> %<\/span> len<\/span>(<\/span>self)<\/span>)<\/span><\/p>\n

    \u6211\u4eec\u9009\u62e9\u5728\u53d1\u751f\u5f02\u5e38\u7684\u65f6\u5019\u8bb0\u5f55\u5f02\u5e38\u5e76\u9012\u5f52\u8c03\u7528\u4e0b\u4e00\u6b21\u5c06\u8981\u8bbf\u95ee\u7684\u6570\u636e，\u76f4\u5230\u8fd4\u56de\u6ca1\u6709\u5f02\u5e38\u7684\u6570\u636e，\u8fd9\u4e0d\u4f1a\u518d\u629b\u51fa\u5f02\u5e38\u4e86，\u800c\u662f\u4f1a\u8fd4\u56de\u4e0b\u4e00\u7ec4\u5c06\u8981\u8bbf\u95ee\u6570\u636e\u7684\u7ed3\u679c，\u8fd9\u542c\u8d77\u6765\u5f88\u4e0d\u9519，\u4f46\u662f\u5b83\u4f1a\u9020\u6210\u4e00\u4e2a\u95ee\u9898，\u6211\u4eec\u5e76\u6ca1\u6709\u4e8b\u5148\u901a\u77e5\u6a21\u578b\u6211\u8fd4\u56de\u7684\u662f\u4e0b\u4e00\u7ec4\u6570\u636e，\u60a8\u4e0d\u7528\u518d\u8bbf\u95ee\u5566，\u6a21\u578b\u4f1a\u4e25\u683c\u6309\u7167\u6279\u6b21\u7ee7\u7eed\u8fd4\u56de\u4e0b\u4e00\u4e2a\u6570\u636e，\u5b9e\u9645\u4e0a\u8fd9\u53ef\u80fd\u5bfc\u81f4\u4e00\u7ec4\u6570\u636e\u88ab\u591a\u6b21\u8bbf\u95ee\u3002\u60a8\u53ef\u4ee5\u8003\u8651\u4e3aSubSet\u52a0\u4e00\u4e2abool\u6210\u5458\u5bf9\u8c61\u957f\u5ea6\u662fdataset，\u8bb0\u5f55\u6bcf\u6b21\u662f\u5426\u88ab\u8bbf\u95ee，\u5982\u679c\u88ab\u8bbf\u95ee\u5219\u901a\u77e5\u6a21\u578b\u7ee7\u7eed\u4e0b\u4e00\u6b21，\u5982\u679c\u6ca1\u6709\u5219\u5728return\u4e4b\u524d\u6807\u8bb0\u4e3a\u8bbf\u95ee\u5e76\u6b63\u5e38return<\/p>\n

    if<\/span> visted[<\/span>index]<\/span>:<\/span>
    \n callTonext(<\/span>model)<\/span>
    \n return<\/span>
    \nimg,<\/span>label =<\/span> self.<\/span>dataset[<\/span>index]<\/span>
    \nif<\/span> self.<\/span>transform is<\/span> not<\/span> None<\/span>:<\/span>
    \n img =<\/span> self.<\/span>transform(<\/span>img)<\/span>
    \nvisted[<\/span>index]<\/span> =<\/span> True<\/span>
    \nreturn<\/span> img,<\/span>label<\/p>\n

    \u4e8b\u5b9e\u4e0a，\u6211\u5199\u7684\u66f4\u50cf\u662f\u4f2a\u4ee3\u7801，\u6211\u4e0d\u6e05\u695a\u6a21\u578b\u662f\u5426\u80fd\u5728\u4e0d\u8fd4\u56de\u4efb\u4f55\u6570\u636e\u7684\u60c5\u51b5\u4e0b\u6267\u884c\u4e0b\u4e00\u6b21\u8bad\u7ec3\u800c\u4e0d\u629b\u51fa\u5f02\u5e38，\u5982\u679c\u6211\u4eec\u8fd9\u6837\u505a\u7684\u8bdd，\u662f\u5426\u4f1a\u4e3b\u52a8\u5f15\u53d1\u6a21\u578b\u7684\u5f02\u5e38\u5462，\u5982\u679c\u611f\u5174\u8da3\u7684\u8bdd\u60a8\u53ef\u4ee5\u67e5\u9605\u66f4\u591a\u8d44\u6599\u63a2\u8ba8\u2026\u2026\u4f46\u76ee\u524d\u800c\u8a00，\u8fd9\u662f\u9632\u6f0f\u6d1e\u90e8\u7f72\u4e2d\u6bd4\u8f83\u7a33\u59a5\u7684\u7b56\u7565，\u5373\u4f7f\u53ef\u80fd\u5e26\u6765\u591a\u6b21\u8bbf\u95ee，\u4f46\u662f\u5b83\u53ef\u4ee5\u968f\u7740\u672c\u6b21\u5f02\u5e38\u7684\u89e3\u51b3\u4e00\u8d77\u88ab\u89e3\u51b3，\u5176\u5b9e\u6362\u79cd\u89d2\u5ea6\u601d\u8003，\u5f02\u5e38\u90fd\u629b\u51fa\u4e86，\u6211\u4eec\u5df2\u7ecf\u8ba9\u9519\u8bef\u5f15\u53d1\u7684\u4ee3\u4ef7\u964d\u5230\u5f88\u5c0f\u4e86<\/p>\n

    code<\/h3>\n

    # data_prepare<\/span>
    \nimport<\/span> os
    \nimport<\/span> tarfile
    \nimport<\/span> urllib<\/p>\n

    from<\/span> PIL import<\/span> Image
    \nfrom<\/span> django.<\/span>urls import<\/span> path
    \nfrom<\/span> numpy.<\/span>f2py.<\/span>auxfuncs import<\/span> throw_error
    \nfrom<\/span> sympy.<\/span>codegen.<\/span>ast import<\/span> none
    \nfrom<\/span> sympy.<\/span>stats.<\/span>sampling.<\/span>sample_scipy import<\/span> scipy
    \nimport<\/span> scipy.<\/span>io
    \nfrom<\/span> torch.<\/span>utils.<\/span>data import<\/span> Dataset<\/p>\n

    def<\/span> download_dataset<\/span>(<\/span>)<\/span> –<\/span>><\/span> None<\/span>:<\/span>
    \n image_url =<\/span> "https:\/\/www.robots.ox.ac.uk\/~vgg\/data\/flowers\/102\/102flowers.tgz"<\/span>
    \n labels_url =<\/span> "https:\/\/www.robots.ox.ac.uk\/~vgg\/data\/flowers\/102\/imagelabels.mat"<\/span><\/p>\n

    os.<\/span>makedirs(<\/span>"flowers"<\/span>,<\/span>exist_ok=<\/span>True<\/span>)<\/span>
    \n if<\/span> not<\/span> os.<\/span>path.<\/span>exists(<\/span>os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span>"img"<\/span>,<\/span>"102flowers.tgz"<\/span>)<\/span>)<\/span>:<\/span>
    \n urllib.<\/span>request.<\/span>urlretrieve(<\/span>image_url,<\/span> os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span>"img"<\/span>,<\/span>os.<\/span>path.<\/span>basename(<\/span>image_url)<\/span>)<\/span>)<\/span>
    \n with<\/span> tarfile.<\/span>open<\/span>(<\/span>os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span> "img"<\/span>,<\/span> "102flowers.tgz"<\/span>)<\/span>)<\/span> as<\/span> tg:<\/span>
    \n tg.<\/span>extractall(<\/span>os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span> "img"<\/span>)<\/span>)<\/span>
    \n if<\/span> not<\/span> os.<\/span>path.<\/span>exists(<\/span>os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span>"imagelabels.mat"<\/span>)<\/span>)<\/span>:<\/span>
    \n urllib.<\/span>request.<\/span>urlretrieve(<\/span>labels_url,<\/span> os.<\/span>path.<\/span>join(<\/span>"flowers"<\/span>,<\/span>os.<\/span>path.<\/span>basename(<\/span>labels_url)<\/span>)<\/span>)<\/span><\/p>\n

    class<\/span> OxfordFlowersDataSet<\/span>(<\/span>Dataset)<\/span>:<\/span>
    \n def<\/span> __init__<\/span>(<\/span>self,<\/span> root,<\/span> target_transform=<\/span>None<\/span>)<\/span>:<\/span>
    \n self.<\/span>root =<\/span> root
    \n self.<\/span>img_dir =<\/span> os.<\/span>path.<\/span>join(<\/span>self.<\/span>root,<\/span>"img"<\/span>,<\/span>"jpg"<\/span>)<\/span><\/p>\n

    labels_mat =<\/span> scipy.<\/span>io.<\/span>loadmat(<\/span>os.<\/span>path.<\/span>join(<\/span>self.<\/span>root,<\/span>"imagelabels.mat"<\/span>)<\/span>)<\/span><\/p>\n

    self.<\/span>labels =<\/span> labels_mat[<\/span>"labels"<\/span>]<\/span>[<\/span>0<\/span>]<\/span> –<\/span> 1<\/span>
    \n def<\/span> __len__<\/span>(<\/span>self)<\/span>:<\/span>
    \n return<\/span> len<\/span>(<\/span>self.<\/span>labels)<\/span>
    \n def<\/span> __getitem__<\/span>(<\/span>self,<\/span> index)<\/span>:<\/span>
    \n try<\/span>:<\/span>
    \n img_name =<\/span> f'image_<\/span>{<\/span>index+<\/span>1<\/span>:<\/span>05d<\/span>}<\/span><\/span>.jpg'<\/span><\/span>
    \n img_path =<\/span> os.<\/span>path.<\/span>join(<\/span>self.<\/span>img_dir,<\/span>img_name)<\/span><\/p>\n

    img =<\/span> Image.<\/span>open<\/span>(<\/span>img_path)<\/span>
    \n labels =<\/span> self.<\/span>labels[<\/span>index]<\/span><\/p>\n

    return<\/span> img,<\/span> labels
    \n except<\/span> Exception as<\/span> e:<\/span>
    \n throw_error(<\/span>e.<\/span>__str__(<\/span>)<\/span>)<\/span>
    \n def<\/span> getImgDir<\/span>(<\/span>self,<\/span>index)<\/span>:<\/span>
    \n return<\/span> os.<\/span>path.<\/span>join(<\/span>self.<\/span>img_dir,<\/span>f'image_<\/span>{<\/span>index+<\/span>1<\/span>:<\/span>05d<\/span>}<\/span><\/span>.jpg'<\/span><\/span>)<\/span>
    \n def<\/span> info<\/span>(<\/span>self)<\/span>:<\/span>
    \n print<\/span>(<\/span>len<\/span>(<\/span>self.<\/span>labels)<\/span>)<\/span>
    \n print<\/span>(<\/span>f"Max label: <\/span>{<\/span>max<\/span>(<\/span>self.<\/span>labels)<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>f"Min label: <\/span>{<\/span>min<\/span>(<\/span>self.<\/span>labels)<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span><\/p>\n

    class<\/span> SubSet<\/span>(<\/span>Dataset)<\/span>:<\/span>
    \n def<\/span> __init__<\/span>(<\/span>self,<\/span> dataset,<\/span>transform =<\/span> None<\/span>)<\/span>:<\/span>
    \n self.<\/span>dataset =<\/span> dataset
    \n self.<\/span>transform =<\/span> transform
    \n self.<\/span>errLog =<\/span> [<\/span>]<\/span><\/p>\n

    def<\/span> __len__<\/span>(<\/span>self)<\/span>:<\/span>
    \n return<\/span> len<\/span>(<\/span>self.<\/span>dataset)<\/span><\/p>\n

    def<\/span> __getitem__<\/span>(<\/span>self,<\/span> index)<\/span>:<\/span>
    \n try<\/span>:<\/span>
    \n img,<\/span>label =<\/span> self.<\/span>dataset[<\/span>index]<\/span>
    \n if<\/span> self.<\/span>transform is<\/span> not<\/span> None<\/span>:<\/span>
    \n img =<\/span> self.<\/span>transform(<\/span>img)<\/span>
    \n return<\/span> img,<\/span>label
    \n except<\/span> Exception as<\/span> e:<\/span>
    \n self.<\/span>errLog.<\/span>append(<\/span>{<\/span>
    \n 'index'<\/span>:<\/span>index,<\/span>
    \n 'error'<\/span>:<\/span>str<\/span>(<\/span>e)<\/span>,<\/span>
    \n 'path'<\/span>:<\/span>"unknow"<\/span> if<\/span> self.<\/span>dataset.<\/span>getImgDir(<\/span>index)<\/span> is<\/span> None<\/span> else<\/span> self.<\/span>dataset.<\/span>getImgDir(<\/span>index)<\/span>,<\/span>
    \n }<\/span>)<\/span><\/p>\n

    return<\/span> self.<\/span>__getitem__(<\/span>index+<\/span>1<\/span> %<\/span> len<\/span>(<\/span>self)<\/span>)<\/span><\/p>\n

    # data_handler<\/span>
    \nfrom<\/span> sympy import<\/span> false
    \nfrom<\/span> torch.<\/span>utils.<\/span>data import<\/span> DataLoader
    \nfrom<\/span> torchvision import<\/span> transforms
    \nimport<\/span> data_prepare
    \nfrom<\/span> data_prepare import<\/span> OxfordFlowersDataSet
    \nfrom<\/span> data_prepare import<\/span> SubSet
    \nfrom<\/span> torch.<\/span>utils.<\/span>data import<\/span> random_split<\/p>\n

    transform_strengthen =<\/span> transforms.<\/span>Compose(<\/span>[<\/span>
    \n transforms.<\/span>RandomHorizontalFlip(<\/span>p=<\/span>0.5<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>RandomRotation(<\/span>degrees=<\/span>10<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>ColorJitter(<\/span>brightness=<\/span>0.2<\/span>)<\/span>,<\/span><\/p>\n

    transforms.<\/span>Resize(<\/span>256<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>CenterCrop(<\/span>224<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>ToTensor(<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>Normalize(<\/span>mean=<\/span>[<\/span>0.485<\/span>,<\/span> 0.456<\/span>,<\/span> 0.406<\/span>]<\/span>,<\/span> std=<\/span>[<\/span>0.229<\/span>,<\/span> 0.224<\/span>,<\/span> 0.225<\/span>]<\/span>)<\/span>,<\/span>
    \n]<\/span>)<\/span><\/p>\n

    transform_base =<\/span> transforms.<\/span>Compose(<\/span>[<\/span>
    \n transforms.<\/span>Resize(<\/span>256<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>CenterCrop(<\/span>224<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>ToTensor(<\/span>)<\/span>,<\/span>
    \n transforms.<\/span>Normalize(<\/span>mean=<\/span>[<\/span>0.485<\/span>,<\/span> 0.456<\/span>,<\/span> 0.406<\/span>]<\/span>,<\/span> std=<\/span>[<\/span>0.229<\/span>,<\/span> 0.224<\/span>,<\/span> 0.225<\/span>]<\/span>)<\/span>,<\/span>
    \n]<\/span>)<\/span><\/p>\n

    data_prepare.<\/span>download_dataset(<\/span>)<\/span>
    \ndataset =<\/span> OxfordFlowersDataSet(<\/span>'flowers'<\/span>)<\/span><\/p>\n

    train_data_size =<\/span> int<\/span>(<\/span>len<\/span>(<\/span>dataset)<\/span> *<\/span> 0.7<\/span>)<\/span>
    \nvalid_data_size =<\/span> int<\/span>(<\/span>len<\/span>(<\/span>dataset)<\/span> *<\/span> 0.15<\/span>)<\/span>
    \ntest_data_size =<\/span> len<\/span>(<\/span>dataset)<\/span> –<\/span> train_data_size –<\/span> valid_data_size<\/p>\n

    train_data,<\/span>valid_data,<\/span>test_data =<\/span> random_split(<\/span>dataset,<\/span>[<\/span>train_data_size,<\/span>valid_data_size,<\/span>test_data_size]<\/span>)<\/span><\/p>\n

    train_set =<\/span> SubSet(<\/span>train_data,<\/span>transform_strengthen)<\/span>
    \nvalid_set =<\/span> SubSet(<\/span>valid_data,<\/span>transform_base)<\/span>
    \ntest_set =<\/span> SubSet(<\/span>test_data,<\/span>transform_base)<\/span><\/p>\n

    data_batch =<\/span> DataLoader(<\/span>dataset=<\/span>train_set,<\/span>batch_size=<\/span>32<\/span>,<\/span>shuffle=<\/span>True<\/span>)<\/span>
    \nvalid_batch =<\/span> DataLoader(<\/span>dataset=<\/span>valid_set,<\/span>batch_size=<\/span>32<\/span>,<\/span>shuffle=<\/span>False<\/span>)<\/span>
    \ntest_batch =<\/span> DataLoader(<\/span>dataset=<\/span>test_set,<\/span>batch_size=<\/span>32<\/span>,<\/span>shuffle=<\/span>False<\/span>)<\/span><\/p>\n

    if<\/span> __name__ ==<\/span> "__main__"<\/span>:<\/span>
    \n print<\/span>(<\/span>f'Train Batch size: <\/span>{<\/span>len<\/span>(<\/span>train_data)<\/span>}<\/span><\/span>'<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>f'Validation Batch size: <\/span>{<\/span>len<\/span>(<\/span>valid_data)<\/span>}<\/span><\/span>'<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>f'Test Batch size: <\/span>{<\/span>len<\/span>(<\/span>test_data)<\/span>}<\/span><\/span>'<\/span><\/span>)<\/span>
    \n for<\/span> index,<\/span>(<\/span>images,<\/span>labels)<\/span> in<\/span> enumerate<\/span>(<\/span>data_batch)<\/span>:<\/span>
    \n if<\/span> index ><\/span> 175<\/span>:<\/span>
    \n print<\/span>(<\/span>f'the <\/span>{<\/span>index}<\/span><\/span>th batch size is <\/span>{<\/span>images.<\/span>shape}<\/span><\/span>'<\/span><\/span>)<\/span><\/p>\n

    Congratulations！！！<\/h3>\n

    \u901a\u8fc7\u7b2c\u4e09\u7ae0\u7684\u5b66\u4e60，\u60a8\u638c\u63e1\u4e86<\/p>\n

      \n
    • \u6a21\u578b\u8bad\u7ec3\u7684\u6570\u636e\u6d41\u7a0b\u662f\u600e\u6837\u7684<\/li>\n
    • \u6211\u4eec\u9700\u8981\u89e3\u51b3\u6570\u636e\u6d41\u7a0b\u4e2d\u9047\u5230\u7684\u54ea\u4e9b\u95ee\u9898<\/li>\n
    • \u5982\u4f55\u89e3\u51b3\u5728\u8fd9\u4e9b\u6570\u636e\u6d41\u7a0b\u4e2d\u9047\u5230\u7684\u95ee\u9898<\/li>\n<\/ul>\n

      \u7b2c\u56db\u7ae0<\/h2>\n

      \u6211\u4eec\u5df2\u7ecf\u80fd\u591f\u6b63\u786e\u7684\u5904\u7406\u56fe\u7247\u6570\u636e，\u63a5\u4e0b\u6765，\u6211\u4eec\u662f\u65f6\u5019\u628a\u5b83\u4ea4\u7ed9\u6a21\u578b，\u8ba9\u8bad\u7ec3\u5668\u4ece\u56fe\u7247\u63d0\u53d6\u7279\u5f81\u5e76\u8bad\u7ec3\u51fa\u53ef\u4ee5\u8bc6\u522b\u56fe\u50cf\u7684\u6a21\u578b\u4e86\u3002\u6211\u4eec\u5e73\u5e38\u662f\u600e\u4e48\u8bc6\u522b\u56fe\u50cf\u5462？\u8fd9\u8ba9\u6211\u60f3\u8d77\u6765\u4e86\u591a\u5e74\u524d\u5728\u8bed\u6587\u8bfe\u672c\u4e0a\u5b66\u7684\u4e00\u4e2a\u6c11\u95f4\u6545\u4e8b\u5e76\u88ab\u300a\u94c1\u9f7f\u94dc\u7259\u7eaa\u6653\u5c9a\u300b\u6f14\u7ece\u7684\u60c5\u666f<\/p>\n

      \u300a\u94c1\u9f7f\u94dc\u7259\u7eaa\u6653\u5c9a\u300b\u4e2d\u7ecf\u5178\u5bf9\u8bdd<\/p>\n

      \u7eaa\u6653\u5c9a\u5f53\u65f6\u4e3a\u4f8d\u90ce\u4e00\u804c\u3002 \u548c\u73c5：\u8fd9\u662f\u4f55\u7269\u554a，\u662f\u72fc\u662f\u72d7（\u4f8d\u90ce\u662f\u72d7）？<\/p>\n

      \u5176\u4e2d\u4e00\u4e2a\u5fa1\u53f2\u9644\u548c：\u4f8d\u90ce，\u662f\u72d7<\/p>\n

      \u7eaa\u6653\u5c9a：\u60a8\u5802\u5802\u4e00\u5927\u5b66\u58eb，\u662f\u72fc\u662f\u72d7\u5206\u4e0d\u6e05，\u6211\u6559\u6559\u60a8，<\/p>\n

      \u201c\u8fd9\u5c3e\u5df4\u4e0b\u6446\u7684\u72fc，\u4e0a\u7ad6\u7684\u662f\u72d7，\u5c1a\u4e66\u662f\u72d7\u201c<\/p>\n

      \u5fa1\u53f2：\u72fc\u5403\u8089，\u72d7\u5403\u7caa，\u5b83\u5403\u8089，\u4f8d\u90ce\u662f\u72d7，\u6beb\u65e0\u7591\u95ee<\/p>\n

      \u7eaa\u6653\u5c9a：\u72d7\u548c\u4eba\u4e0d\u4e00\u6837，\u5b83\u662f\u9047\u8089\u5403\u8089，\u9047\u5c4e\u5403\u5c4e，\u5fa1\u53f2\u5403\u5c4e\u3002<\/p>\n

      \u6587\u7ae0\u6765\u81ea\u300a\u94c1\u9f7f\u94dc\u7259\u7eaa\u6653\u5c9a\u300b\u4e2d\u7ecf\u5178\u5bf9\u8bdd – \u77e5\u4e4e<\/p>\n

      \u8fd9\u662f\u4e00\u4ef6\u5c0f\u8da3\u4e8b，\u4f46\u662f\u6211\u4eec\u80fd\u4ece\u4e2d\u770b\u5230\u5404\u4f4d\u5b98\u5458\u662f\u5982\u4f55\u5206\u8fa8\u72fc\u548c\u72d7\u7684，\u548c\u73c5\u7b49\u5b98\u5458\u901a\u8fc7\u5c3e\u5df4\u8fd9\u4e00\u7279\u5f81\u5224\u65ad，\u7eaa\u6653\u5c9a\u5219\u901a\u8fc7\u751f\u6d3b\u4e60\u6027\u6765\u5224\u65ad，\u5b83\u4eec\u90fd\u662f\u72fc\u548c\u72d7\u8eab\u4e0a\u7684\u4e00\u4e9b\u663e\u8457\u7279\u5f81，\u4eba\u53ef\u4ee5\u901a\u8fc7\u7279\u5f81\u5bf9\u4e00\u5f20\u7167\u7247\u8fdb\u884c\u5206\u7c7b，\u5377\u79ef\u795e\u7ecf\u7f51\u7edc（CNN）\u6b63\u662f\u5728\u8fd9\u4e00\u751f\u7269\u5b66\u4e8b\u5b9e\u4e0b\u542f\u53d1，\u901a\u8fc7\u5377\u79ef\u6838\u7b49\u5de5\u5177\u8fdb\u884c\u7279\u5f81\u626b\u63cf，\u518d\u7ecf\u8fc7\u6fc0\u6d3b\u3001\u6c60\u5316\u7b49\u4e00\u7cfb\u5217\u6b65\u9aa4\u5224\u65ad\u4e00\u5f20\u7167\u7247\u5c5e\u4e8e\u4ec0\u4e48\u7c7b\u522b<\/p>\n

      CNN\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u2014\u2014\u5377\u79ef<\/h3>\n

      \u539f\u7406<\/h4>\n

      \u5728\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u4e2d，\u6211\u4eec\u4e0d\u518d\u805a\u7126\u4e8e\u5206\u6790\u4e00\u4e2a\u4e2a\u50cf\u7d20\u70b9，\u800c\u662f\u5206\u6790\u4e00\u4e2a\u533a\u57df\u662f\u5426\u80fd\u548c\u5377\u79ef\u6838\u7684\u7279\u5f81\u6709\u8f83\u9ad8\u7684\u76f8\u4f3c\u5ea6\u3002\u5c31\u50cf\u6211\u4eec\u5173\u6ce8\u56fe\u7247\u65f6\u4e0d\u4f1a\u5173\u6ce8\u5185\u90e8\u7684\u4e00\u4e2a\u50cf\u7d20\u70b9\u662f\u4ec0\u4e48，\u800c\u662f\u5173\u6ce8\u56fe\u7247\u4e2d\u6709\u4ec0\u4e48（\u7279\u5f81），\u9664\u975e\u60a8\u5728\u5b66\u4e60\u56fe\u7247\u76f8\u5173\u7684\u5de5\u4f5c\u539f\u7406\u6216\u8005\u5bf9\u56fe\u7247\u8fdb\u884c\u50cf\u7d20\u7ea7\u7684\u5904\u7406<\/p>\n

      \u5377\u79ef\u6838\u80fd\u591f\u5e2e\u52a9\u6211\u4eec\u5728\u56fe\u50cf\u4e2d\u626b\u63cf\u7279\u5f81，\u5b83\u53c8\u88ab\u79f0\u4e3a\u6ee4\u6ce2\u5668，\u5c31\u50cf\u6211\u4eec\u4f1a\u7740\u91cd\u5173\u6ce8\u56fe\u50cf\u7684\u67d0\u4e00\u4e2a\u5177\u4f53\u7279\u5f81（\u6211\u4eec\u7684\u5927\u8111\u5e76\u884c\u5904\u7406\u80fd\u529b\u6781\u5f3a，\u80fd\u591f\u77ac\u95f4\u6355\u6349\u5f88\u591a\u7279\u5f81\u5e76\u7efc\u5408\u5206\u6790），\u5377\u79ef\u6838\u5c31\u50cf\u5927\u8111\u4e2d\u7684\u4e00\u4e2a\u5206\u6790\u5668，\u5bf9\u4e8e\u4e00\u5f20\u732b\u7684\u56fe\u7247，\u4ed6\u53ef\u80fd\u4f1a\u6709\u4e00\u4e2a\u4e13\u95e8\u7684\u5377\u79ef\u6838\u5206\u6790\u732b\u7684\u8033\u6735，\u6216\u8005\u5206\u6790\u732b\u7684\u722a\u5b50\u3002\u8fd9\u6837\u7684\u5206\u6790\u5de5\u4f5c\u662f\u7531\u4e00\u4e2a\u7279\u5b9a\u7684\u77e9\u9635\u6765\u5b8c\u6210\u7684 <\/p>\n

      [<\/p>\n

      \u2212<\/p>\n

      1<\/p>\n

      0<\/p>\n

      1<\/p>\n

      \u2212<\/p>\n

      2<\/p>\n

      0<\/p>\n

      2<\/p>\n

      \u2212<\/p>\n

      1<\/p>\n

      0<\/p>\n

      1<\/p>\n

      ]<\/p>\n

      \\\\begin{bmatrix} -1 && 0 && 1\\\\\\\\ -2 && 0 && 2\\\\\\\\ -1 && 0 && 1 \\\\end{bmatrix} <\/p>\n

      <\/span><\/span><\/span><\/p>\n

      <\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2212<\/span>1<\/span><\/span><\/span><\/span>\u2212<\/span>2<\/span><\/span><\/span><\/span>\u2212<\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>0<\/span><\/span><\/span><\/span>0<\/span><\/span><\/span><\/span>0<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>1<\/span><\/span><\/span><\/span>2<\/span><\/span><\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n

      <\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u8fd9\u770b\u8d77\u6765\u53ea\u662f\u4e00\u4e2a\u77e9\u9635，\u5b83\u8be5\u5982\u4f55\u4f5c\u7528\u4e8e\u56fe\u50cf\u5462，\u5b83\u7684\u5de5\u4f5c\u539f\u7406\u662f\u8fd9\u6837\u7684<\/p>\n

    • \u521d\u59cb：\u5c06\u5377\u79ef\u6838\u4f5c\u7528\u4e8e\u56fe\u7247\u7684\u5176\u5b9e\u533a\u57df，\u5de6\u4e0a\u89d2 3 * 3\u7684\u50cf\u7d20\u7f51\u683c\u3010\u5e76\u4e0d\u4e25\u683c\u4e00\u81f4，\u6709\u65f6\u4f1a\u8ba9\u5377\u79ef\u6838\u4e2d\u5fc3\u4f4d\u4e8e\u6700\u5de6\u4e0a\u89d2\u7684\u4e00\u4e2a\u50cf\u7d20，\u5e76\u5c06\u5176\u4ed6\u8d85\u51fa\u56fe\u7247\u533a\u57df\u7684\u50cf\u7d20\u8bbe\u7f6e\u4e3a\u4e00\u4e2a\u56fa\u5b9a\u7684\u503c，\u4e00\u822c\u662f0，\u8fd9\u4f7f\u5f97\u5b83\u53ef\u4ee5\u548c\u5176\u4ed6\u975e\u7279\u6b8a\u60c5\u51b5\u4e00\u6837\u8fdb\u884c\u5377\u79ef\u5de5\u4f5c\u3011（\u8d77\u59cb\u8fb9\u7f18）<\/li>\n
    • \u5bf9\u56fe\u50cf\u533a\u57df\u3010\u622a\u53d6\u548c\u5377\u79ef\u6838\u4e00\u6837\u5927\u5c0f\u7684\u4e00\u4e2a\u77e9\u9635\u3011\u548c\u5377\u79ef\u6838\u70b9\u4e58\u5e76\u5c06\u6240\u6709\u70b9\u4e58\u7ed3\u679c\u76f8\u52a0<\/li>\n
    • \u8ba9\u5377\u79ef\u6838\u5728\u56fe\u50cf\u4e0a\u4ee5\u56fa\u5b9a\u7684\u6b65\u957f\u5411\u53f3\u6ed1\u52a8\u7ee7\u7eed\u8fdb\u884c\u7b2c\u4e8c\u6b65，\u76f4\u5230\u8fbe\u5230\u56fe\u7247\u53f3\u4fa7\u8fb9\u7f18\u540e，\u8ba9\u56fe\u7247\u56de\u5230\u8be5\u64cd\u4f5c\u7684\u8d77\u59cb\u8fb9\u7f18，\u4ee5\u56fa\u5b9a\u6b65\u957f\u4e0b\u6ed1，\u66f4\u65b0\u8d77\u59cb\u8fb9\u7f18\u4e3a\u5f53\u524d\u56fe\u50cf\u533a\u57df\u4e0e\u5377\u79ef\u6838\u4f5c\u7528\u533a\u57df，\u7ee7\u7eed\u8fdb\u884c\u7b2c\u4e8c\u6b65<\/li>\n
    • \u5f53\u56fe\u50cf\u7684\u4e0b\u4e00\u6b21\u56fa\u5b9a\u6b65\u957f\u7684\u5377\u79ef\u6838\u4e2d\u5fc3\u7684\u884c\u7d22\u5f15\u8d85\u8fc7\u56fe\u50cf\u5e95\u90e8\u7684\u884c\u7d22\u5f15\u65f6，\u7ed3\u675f\u5377\u79ef\u64cd\u4f5c \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \u8fd9\u770b\u8d77\u6765\u50cf\u662f\u77e9\u9635\u7248\u7684\u6ed1\u52a8\u7a97\u53e3，\u4e0d\u8fc7\u6211\u4eec\u4e3a\u4ec0\u4e48\u8981\u8fd9\u4e48\u505a，\u6216\u8bb8\u6211\u4eec\u53ef\u4ee5\u6a21\u62dfCNN\u7684\u5de5\u4f5c\u8fc7\u7a0b，\u770b\u770b\u7a76\u7adf\u53d1\u751f\u4e86\u4ec0\u4e48<\/li>\n

      \u6211\u4eec\u5c1d\u8bd5\u5c06\u4e00\u4e2a\u5377\u79ef\u6838\u5e94\u7528\u5728\u4e0d\u540c\u7684\u56fe\u50cf\u622a\u53d6\u77e9\u9635\u5462，\u89c2\u5bdf\u4e0d\u540c\u7684\u56fe\u50cf\u7684\u8ba1\u7b97\u7ed3\u679c\u4f1a\u6709\u4ec0\u4e48\u4e0d\u540c $$ \\\\begin{bmatrix} 1 && 1 && 1\\\\ 1 && 1 && 1\\\\ 1 && 1 && 1 \\\\end{bmatrix} \\\\cdot \\\\begin{bmatrix} -1 && 0 && 1\\\\ -2 && 0 && 2\\\\ -1 && 0 && 1 \\\\end{bmatrix}<\/h2>\n

      \\\\begin{bmatrix} -1 && 0 && 1\\\\ -2 && 0 && 2\\\\ -1 && 0 && 1 \\\\end{bmatrix} ->0\\\\ \\\\ \\\\begin{bmatrix} -1 && 0 && 1\\\\ -2 && 0 && 2\\\\ -1 && 0 && 1 \\\\end{bmatrix} \\\\cdot \\\\begin{bmatrix} -1 && 0 && 1\\\\ -2 && 0 && 2\\\\ -1 && 0 && 1 \\\\end{bmatrix}<\/h2>\n

      \\\\begin{bmatrix} 1 && 0 && 1\\\\ 4 && 0 && 4\\\\ 1 && 0 && 1 \\\\end{bmatrix} ->12 \\\\ \\\\ \\\\begin{bmatrix} -1 && 0 && 1\\\\ -2 && 0 && 2\\\\ -1 && 0 && 1 \\\\end{bmatrix} \\\\cdot \\\\begin{bmatrix} 1 && 0 && -1\\\\ 2 && 0 && -2\\\\ 1 && 0 && -1 \\\\end{bmatrix}<\/h2>\n

      \\\\begin{bmatrix} -1 && 0 && -1\\\\ -4 && 0 && -4\\\\ -1 && 0 && -1 \\\\end{bmatrix} ->-12 $$<\/p>\n

      \u5982\u679c\u6211\u4eec\u901a\u8fc7\u6570\u503c\u6bd4\u8f83\u77e9\u9635\u70b9\u4e58\u540e\u76f8\u52a0\u7684\u7ed3\u679c，\u6211\u4eec\u53ef\u4ee5\u770b\u5230\u5e38\u6570\u77e9\u9635\u7684\u8ba1\u7b97\u7ed3\u679c\u4e3a0，\u800c\u4e0e\u5377\u79ef\u6838\u76f8\u7b49\u7684\u77e9\u9635，\u4ed6\u4eec\u7684\u8ba1\u7b97\u7ed3\u679c\u6570\u503c\u8f83\u9ad8，\u800c\u4e0e\u5377\u79ef\u6838\u5728\u6c34\u5e73\u7ffb\u8f6c\u540e\u7684\u77e9\u9635，\u4e0e\u5b83\u8fdb\u884c\u8ba1\u7b97\u540e\u7684\u7ed3\u679c\u8f83\u4f4e，\u5982\u679c\u6211\u4eec\u6fc0\u6d3b\u4e0e\u5377\u79ef\u6838\u8fd0\u7b97\u6570\u503c\u8f83\u9ad8\u7684\u77e9\u9635\u3010\u5728\u8fd9\u91cc\u8868\u73b0\u4e3a\u4e0e\u5377\u79ef\u6838\u76f8\u7b49\u3011，\u90a3\u4e48\u5f53\u5377\u79ef\u6838\u5728\u539f\u59cb\u56fe\u50cf\u4e0a\u6ed1\u52a8\u65f6，\u5b83\u5bf9\u4e8e\u5377\u79ef\u6838\u76f8\u7b49\u7684\u54cd\u5e94\u4f1a\u66f4\u6fc0\u70c8，\u4ece\u800c\u4f7f\u6a21\u578b\u627e\u5230\u6b63\u786e\u7684\u7279\u5f81（\u4e0e\u5377\u79ef\u6838\u9ad8\u5ea6\u76f8\u4f3c），\u4e8b\u5b9e\u4e0a，\u8fd9\u79cd\u8868\u793a\u5e76\u4e0d\u5b8c\u5168\u5728\u4f53\u73b0\u5728\u4e0e\u5377\u79ef\u6838\u6570\u503c\u5168\u7b49\u7684\u56fe\u50cf\u622a\u53d6，\u6211\u4eec\u53ef\u4ee5\u8ba9\u4e2d\u95f4\u7684\u77e9\u9635\u518d\u901a\u7528\u4e00\u4e9b $$ \\\\begin{bmatrix} -ax && 0 && dx\\\\ -by && 0 && ey\\\\ -cz && 0 && fz \\\\end{bmatrix} \\\\cdot \\\\begin{bmatrix} -x && 0 && x\\\\ -y && 0 && y\\\\ -z && 0 && z \\\\end{bmatrix}<\/h2>\n

      \\\\begin{bmatrix} ax^2 && 0 && dx^2\\\\ by^2 && 0 && ey^2\\\\ cz^2 && 0 && fz^2 \\\\end{bmatrix} ->(a+d)x2;+;(b+e)y<\/sup>2;+;(c+f)z^2\\\\ \\\\ a\u3001b\u3001c\u3001d\u3001e\u3001f;\u5747\u4e3a\u6b63\u6570 KaTeX parse error: Expected 'EOF', got '#' at position 85: \u2026\u7b2c\u4e00\u7ae0\u2014\u2014\u4f18\u5316\u5355\u795e\u7ecf\u5143\u6a21\u578b](#\u0332\u4f18\u5316\u5355\u795e\u7ecf\u5143\u6a21\u578b))\u6fc0\u6d3b <\/span> \\\\begin{bmatrix} ax && 0 && -dx\\\\ by && 0 && -ey\\\\ cz && 0 && -fz \\\\end{bmatrix} \\\\cdot \\\\begin{bmatrix} -x && 0 && x\\\\ -y && 0 && y\\\\ -z && 0 && z \\\\end{bmatrix}<\/h2>\n

      \\\\begin{bmatrix} -ax^2 && 0 && -dx^2\\\\ -by^2 && 0 && -ey^2\\\\ -cz^2 && 0 && -fz^2 \\\\end{bmatrix} ->-(a+d)x2;-;(b+e)y<\/sup>2;-;(c+f)z^2\\\\\\\\ a\u3001b\u3001c\u3001d\u3001e\u3001f;\u5747\u4e3a\u6b63\u6570 $$ \u76f8\u53cd，\u5f53\u7cfb\u7edf\u4e3a\u6b63\u6570\u3001\u5373\u56fe\u50cf\u622a\u53d6\u65b9\u5411\u4e0e\u5377\u79ef\u6838\u65b9\u5411\u5b8c\u5168\u76f8\u53cd\u65f6，\u6211\u4eec\u80fd\u5f97\u5230\u6700\u5c0f\u7684\u503c，\u800c\u4e14\u5f80\u53cd\u65b9\u5411\u504f\u79bb\u7684\u8d8a\u591a，\u503c\u8d8a\u5c0f，\u5b83\u4f1a\u88abRelu\u6fc0\u6d3b\u51fd\u6570\u8bbe\u7f6e\u4e3a0\u3002<\/p>\n

      \u5982\u679c\u5377\u79ef\u6838\u4e2d\u6570\u503c\u7684\u6b63\u8d1f\u4ee3\u8868\u7279\u5f81\u4e00\u81f4\u6216\u8005\u76f8\u53cd\u8fd9\u79cd\u65b9\u5411，\u90a3\u4e48\u6570\u503c\u5c31\u7c7b\u4f3c\u6743\u91cd，\u5982\u679c\u5728\u6743\u91cd\u8f83\u5927\u7684\u5730\u65b9，\u56fe\u7247\u622a\u53d6\u7684\u67d0\u4e00\u50cf\u7d20\u72af\u4e86\u65b9\u5411\u7684\u9519\u8bef，\u90a3\u4e48\u8fd9\u5bf9\u6574\u4f53\u7ed3\u679c\u7684\u60e9\u7f5a\u7a0b\u5ea6\u4f1a\u8f83\u5927（\u66f4\u52a0\u504f\u79bb\u7279\u5f81\u76f8\u4f3c）\u5982\u679c\u50cf\u7d20\u5728\u6743\u91cd\u8f83\u5927\u7684\u5730\u65b9\u65b9\u5411\u5bf9\u4e86，\u90a3\u4e48\u8fd9\u5bf9\u6574\u4f53\u7ed3\u679c\u7684\u5956\u52b1\u529b\u5ea6\u4e5f\u4f1a\u8d8a\u5927 $$ \\\\begin{bmatrix} -1 && 0 && 1\\\\ 2 && 0 && -2\\\\ -1 && 0 && 1 \\\\end{bmatrix} \\\\cdot \\\\begin{bmatrix} -1 && 0 && 1\\\\ -2 && 0 && 2\\\\ -1 && 0 && 1 \\\\end{bmatrix}<\/h2>\n

      \\\\begin{bmatrix} 1 && 0 && 1\\\\ -4 && 0 && -4\\\\ 1 && 0 && 1 \\\\end{bmatrix} ->-4 <\/p>\n

      \u5bf9\u4e8e\u8fd9\u4e2a\u4f8b\u5b50，\u56fe\u50cf\u622a\u53d6\u5728\u77e9\u9635\u70b9\u4e58\u7684\u8bb8\u591a\u65b9\u5411\u4e0e\u5377\u79ef\u6838\u4fdd\u6301\u4e00\u81f4，\u4f46\u662f\u5728\u6743\u91cd\u6700\u5927\u7684\u4e24\u4e2a\u65b9\u5411\u76f8\u53cd，\u8fd9\u5bfc\u81f4\u539f\u672c\u8f83\u5927\u7684\u6570\u503c\u2014\u2014<\/p>\n

      12<\/p>\n

      \u53d8\u6210\u4e86\u8d1f\u6570<\/p>\n

      \u5bf9\u4e8e\u8fd9\u4e2a\u4f8b\u5b50，\u56fe\u50cf\u622a\u53d6\u5728\u77e9\u9635\u70b9\u4e58\u7684\u8bb8\u591a\u65b9\u5411\u4e0e\u5377\u79ef\u6838\u4fdd\u6301\u4e00\u81f4，\u4f46\u662f\u5728\u6743\u91cd\u6700\u5927\u7684\u4e24\u4e2a\u65b9\u5411\u76f8\u53cd，\u8fd9\u5bfc\u81f4\u539f\u672c\u8f83\u5927\u7684\u6570\u503c\u2014\u201412\u53d8\u6210\u4e86\u8d1f\u6570 <\/p>\n

      <\/span><\/span>\u5bf9\u4e8e\u8fd9\u4e2a\u4f8b\u5b50，\u56fe\u50cf\u622a\u53d6\u5728\u77e9\u9635\u70b9\u4e58\u7684\u8bb8\u591a\u65b9\u5411\u4e0e\u5377\u79ef\u6838\u4fdd\u6301\u4e00\u81f4，\u4f46\u662f\u5728\u6743\u91cd\u6700\u5927\u7684\u4e24\u4e2a\u65b9\u5411\u76f8\u53cd，\u8fd9\u5bfc\u81f4\u539f\u672c\u8f83\u5927\u7684\u6570\u503c<\/span>\u2014\u201412<\/span>\u53d8\u6210\u4e86\u8d1f\u6570<\/span><\/span><\/span><\/span><\/span><\/span> \\\\begin{bmatrix} 1 && 0 && -1\\\\ -2 && 0 && 2\\\\ 1 && 0 && -1 \\\\end{bmatrix} \\\\cdot \\\\begin{bmatrix} -1 && 0 && 1\\\\ -2 && 0 && 2\\\\ -1 && 0 && 1 \\\\end{bmatrix}<\/h2>\n

      \\\\begin{bmatrix} -1 && 0 && -1\\\\ 4 && 0 && 4\\\\ -1 && 0 && -1 \\\\end{bmatrix} ->4 $$ \u5bf9\u4e8e\u8fd9\u4e2a\u4f8b\u5b50，\u56fe\u50cf\u622a\u53d6\u5728\u77e9\u9635\u70b9\u4e58\u7684\u8bb8\u591a\u65b9\u5411\u4e0e\u5377\u79ef\u6838\u76f8\u53cd，\u4f46\u662f\u5728\u6743\u91cd\u6700\u5927\u7684\u4e24\u4e2a\u65b9\u5411\u4e00\u81f4，\u8fd9\u5bfc\u81f4\u539f\u672c\u8f83\u5c0f\u7684\u6570\u503c\u53d8\u5f97\u8f83\u5927\u2014\u2014-12\u53d8\u6210\u4e86\u6b63\u6570<\/p>\n

      tips:\u5728\u65e5\u5e38CNN\u8bad\u7ec3\u4e2d，\u6211\u4eec\u4e00\u822c\u91c7\u7528\u5c0f\u5377\u79ef\u6838(3*3 | 5*5)，\u56e0\u4e3a\u5728\u8bb8\u591a\u56fe\u7247\u4e2d，\u76f8\u90bb\u50cf\u7d20\u4e4b\u95f4\u7684\u5dee\u8ddd\u4e0d\u4f1a\u7279\u522b\u5927，\u4ee5\u6b64\u4fdd\u6301\u8f83\u4e3a\u7a33\u5b9a\u7684\u56fe\u50cf\u622a\u53d6\u4e0e\u5377\u79ef\u6838\u76f8\u4e58，\u907f\u514d\u7279\u522b\u5927\u7684\u6743\u91cd\u7684\u5e72\u6270<\/p>\n

      pytorch\u5bf9\u5e94<\/h4>\n

      \u60a8\u4e0d\u7528\u4eb2\u81ea\u4e3a\u6bcf\u4e00\u4e2a\u7279\u5f81\u6d89\u53ca\u5377\u79ef\u6838，pytorch\u5728\u6a21\u578b\u8bad\u7ec3\u7684\u8fc7\u7a0b\u4e2d\u4f1a\u901a\u8fc7\u4e00\u7cfb\u5217\u8c03\u6574\u2014\u2014\u77eb\u6b63（\u901a\u8fc7\u6539\u53d8\u53c2\u6570\u548c\u5e94\u7528\u635f\u5931\u51fd\u6570\u7b49），\u8ba9\u6211\u4eec\u770b\u770b\u5982\u4f55\u5728pytorch\u4e2d\u505a\u5230\u8fd9\u4e00\u70b9<\/p>\n

      from<\/span> torch import<\/span> nn
      \nconv_layer =<\/span> nn.<\/span>Conv2d(<\/span>
      \n in_channels=<\/span>3<\/span>,<\/span>
      \n out_channels=<\/span>16<\/span>,<\/span>
      \n kernel_size=<\/span>3<\/span>,<\/span>
      \n stride=<\/span>1<\/span>,<\/span>
      \n padding=<\/span>1<\/span>,<\/span>
      \n)<\/span><\/p>\n

      torch\u6a21\u5757\u5f53\u4e2d\u7684nn\u6a21\u5757\u7684Conv2d\u65b9\u6cd5\u53ef\u4ee5\u5e2e\u52a9\u6211\u4eec\u6784\u5efa\u5377\u79ef\u5c42，\u5176\u4e2d\u7684\u53c2\u6570<\/p>\n

      in_channels：\u8fd9\u4ee3\u8868\u8f93\u5165\u56fe\u7247\u7684\u901a\u9053\u6570，\u5bf9\u4e8e\u6211\u4eecRGB\u56fe\u50cf\u6765\u8bf4，\u6b63\u597d\u662fR\u3001G\u3001B\u4e09\u4e2a\u901a\u9053<\/p>\n

      out_channels：\u8fd9\u4ee3\u8868\u8f93\u51fa\u56fe\u7247\u7684\u901a\u9053\u6570，\u6bcf\u4e00\u4e2a\u901a\u9053\u5b58\u50a8\u4e00\u4e2a\u5377\u79ef\u6838\u5377\u79ef\u540e\u7684\u7ed3\u679c，\u603b\u5171\u4f1a\u6709out_channels\u4e2a\u903b\u8f91\u5377\u79ef\u6838（\u3010in_channels * \u5355\u901a\u9053\u5377\u79ef\u6838\u3011\u4f5c\u4e3a\u4e00\u4e2a\u6574\u4f53\u4f5c\u7528\u5728\u4e00\u5f20\u56fe\u7247\u4e0a）<\/p>\n

      kernel_size：\u5377\u79ef\u6838\u5927\u5c0f，\u662fkerner_size * kernersize\u5927\u5c0f\u7684\u5355\u901a\u9053\u5377\u79ef\u6838<\/p>\n

      stride：\u6b65\u957f，\u5377\u79ef\u6838\u6bcf\u6b21\u5728\u56fe\u50cf\u4e0a\u5411\u53f3\u548c\u5411\u4e0b\u6ed1\u52a8\u7684\u8ddd\u79bb，\u5bf9\u4e8e\u6b65\u957f\u6765\u8bf4，\u5982\u679c\u60a8\u8bbe\u7f6e\u7684\u8f83\u5927，\u5219\u5bf9\u5e94\u7684\u5377\u79ef\u5b8c\u6210\u901f\u5ea6\u4e5f\u4f1a\u5feb，\u4f46\u662f\u8fd9\u4f1a\u8ba9\u6a21\u578b\u9519\u5931\u4e00\u4e9b\u50cf\u7d20\u7684\u5377\u79ef\u8ba1\u7b97，\u5bfc\u81f4\u9519\u8fc7\u4e00\u4e9b\u7ec6\u8282\u7279\u5f81<\/p>\n

      padding：\u5185\u8fb9\u8ddd，\u5f53\u60a8\u60f3\u8981\u8ba9\u5377\u79ef\u6838\u4e2d\u5fc3\u4f4d\u4e8e\u56fe\u50cf\u89d2\u843d\u7684\u65f6\u5019，\u60a8\u9700\u8981\u8bbe\u7f6e，\u4ee5\u907f\u514d\u77e9\u9635\u884c\u7d22\u5f15 + \u77e9\u9635\u5217\u7d22\u5f15 < \u77e9\u9635\u4e2d\u5fc3\u7684\u884c\u7d22\u5f15 + \u77e9\u9635\u4e2d\u5fc3\u5217\u7d22\u5f15\u65f6\u65e0\u6cd5\u627e\u5230\u6709\u6548\u7684\u50cf\u7d20\u503c\u76f8\u4e58，\u503c\u4e00\u822c\u8868\u793a\u9700\u8981\u591a\u5c11\u5c42padding，\u4e00\u822c\u662f(kernel_size – 1) \/ 2<\/p>\n

      CNN\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u2014\u2014\u6c60\u5316<\/h3>\n

      \u63a5\u4e0b\u6765\u8ba9\u6211\u4eec\u653e\u677e\u5fc3\u60c5，\u6765\u4e00\u573a\u5377\u79ef\u4e4b\u95f4\u7684\u5927\u6bd4\u62fc，\u80dc\u8005\u6765\u81ea\u6c60\u5316\u6bd4\u62fc\u7684\u4f7c\u4f7c\u8005，\u4ed6\u4eec\u5c06\u8fdb\u5165\u4e0b\u4e00\u8f6e\u7684\u5377\u79ef<\/p>\n

      \u54c8\u54c8，\u8fd9\u542c\u8d77\u6765\u4ee4\u4eba\u5fc3\u6f6e\u6f8e\u6e43，\u5b9e\u9645\u4e0a，\u8fd9\u4fbf\u662f\u6c60\u5316\u6240\u505a\u7684\u4e8b\u60c5，\u4ed6\u50cf\u662f\u4e00\u4e2a\u56f4\u680f，\u5c06\u5377\u79ef\u540e\u7684\u56fe\u50cf\u56f4\u57ce\u4e00\u4e2a\u4e2a\u51b3\u6597\u573a，\u7b26\u5408\u5224\u8d62\u89c4\u5219\u7684\u9009\u624b，\u8fdb\u5165\u4e0b\u4e00\u6b21\u6bd4\u8d5b，\u8fd9\u662f\u6765\u81ea\u4e8e\u8bad\u7ec3\u6a21\u578b\u7684\u4f18\u4e2d\u9009\u4f18，\u60a8\u4f5c\u4e3a\u88c1\u5224，\u53ef\u4ee5\u5236\u5b9a\u5224\u8d62\u89c4\u5219，\u5e38\u89c1\u7684\u89c4\u5219\u662f\u4ece\u91cc\u9762\u9009\u503c\u6700\u5927\u7684\u90a3\u4e00\u4f4d\u50cf\u7d20\u9009\u624b，\u6700\u4ec1\u6148\u7684\u89c4\u5219，\u5f53\u7136\u662f\u5168\u90e8\u63a5\u6536\u5566，\u8fd9\u4e2a\u65f6\u5019，\u51b3\u6597\u5708（\u6216\u8005\u662f\u56e2\u961f）\u4e2d\u7684\u6bcf\u4e2a\u50cf\u7d20\u90fd\u8d21\u732e\u81ea\u5df1\u7684\u4e00\u4efd\u529b\u91cf，\u53d6\u5e73\u5747\u503c\u4e0a\u4ea4\u7ed9\u8bad\u7ec3\u6a21\u578b\u3002\u6a21\u578b\u671f\u671b\u80fd\u591f\u9009\u51fa\u548c\u7279\u5f81\u6700\u76f8\u8fd1\u7684，\u6bd4\u8d5b\u662f\u6b8b\u9177\u7684，\u6700\u5e38\u89c1\u7684\u6a21\u5f0f\u5c31\u662f\u201c\u51b3\u6597\u573a\u201d\u6a21\u5f0f，\u56f4\u680f\u53d6\u6700\u5927\u503c，\u81f3\u4e8e\u56f4\u680f\u7684\u5927\u5c0f\u5c31\u662fkernelsize * kernelsize\u7684\u6c60\u5316\u6838\u5927\u5c0f\u4e86 \"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

      \u6c60\u5316\u4f1a\u51cf\u5c0f\u56fe\u50cf\u9762\u79ef，\u8ba9\u56fe\u50cf\u91cd\u70b9\u5173\u6ce8\u7279\u5f81\u76f8\u4f3c\u7684\u90e8\u5206，\u4ece\u800c\u8fbe\u5230\u51cf\u5c11\u566a\u58f0\u4ee5\u53ca\u63d0\u5347\u6a21\u578b\u8bad\u7ec3\u901f\u5ea6\u7684\u6548\u679c<\/p>\n

      CNN\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u2014\u2014\u5b8c\u6574\u67b6\u6784\u5b9e\u73b0<\/h3>\n

      class<\/span> SimpleCNN<\/span>(<\/span>nn.<\/span>Module)<\/span>:<\/span>
      \n def<\/span> __init__<\/span>(<\/span>self)<\/span>:<\/span>
      \n super<\/span>(<\/span>)<\/span>.<\/span>__init__(<\/span>)<\/span>
      \n self.<\/span>conv1 =<\/span> nn.<\/span>Conv2d(<\/span>3<\/span>,<\/span> 32<\/span>,<\/span> kernel_size=<\/span>3<\/span>,<\/span>padding=<\/span>1<\/span>)<\/span>
      \n self.<\/span>relu1 =<\/span> nn.<\/span>ReLU(<\/span>)<\/span>
      \n self.<\/span>pool1 =<\/span> nn.<\/span>MaxPool2d(<\/span>kernel_size=<\/span>2<\/span>,<\/span>stride=<\/span>2<\/span>)<\/span><\/p>\n

      self.<\/span>conv2 =<\/span> nn.<\/span>Conv2d(<\/span>32<\/span>,<\/span> 64<\/span>,<\/span> kernel_size=<\/span>3<\/span>,<\/span> padding=<\/span>1<\/span>)<\/span>
      \n self.<\/span>relu2 =<\/span> nn.<\/span>ReLU(<\/span>)<\/span>
      \n self.<\/span>pool2 =<\/span> nn.<\/span>MaxPool2d(<\/span>kernel_size=<\/span>2<\/span>,<\/span>stride=<\/span>2<\/span>)<\/span><\/p>\n

      self.<\/span>conv3 =<\/span> nn.<\/span>Conv2d(<\/span>64<\/span>,<\/span> 128<\/span>,<\/span> kernel_size=<\/span>3<\/span>,<\/span> padding=<\/span>1<\/span>)<\/span>
      \n self.<\/span>relu3 =<\/span> nn.<\/span>ReLU(<\/span>)<\/span>
      \n self.<\/span>pool3 =<\/span> nn.<\/span>MaxPool2d(<\/span>kernel_size=<\/span>2<\/span>,<\/span>stride=<\/span>2<\/span>)<\/span><\/p>\n

      self.<\/span>flatten =<\/span> nn.<\/span>Flatten(<\/span>)<\/span><\/p>\n

      self.<\/span>fc1 =<\/span> nn.<\/span>Linear(<\/span>128<\/span> *<\/span> 28<\/span> *<\/span> 28<\/span>,<\/span> 512<\/span>)<\/span>
      \n self.<\/span>relu4 =<\/span> nn.<\/span>ReLU(<\/span>)<\/span>
      \n self.<\/span>dropout =<\/span> nn.<\/span>Dropout(<\/span>0.5<\/span>)<\/span>
      \n self.<\/span>fc2 =<\/span> nn.<\/span>Linear(<\/span>512<\/span>,<\/span> 102<\/span>)<\/span><\/p>\n

      \u6211\u4eec\u5b9a\u4e49\u4e00\u4e2a\u7c7b，\u8ba9\u5b83\u7ee7\u627f\u81eann,Module，\u8fd9\u610f\u5473\u7740\u5728\u6b63\u5f0f\u8bad\u7ec3\u7684\u65f6\u5019，\u5b83\u4f1a\u88ab\u5b9e\u4f8b\u5316\u4e3a\u4e00\u4e2a\u6a21\u578b\u5bf9\u8c61，\u5173\u4e8e\u5b8c\u6574\u7684\u6a21\u578b\u6784\u5efa，\u60a8\u53ef\u4ee5\u67e5\u770bML&\u6a21\u578b（\u6784\u5efa\u6a21\u578b\u3001\u8bad\u7ec3）,\u6a21\u578b\u6784\u5efa\u7684\u6b65\u9aa4\u662f\u76f8\u540c\u7684，\u90fd\u662f\u5148\u901a\u8fc7__init__\u6784\u9020\u51fd\u6570\u5b9a\u4e49\u6a21\u578b\u5c42\u4ee5\u53ca\u6bcf\u4e00\u5c42\u6240\u8981\u505a\u7684\u5de5\u4f5c，\u7136\u540e\u5728forward\u51fd\u6570\u4e2d\u5e94\u7528\u6a21\u578b\u5c42\u8fdb\u884c\u8bad\u7ec3,\u5173\u4e8enn.MaxPool2d\u8fd9\u4e2a\u51fd\u6570，\u5b83\u662f\u7528\u6765\u6784\u5efa\u6c60\u5316\u5c42\u7684，\u53c2\u6570kernel_size\u4ee3\u8868\u6c60\u5316\u6838\u7684\u5927\u5c0f，stride\u4ee3\u8868\u6b65\u957f，\u4ed6\u4e00\u822c\u548ckernel_size\u4fdd\u6301\u4e00\u81f4，\u5b83\u7684\u4f5c\u7528\u548c\u5377\u79ef\u7684\u6309\u6b65\u957f\u6ed1\u52a8\u662f\u4e00\u6837\u7684，\u5f53\u6b65\u957f\u7b49\u4e8e\u6838\u7684\u5c3a\u5bf8\u65f6，\u521a\u597d\u5f80\u65b9\u5411\u6ed1\u52a8\u4e00\u4e2a\u6838\u7684\u5927\u5c0f（\u6ed1\u8fc7\u81ea\u8eab），\u4ece\u800c\u907f\u514d\u91cd\u53e0\u3010\u91cd\u53e0\u53ef\u80fd\u5bfc\u81f4\u4e00\u4e2a\u7279\u5f81\u503c\u88ab\u591a\u6b21\u5305\u542b\u3011，\u6700\u540e\u7684\u6b65\u9aa4\u5c31\u662f\u5c06\u6211\u4eec\u7684\u591a\u901a\u9053\u56fe\u50cf\u5e94\u7528\u4e8e\u5168\u8fde\u63a5\u5c42，\u5168\u8fde\u63a5\u5c42\u63a5\u6536\u56fe\u50cf，\u5e76\u6784\u5efa\u591a\u9879\u5f0f\u6765\u62df\u5408\u6a21\u578b\u7279\u5f81<\/p>\n

      \u5728\u6574\u4e2a\u6a21\u578b\u6784\u5efa\u7684\u8fc7\u7a0b\u4e2d，\u6211\u4eec\u9700\u8981\u989d\u5916\u5173\u6ce8\u4e00\u4e9b\u53c2\u6570\u8bbe\u7f6e<\/p>\n

    • \n

      conv1\u7684in_channel\u7b49\u4e8e\u539f\u59cb\u56fe\u50cf\u7684channel，\u800c\u6bcf\u4e00\u6b21\u7684out_channel\u4f5c\u4e3a\u4e0b\u4e00\u6b21\u5377\u79ef\u7684in_channel，\u6211\u4eec\u5728forward\u51fd\u6570\u4e2d\u4e00\u822c\u6309\u7167\u987a\u5e8f\u6784\u5efa，\u7b2c\u4e00\u5c42\u7684\u5377\u79ef\u540e\u7684\u7ed3\u679c\u7279\u5f81\u56fe\u5728\u6c60\u5316\u540e\u4f1a\u8fdb\u5165\u4e0b\u4e00\u8f6e\u5377\u79ef，\u786e\u4fdd\u4e0a\u4e00\u8f6e\u7684\u5377\u79ef\u8f93\u51fa\u901a\u9053\u6570\u4e0e\u5f53\u524d\u8f6e\u7684\u5377\u79ef\u8f93\u5165\u901a\u9053\u6570\u4fdd\u6301\u4e00\u81f4\u3010\u6c60\u5316\u5e76\u4e0d\u6539\u53d8\u901a\u9053\u5927\u5c0f\u3011<\/p>\n<\/li>\n

    • \n

      \u5168\u8fde\u63a5\u5c42\u7684\u7b2c\u4e00\u5c42\u8f93\u5165\u8282\u70b9\u6570\u5e94\u8be5\u7b49\u4e8e\u5f53\u524d\u901a\u9053\u6570 * \u56fe\u5f62\u5c3a\u5bf8（height * width），\u6bcf\u7ecf\u8fc7\u4e00\u5c42\u6c60\u5316，\u56fe\u50cf\u5c3a\u5bf8\u90fd\u4f1a\u53d8\u5316\u4e3a<\/p>\n

      \u56fe\u50cf\u5c3a\u5bf8
      \n \u2005\u200a<\/p>\n

      =
      \n \u2005\u200a<\/p>\n

      h<\/p>\n

      e<\/p>\n

      i<\/p>\n

      g<\/p>\n

      h<\/p>\n

      t<\/p>\n

      :<\/p>\n

      o<\/p>\n

      r<\/p>\n

      i<\/p>\n

      g<\/p>\n

      i<\/p>\n

      n<\/p>\n

      H<\/p>\n

      e<\/p>\n

      i<\/p>\n

      g<\/p>\n

      h<\/p>\n

      t<\/p>\n

      p<\/p>\n

      o<\/p>\n

      o<\/p>\n

      l<\/p>\n

      k<\/p>\n

      e<\/p>\n

      r<\/p>\n

      n<\/p>\n

      e<\/p>\n

      l<\/p>\n

      \u2005\u200a<\/p>\n

      \u2217
      \n \u2005\u200a<\/p>\n

      w<\/p>\n

      i<\/p>\n

      d<\/p>\n

      t<\/p>\n

      h<\/p>\n

      :<\/p>\n

      o<\/p>\n

      r<\/p>\n

      i<\/p>\n

      g<\/p>\n

      i<\/p>\n

      n<\/p>\n

      W<\/p>\n

      i<\/p>\n

      d<\/p>\n

      t<\/p>\n

      h<\/p>\n

      p<\/p>\n

      o<\/p>\n

      o<\/p>\n

      l<\/p>\n

      k<\/p>\n

      e<\/p>\n

      r<\/p>\n

      n<\/p>\n

      e<\/p>\n

      l<\/p>\n

      {\u56fe\u50cf\u5c3a\u5bf8\\\\;=\\\\;height:\\\\dfrac{originHeight}{poolkernel}\\\\;*\\\\;width:\\\\dfrac{originWidth}{poolkernel}}<\/p>\n

      <\/span><\/span>\u56fe\u50cf\u5c3a\u5bf8<\/span><\/span><\/span>=<\/span><\/span><\/span>h<\/span>e<\/span>i<\/span>g<\/span>h<\/span>t<\/span><\/span>:<\/span><\/span><\/span><\/span>p<\/span>oo<\/span>l<\/span>k<\/span>er<\/span>n<\/span>e<\/span>l<\/span><\/span><\/span><\/span><\/span><\/span><\/span>or<\/span>i<\/span>g<\/span>in<\/span>H<\/span>e<\/span>i<\/span>g<\/span>h<\/span>t<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span>w<\/span>i<\/span>d<\/span>t<\/span>h<\/span><\/span>:<\/span><\/span><\/span><\/span>p<\/span>oo<\/span>l<\/span>k<\/span>er<\/span>n<\/span>e<\/span>l<\/span><\/span><\/span><\/span><\/span><\/span><\/span>or<\/span>i<\/span>g<\/span>inW<\/span>i<\/span>d<\/span>t<\/span>h<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<\/li>\n

    • \n

      \u6700\u540e\u4e00\u5c42\u8fde\u63a5\u5c42\u4e3a\u8f93\u51fa\u5c42，\u8f93\u51fa\u8282\u70b9\u6570\u5e94\u8be5\u7b49\u4e8e\u5206\u7c7b\u6570\u91cf\u3010\u5728\u5206\u7c7b\u4efb\u52a1\u4e2d\u3011<\/p>\n<\/li>\n

      tips：\u6fc0\u6d3b\u51fd\u6570\u662f\u4e3a\u4e86\u62df\u5408\u975e\u7ebf\u6027\u7684\u591a\u79cd\u60c5\u51b5\u800c\u505a\u7684\u7b56\u7565，\u5f53\u8f93\u51fa\u5c42\u8f93\u51fa\u8282\u70b9\u540e，\u6211\u4eec\u5e76\u4e0d\u9700\u8981\u518d\u8bbe\u7f6e\u6fc0\u6d3b\u51fd\u6570，\u6b64\u65f6\u51fd\u6570\u4e0d\u518d\u62df\u5408，\u800c\u662f\u4f5c\u4e3a\u4e00\u6b21\u8bad\u7ec3\u7ed3\u679c\u8f93\u51fa<\/p>\n

      tips：\u60a8\u9700\u8981\u5c06\u56fe\u50cf\u5c55\u5e73\u4e3a\u4e00\u7ef4\u6570\u636e，\u8fd9\u662f\u5168\u8fde\u63a5\u8282\u70b9\u53ef\u4ee5\u63a5\u53d7\u7684\u8f93\u5165\u901a\u9053shape，\u5c55\u5e73\u540e\u7684shape\u867d\u7136\u4e0e\u539f\u56fe\u50cf\u4e0d\u4e00\u81f4，\u4f46\u662f\u603b\u50cf\u7d20\u6570\u91cf\u662f\u4e00\u81f4\u7684，\u4e3a\u5f53\u524d\u901a\u9053\u6570 * \u56fe\u5f62\u5c3a\u5bf8（height * width）<\/p>\n

      Dropout\u5728\u6a21\u578b\u6784\u5efa\u4e2d\u6709\u4ec0\u4e48\u4f5c\u7528?<\/p>\n

      \u6211\u4eec\u5728\u770b\u5230\u63cf\u5199\u590f\u5929\u7684\u8bd7\u53e5\u65f6，\u4f1a\u542c\u5230\u201c\u5c0f\u8377\u624d\u9732\u5c16\u5c16\u89d2，\u65e9\u6709\u873b\u8713\u7acb\u4e0a\u5934\u201d，\u597d\u50cf\u4e00\u60f3\u5230\u8377\u82b1，\u8111\u6d77\u4e2d\u5c31\u6d6e\u73b0\u51fa\u70c8\u65e5\u708e\u708e\u7684\u590f\u65e5，\u6c60\u5858\u91cc\u7684\u90a3\u4e00\u7247\u906e\u9633\u4f1e，\u53ef\u662f，\u5982\u679c\u8377\u82b1\u4e0d\u51fa\u73b0\u5728\u6c60\u5858，\u800c\u662f\u51fa\u73b0\u5728\u2026\u2026\u51b0\u7bb1\u5462😀<\/p>\n

      \u6211\u5c0f\u65f6\u5019\u559c\u6b22\u5403\u8377\u53f6\u996d，\u5546\u8d29\u4eec\u53ef\u80fd\u4f1a\u4ece\u51b0\u7bb1\u62ff\u51fa\u81ea\u5df1\u5305\u597d\u7684\u8377\u53f6\u996d\u653e\u5230\u84b8\u7b3c\u91cc\u84b8\u719f，\u6a21\u578b\u5728\u770b\u5230\u8377\u53f6\u996d\u7684\u8377\u53f6\u65f6\u4f1a\u8bc6\u522b\u5230\u8fd9\u662f\u8377\u53f6\u561b？\u8fd9\u5c31\u6d89\u53ca\u5230\u6a21\u578b\u7684\u8bc6\u522b\u6a21\u5f0f\u4e86，\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d，\u5982\u679c\u8377\u53f6\u603b\u662f\u51fa\u73b0\u5728\u70c8\u65e5\u708e\u708e\u7684\u6c60\u5858\u91cc，\u6a21\u578b\u5c31\u4f1a\u8ba4\u4e3a\u6709\u590f\u5929\u7684\u56fe\u7247\u5c31\u662f\u8377\u82b1，\u8fd9\u79cd\u73b0\u8c61\u88ab\u79f0\u4e3a\u534f\u540c\u9002\u5e94\u73b0\u8c61，Dropout\u5c31\u662f\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\u7684，\u5b83\u4f1a\u5728\u8bad\u7ec3\u7684\u65f6\u5019\u968f\u673a\u5173\u95ed\u4e00\u90e8\u5206\u795e\u7ecf\u5143，\u4ee5\u907f\u514d\u5176\u4ed6\u795e\u7ecf\u5143\u6216\u8005\u8bc6\u522b\u6a21\u5f0f\u8fc7\u5206\u4f9d\u8d56\u67d0\u4e00\u5355\u72ec\u7684\u795e\u7ecf\u5143\u5224\u65ad，\u6bd4\u5982\u5b83\u4f1a\u5173\u95ed\u9633\u5149\u5f3a\u5ea6\u795e\u7ecf\u5143，\u8bad\u7ec3\u5728\u4e0d\u4f9d\u8d56\u8fd9\u4e00\u795e\u7ecf\u5143\u7684\u60c5\u51b5\u4e0b，\u6a21\u578b\u662f\u5426\u80fd\u591f\u51c6\u5907\u8bc6\u522b\u8377\u82b1，\u4ee5\u9f13\u52b1\u795e\u7ecf\u5143\u4e0d\u4f9d\u8d56\u5176\u4ed6\u795e\u7ecf\u5143\u7684\u7ed3\u679c\u3010\u8ba9\u5355\u4e00\u795e\u7ecf\u5143\u7684\u6743\u91cd\u5360\u6bd4\u4e0d\u518d\u8fc7\u4e8e\u5de8\u5927\u4ee5\u4ea7\u751f\u51b3\u5b9a\u6027\u4f5c\u7528\u3011<\/p>\n

      from<\/span> torch import<\/span> nn
      \nclass<\/span> SimpleCNN<\/span>(<\/span>nn.<\/span>Moudle)<\/span>:<\/span>
      \n def<\/span> __init<\/span>(<\/span>self)<\/span>:<\/span>
      \nsuper<\/span>.<\/span>__init__(<\/span>)<\/span>
      \n #……<\/span>
      \n def<\/span> forward<\/span>(<\/span>self,<\/span> x)<\/span>:<\/span>
      \n x =<\/span> self.<\/span>conv1(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>relu1(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>pool1(<\/span>x)<\/span><\/p>\n

      x =<\/span> self.<\/span>conv2(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>relu2(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>pool2(<\/span>x)<\/span><\/p>\n

      x =<\/span> self.<\/span>conv3(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>relu3(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>pool3(<\/span>x)<\/span><\/p>\n

      x =<\/span> self.<\/span>flatten(<\/span>x)<\/span><\/p>\n

      x =<\/span> self.<\/span>fc1(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>relu4(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>dropout(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>fc2(<\/span>x)<\/span>
      \n return<\/span> x<\/p>\n

      \u5728forward\u4e2d，\u6211\u4eec\u5c31\u53ef\u4ee5\u76f4\u63a5\u5e94\u7528\u5728__init__\u4e2d\u6784\u5efa\u7684\u90a3\u4e9b\u8282\u70b9\u4e86<\/p>\n

      \u60a8\u8fd8\u8bb0\u5f97\u6211\u4eec\u5728\u7b2c\u4e8c\u7ae0\u63d0\u5230\u7684[\u4ea4\u53c9\u71b5\u635f\u5931\u51fd\u6570](#\u4ea4\u53c9\u71b5\u635f\u5931\u51fd\u6570(Cross Entropy loss))\u548cAdam\u4f18\u5316\u5668\u561b,\u5b9a\u4e49\u597d\u8fd9\u4e9b\u8bca\u65ad\u2014\u2014\u8c03\u6574\u5de5\u5177\u4e4b\u540e（\u6211\u4eec\u5c31\u53ef\u4ee5\u5f00\u59cb\u5c1d\u8bd5\u8bad\u7ec3\u4e00\u4e2a\u771f\u5b9e\u53ef\u7528\u7684\u6a21\u578b\u4e86）<\/p>\n

      adam =<\/span> optim.<\/span>Adam(<\/span>model.<\/span>parameters(<\/span>)<\/span>,<\/span>lr=<\/span>0.001<\/span>)<\/span>
      \nloss_func =<\/span> nn.<\/span>CrossEntropyLoss(<\/span>)<\/span><\/p>\n

      code<\/h5>\n

      import<\/span> torch
      \nfrom<\/span> sympy import<\/span> gamma
      \nfrom<\/span> torch import<\/span> nn,<\/span> device
      \nfrom<\/span> torch.<\/span>utils.<\/span>data import<\/span> random_split,<\/span> DataLoader
      \nfrom<\/span> torchvision import<\/span> transforms<\/p>\n

      from<\/span> cnn import<\/span> *<\/span>
      \nfrom<\/span> torch import<\/span> optim
      \nfrom<\/span> data_prepare import<\/span> *<\/span><\/p>\n

      def<\/span> model_init<\/span>(<\/span>)<\/span>:<\/span>
      \n # \u6570\u636e\u51c6\u5907<\/span>
      \n download_dataset(<\/span>)<\/span>
      \n all_dataset =<\/span> OxfordFlowersDataSet(<\/span>'flowers'<\/span>)<\/span><\/p>\n

      train_size =<\/span> int<\/span>(<\/span>len<\/span>(<\/span>all_dataset)<\/span> *<\/span> 0.7<\/span>)<\/span>
      \n valid_size =<\/span> int<\/span>(<\/span>len<\/span>(<\/span>all_dataset)<\/span> *<\/span> 0.15<\/span>)<\/span>
      \n test_size =<\/span> len<\/span>(<\/span>all_dataset)<\/span> –<\/span> train_size –<\/span> valid_size<\/p>\n

      transform_simple =<\/span> transforms.<\/span>Compose(<\/span>[<\/span>
      \n transforms.<\/span>Resize(<\/span>256<\/span>)<\/span>,<\/span>
      \n transforms.<\/span>CenterCrop(<\/span>224<\/span>)<\/span>,<\/span>
      \n transforms.<\/span>ToTensor(<\/span>)<\/span>,<\/span>
      \n transforms.<\/span>Normalize(<\/span>mean=<\/span>[<\/span>0.485<\/span>,<\/span> 0.456<\/span>,<\/span> 0.406<\/span>]<\/span>,<\/span>std=<\/span>[<\/span>0.229<\/span>,<\/span> 0.224<\/span>,<\/span> 0.225<\/span>]<\/span>)<\/span>
      \n ]<\/span>)<\/span><\/p>\n

      transform_strengthen =<\/span> transforms.<\/span>Compose(<\/span>[<\/span>
      \n transforms.<\/span>RandomHorizontalFlip(<\/span>p=<\/span>0.5<\/span>)<\/span>,<\/span>
      \n transforms.<\/span>RandomRotation(<\/span>degrees=<\/span>10<\/span>)<\/span>,<\/span>
      \n transforms.<\/span>ColorJitter(<\/span>brightness=<\/span>0.2<\/span>)<\/span>,<\/span>
      \n transforms.<\/span>Resize(<\/span>256<\/span>)<\/span>,<\/span>
      \n transforms.<\/span>CenterCrop(<\/span>224<\/span>)<\/span>,<\/span>
      \n transforms.<\/span>ToTensor(<\/span>)<\/span>,<\/span>
      \n transforms.<\/span>Normalize(<\/span>mean=<\/span>[<\/span>0.485<\/span>,<\/span> 0.456<\/span>,<\/span> 0.406<\/span>]<\/span>,<\/span> std=<\/span>[<\/span>0.229<\/span>,<\/span> 0.224<\/span>,<\/span> 0.225<\/span>]<\/span>)<\/span>
      \n ]<\/span>)<\/span><\/p>\n

      train_origin_set,<\/span>valid_origin_set,<\/span>test_origin_set =<\/span> random_split(<\/span>all_dataset,<\/span>[<\/span>train_size,<\/span>valid_size,<\/span>test_size]<\/span>)<\/span><\/p>\n

      train_use_set =<\/span> SubSet(<\/span>train_origin_set,<\/span>transform_strengthen)<\/span>
      \n valid_use_set =<\/span> SubSet(<\/span>valid_origin_set,<\/span>transform_simple)<\/span>
      \n test_use_set =<\/span> SubSet(<\/span>test_origin_set,<\/span>transform_simple)<\/span><\/p>\n

      train_loader =<\/span> DataLoader(<\/span>train_use_set,<\/span>batch_size=<\/span>32<\/span>,<\/span>shuffle=<\/span>True<\/span>)<\/span>
      \n valid_loader =<\/span> DataLoader(<\/span>valid_use_set,<\/span>batch_size=<\/span>32<\/span>,<\/span>shuffle=<\/span>False<\/span>)<\/span>
      \n test_loader =<\/span> DataLoader(<\/span>test_use_set,<\/span>batch_size=<\/span>32<\/span>,<\/span>shuffle=<\/span>False<\/span>)<\/span><\/p>\n

      #\u5b9a\u4e49\u8bad\u7ec3\u6240\u9700\u7684\u5de5\u5177<\/span>
      \n device =<\/span> torch.<\/span>device(<\/span>"cuda"<\/span> if<\/span> torch.<\/span>cuda.<\/span>is_available(<\/span>)<\/span> else<\/span> "cpu"<\/span>)<\/span>
      \n model =<\/span> SimpleCNN(<\/span>)<\/span>.<\/span>to(<\/span>device)<\/span>
      \n optimizer =<\/span> optim.<\/span>Adam(<\/span>model.<\/span>parameters(<\/span>)<\/span>,<\/span> lr=<\/span>0.001<\/span>,<\/span>weight_decay=<\/span>1e-4<\/span>)<\/span>
      \n loss_func =<\/span> nn.<\/span>CrossEntropyLoss(<\/span>)<\/span><\/p>\n

      return<\/span> train_loader,<\/span>valid_loader,<\/span>test_loader,<\/span>model,<\/span>device,<\/span>optimizer,<\/span>loss_func<\/p>\n

      def<\/span> train_epoch<\/span>(<\/span>train_loader,<\/span>model,<\/span>device,<\/span>optimizer,<\/span>loss_func)<\/span>:<\/span><\/p>\n

      train_counter =<\/span> 0<\/span>
      \n running_loss =<\/span> 0.0<\/span>
      \n running_correct =<\/span> 0.0<\/span>
      \n record_print =<\/span> 0<\/span>
      \n for<\/span> index,<\/span>(<\/span>img,<\/span>label)<\/span> in<\/span> enumerate<\/span>(<\/span>train_loader)<\/span>:<\/span>
      \n img =<\/span> img.<\/span>to(<\/span>device)<\/span>
      \n label =<\/span> label.<\/span>to(<\/span>device)<\/span><\/p>\n

      optimizer.<\/span>zero_grad(<\/span>)<\/span>
      \n output =<\/span> model(<\/span>img)<\/span>
      \n loss =<\/span> loss_func(<\/span>output,<\/span>label)<\/span>
      \n loss.<\/span>backward(<\/span>)<\/span>
      \n optimizer.<\/span>step(<\/span>)<\/span><\/p>\n

      running_loss +=<\/span> loss.<\/span>item(<\/span>)<\/span>
      \n _,<\/span>predict =<\/span> output.<\/span>max<\/span>(<\/span>1<\/span>)<\/span>
      \n running_correct +=<\/span> (<\/span>predict ==<\/span> label)<\/span>.<\/span>sum<\/span>(<\/span>)<\/span>.<\/span>item(<\/span>)<\/span>
      \n train_counter +=<\/span> label.<\/span>size(<\/span>0<\/span>)<\/span>
      \n record_print +=<\/span> label.<\/span>size(<\/span>0<\/span>)<\/span><\/p>\n

      if<\/span> train_counter %<\/span> 800<\/span> ==<\/span> 0<\/span> or<\/span> index ==<\/span> len<\/span>(<\/span>train_loader)<\/span> –<\/span> 1<\/span>:<\/span>
      \n accuracy =<\/span> (<\/span>running_correct \/<\/span> record_print)<\/span> *<\/span> 100<\/span>
      \n loss =<\/span> running_loss \/<\/span> record_print<\/p>\n

      print<\/span>(<\/span>f'the <\/span>{<\/span>train_counter}<\/span><\/span> batch : accuracy is <\/span>{<\/span>accuracy:<\/span>.3f<\/span>}<\/span><\/span>% and loss is <\/span>{<\/span>loss:<\/span>.3f<\/span>}<\/span><\/span>'<\/span><\/span>)<\/span><\/p>\n

      running_correct =<\/span> 0<\/span>
      \n running_loss =<\/span> 0<\/span>
      \n record_print =<\/span> 0<\/span><\/p>\n

      def<\/span> valid_epoch<\/span>(<\/span>valid_loader,<\/span>model,<\/span>device)<\/span>:<\/span>
      \n model.<\/span>eval<\/span>(<\/span>)<\/span>
      \n correct =<\/span> 0<\/span>
      \n total =<\/span> 0<\/span>
      \n with<\/span> torch.<\/span>no_grad(<\/span>)<\/span>:<\/span>
      \n for<\/span> img,<\/span>label in<\/span> valid_loader:<\/span>
      \n img =<\/span> img.<\/span>to(<\/span>device)<\/span>
      \n label =<\/span> label.<\/span>to(<\/span>device)<\/span>
      \n output =<\/span> model(<\/span>img)<\/span>
      \n _,<\/span>predict =<\/span> output.<\/span>max<\/span>(<\/span>1<\/span>)<\/span>
      \n correct +=<\/span> (<\/span>predict ==<\/span> label)<\/span>.<\/span>sum<\/span>(<\/span>)<\/span>.<\/span>item(<\/span>)<\/span>
      \n total +=<\/span> label.<\/span>size(<\/span>0<\/span>)<\/span>
      \n accuracy =<\/span> (<\/span>correct \/<\/span> total)<\/span> *<\/span> 100<\/span>
      \n return<\/span> accuracy<\/p>\n

      if<\/span> __name__ ==<\/span> '__main__'<\/span>:<\/span>
      \n train_loader,<\/span>valid_loader,<\/span>test_loader,<\/span>model,<\/span>device,<\/span>optimizer,<\/span>loss_function=<\/span> model_init(<\/span>)<\/span>
      \n epochs =<\/span> 10<\/span>
      \n for<\/span> item in<\/span> range<\/span>(<\/span>epochs)<\/span>:<\/span>
      \n train_epoch(<\/span>train_loader,<\/span>model,<\/span>device,<\/span>optimizer,<\/span>loss_function)<\/span>
      \n print<\/span>(<\/span>f'the <\/span>{<\/span>item +<\/span> 1<\/span>}<\/span><\/span> epoch\\\\'s accuracy is <\/span>{<\/span>valid_epoch(<\/span>valid_loader,<\/span>model,<\/span>device)<\/span>}<\/span><\/span>% '<\/span><\/span>)<\/span><\/p>\n

      \u9759\u6001\u6784\u5efa\u548c\u52a8\u6001\u8ba1\u7b97\u56fe<\/h3>\n

      \u56de\u987e\u6211\u4eec\u6240\u6709\u8bad\u7ec3\u7684\u6a21\u578b，\u4e0d\u7ba1\u662f\u5feb\u9012\u6a21\u578b\u8fd8\u662fMNIST\u6570\u636e\u96c6\u8bc6\u522b\u6a21\u578b\u751a\u81f3\u662f\u6211\u4eec\u5728\u4e0a\u4e00\u7ae0\u7f16\u5199\u7684Oxford\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u6a21\u578b，\u5b83\u4eec\u90fd\u6709\u4e00\u4e2a\u5171\u540c\u70b9，\u6240\u6709\u6a21\u578b\u7684\u8bad\u7ec3\u8def\u5f84\u662f\u786e\u5b9a\u7684，\u8f93\u5165\u8282\u70b9\u662f\u4ec0\u4e48，\u5148\u7ecf\u5386\u54ea\u4e9b\u5c42，\u540e\u7ecf\u5386\u54ea\u4e9b\u5c42，\u5404\u4e2a\u5c42\u7684\u53c2\u6570\u8be5\u600e\u4e48\u786e\u5b9a，\u5b83\u4eec\u770b\u8d77\u6765\u50cf\u4e00\u4e2a\u7cbe\u5bc6\u7684\u6570\u5b66\u516c\u5f0f，\u8fd9\u5728\u8bad\u7ec3\u6a21\u578b\u4e2d\u88ab\u79f0\u4e3a\u9759\u6001\u56fe<\/p>\n

      \u5bf9\u4e8e\u57fa\u4e8e\u9759\u6001\u56fe\u7684\u6a21\u578b，\u6211\u4eec\u4e5f\u53ef\u4ee5\u4f7f\u7528nn.Sequential()\u65b9\u6cd5\u6765\u5b9a\u4e49<\/p>\n

      nn.<\/span>Sequential(<\/span>
      \nnn.<\/span>Conv2d(<\/span>3<\/span>,<\/span>32<\/span>,<\/span>3<\/span>)<\/span>
      \n nn.<\/span>ReLU(<\/span>)<\/span>,<\/span>
      \n nn.<\/span>Conv2d(<\/span>32<\/span>,<\/span>64<\/span>,<\/span>3<\/span>)<\/span>,<\/span>
      \n nn.<\/span>ReLU(<\/span>)<\/span>,<\/span>
      \n nn.<\/span>Flatten(<\/span>)<\/span>,<\/span>
      \n nn.<\/span>Linear(<\/span>64<\/span> *<\/span> 26<\/span> *<\/span> 26<\/span>,<\/span>10<\/span>)<\/span>
      \n)<\/span><\/p>\n

      \u5b83\u7684\u6784\u5efa\u6d41\u7a0b\u548c\u6211\u4eec\u73b0\u9636\u6bb5\u5728__init__\u51fd\u6570\u4e2d\u6240\u6784\u5efa\u7684\u4e00\u81f4 $$<\/p>\n

    • ;;nn.Conv2d()->out(N_i,C_{out_j});=;bias(C_{out_j});+;\\\\sum_{k=0}^{C_{in}-1}weight(C_{out_j},k);*;input(N_i,k)\\\\<\/li>\n
    • ;;ReLU(x);=;(x)^+;=;max(0,x) $$<\/li>\n
    • \u516c\u5f0f\u4e00\u662f\u5377\u79ef\u7684\u6570\u5b66\u516c\u5f0f，\u5176\u4e2d\n

      N<\/p>\n

      i<\/p>\n

      {N_i}<\/p>\n

      <\/span><\/span>N<\/span><\/span>i<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u8868\u793a\u7b2ci\u5f20\u56fe\u7247，<\/p>\n

      C<\/p>\n

      o<\/p>\n

      u<\/p>\n

      t<\/p>\n

      j<\/p>\n

      {C_{out_{j}}}<\/p>\n

      <\/span><\/span>C<\/span><\/span>o<\/span>u<\/span>t<\/span><\/span>j<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u8868\u793a\u4e00\u4e2a\u903b\u8f91\u5377\u79ef\u6838\u3010channel * \u5355\u5c42\u4e8c\u7ef4\u5377\u79ef\u6838\u3011，\u5728\u7b49\u5f0f\u53f3\u8fb9\u6211\u4eec\u5bf9\u6240\u6709\u7684\u56fe\u50cf\u8f93\u5165\u901a\u9053\u5377\u79ef，\u518d\u5c06\u6240\u6709\u5355\u5c42\u901a\u9053\u7684\u5377\u79ef\u6838\u76f8\u52a0，\u4f7f\u591a\u4e2a\u901a\u9053\u7684\u6570\u503c\u76f8\u52a0\u5f62\u6210\u4e00\u4e2a\u7ed3\u679c\u7279\u5f81\u901a\u9053，\u8fd9\u4e2a\u884c\u4e3a\u5728\u516c\u5f0f\u4e2d\u8868\u793a\u4e3a<\/p>\n

      \u2211<\/p>\n

      k<\/p>\n

      =<\/p>\n

      0<\/p>\n

      C<\/p>\n

      i<\/p>\n

      n<\/p>\n

      \u2212<\/p>\n

      1<\/p>\n

      (<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      )<\/p>\n

      {\\\\sum_{k=0}^{C_{in}-1}(…)}<\/p>\n

      <\/span><\/span>\u2211<\/span><\/span>k<\/span>=<\/span>0<\/span><\/span><\/span><\/span><\/span>C<\/span><\/span>in<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2212<\/span>1<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>(<\/span>…<\/span>)<\/span><\/span><\/span><\/span><\/span><\/span>,\u800c\u6c42\u548c\u5f53\u4e2d\u7684\u6bcf\u4e00\u9879\u5c31\u662f\u903b\u8f91\u5377\u79ef\u6838\u7684\u7b2ck\u5c42\u548c\u56fe\u50cf\u4e2d\u7684\u7b2ck\u5c42\u901a\u9053\u9010\u5143\u7d20\u76f8\u4e58，\u8fd9\u4e2a\u516c\u5f0f\u662f\u5bf9\u4e8e\u5355\u50cf\u7d20\u7684\u5377\u79ef\u5355\u4f4d\u8fd0\u7b97，\u53ef\u4ee5\u4ee5\u884c\u7d22\u5f15\u548c\u5217\u7d22\u5f15\u521b\u5efa\u5faa\u73af，\u4f7f\u5176\u8fd0\u7528\u4e8e\u6574\u4e2a\u56fe\u50cf，bias(<\/p>\n

      C<\/p>\n

      o<\/p>\n

      u<\/p>\n

      t<\/p>\n

      j<\/p>\n

      {C_{out_j}}<\/p>\n

      <\/span><\/span>C<\/span><\/span>o<\/span>u<\/span>t<\/span><\/span>j<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)\u662f\u903b\u8f91\u5377\u79ef\u6838\u7684\u504f\u7f6e\u9879，\u7528\u4e8e\u5bf9\u6574\u4e2a\u5377\u79ef\u8fdb\u884c\u5fae\u8c03<\/li>\n

    • \u516c\u5f0f\u4e8c\u662fReLU\u6fc0\u6d3b\u51fd\u6570\u7684\u6570\u5b66\u516c\u5f0f，\u7528\u4e8e\u89c4\u5b9a\u6700\u5c0f\u503c\u4e3a0\u3010max(0,x)\u3011，\u5728\u903b\u8f91\u4e0a\u7b49\u4ef7\u4e8e\u5c06\u4f46\u795e\u7ecf\u5143\u8f93\u51fa <= 0\u7684\u90e8\u5206\u8bbe\u7f6e\u4e3a0，\u4f7f\u5b83\u4e0d\u5bf9\u6574\u4f53\u7ed3\u679c\u4ea7\u751f\u5f71\u54cd（\u4e0d\u6fc0\u6d3b），\u800c > 0\u7684\u795e\u7ecf\u5143\u8f93\u51fa\u4f1a\u6309\u7167\u5b83\u539f\u672c\u7684\u5de5\u4f5c\u65b9\u5f0f\u5bf9\u6574\u4f53\u7ed3\u679c\u4ea7\u751f\u5f71\u54cd<\/li>\n

      \u5c55\u5e73\u548c\u5168\u8fde\u63a5\u6a21\u578b\u4e5f\u5728\u6570\u5b66\u4e0a\u5206\u522b\u5bf9\u5e94\u77e9\u9635\u53d8\u6362\u548c\u6784\u5efa\u591a\u9879\u5f0f，\u5bf9\u4e8e\u8fd9\u6837\u4e00\u4e2a\u7531nn.Sequential\u6784\u5efa\u7684\u6a21\u578b\u8bad\u7ec3\u8fc7\u7a0b\u6216\u8005\u73b0\u9636\u6bb5\u6211\u4eec\u5728\u7ee7\u627fnn.Module\u7c7b\u4e2d\u6240\u6784\u5efa\u7684\u6a21\u578b\u5c42\u90fd\u662f\u57fa\u4e8e\u8fd9\u6837\u4e00\u79cd\u987a\u5e8f\u7684，\u80fd\u591f\u7528\u786e\u5207\u6570\u5b66\u516c\u5f0f\u8868\u8fbe\u7684\u591a\u5c42\u8bad\u7ec3\u6a21\u578b，\u5982\u679c\u5c06\u8fd9\u6837\u4e00\u4e2a\u8fc7\u7a0b\u753b\u6210\u4e00\u4e2a\u6982\u5ff5\u56fe，\u5b83\u5927\u6982\u4f1a\u662f\u8fd9\u6837\u5b50\u7684<\/p>\n

      #mermaid-svg-ATRKOz9tpoZZAzl4{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-ATRKOz9tpoZZAzl4 .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-ATRKOz9tpoZZAzl4 .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-ATRKOz9tpoZZAzl4 .error-icon{fill:#552222;}#mermaid-svg-ATRKOz9tpoZZAzl4 .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-ATRKOz9tpoZZAzl4 .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-ATRKOz9tpoZZAzl4 .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-ATRKOz9tpoZZAzl4 .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-ATRKOz9tpoZZAzl4 .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-ATRKOz9tpoZZAzl4 .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-ATRKOz9tpoZZAzl4 .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-ATRKOz9tpoZZAzl4 .marker{fill:#333333;stroke:#333333;}#mermaid-svg-ATRKOz9tpoZZAzl4 .marker.cross{stroke:#333333;}#mermaid-svg-ATRKOz9tpoZZAzl4 svg{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-ATRKOz9tpoZZAzl4 p{margin:0;}#mermaid-svg-ATRKOz9tpoZZAzl4 .label{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-ATRKOz9tpoZZAzl4 .cluster-label text{fill:#333;}#mermaid-svg-ATRKOz9tpoZZAzl4 .cluster-label span{color:#333;}#mermaid-svg-ATRKOz9tpoZZAzl4 .cluster-label span p{background-color:transparent;}#mermaid-svg-ATRKOz9tpoZZAzl4 .label text,#mermaid-svg-ATRKOz9tpoZZAzl4 span{fill:#333;color:#333;}#mermaid-svg-ATRKOz9tpoZZAzl4 .node rect,#mermaid-svg-ATRKOz9tpoZZAzl4 .node circle,#mermaid-svg-ATRKOz9tpoZZAzl4 .node ellipse,#mermaid-svg-ATRKOz9tpoZZAzl4 .node polygon,#mermaid-svg-ATRKOz9tpoZZAzl4 .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-ATRKOz9tpoZZAzl4 .rough-node .label text,#mermaid-svg-ATRKOz9tpoZZAzl4 .node .label text,#mermaid-svg-ATRKOz9tpoZZAzl4 .image-shape .label,#mermaid-svg-ATRKOz9tpoZZAzl4 .icon-shape .label{text-anchor:middle;}#mermaid-svg-ATRKOz9tpoZZAzl4 .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-ATRKOz9tpoZZAzl4 .rough-node .label,#mermaid-svg-ATRKOz9tpoZZAzl4 .node .label,#mermaid-svg-ATRKOz9tpoZZAzl4 .image-shape .label,#mermaid-svg-ATRKOz9tpoZZAzl4 .icon-shape .label{text-align:center;}#mermaid-svg-ATRKOz9tpoZZAzl4 .node.clickable{cursor:pointer;}#mermaid-svg-ATRKOz9tpoZZAzl4 .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-ATRKOz9tpoZZAzl4 .arrowheadPath{fill:#333333;}#mermaid-svg-ATRKOz9tpoZZAzl4 .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-ATRKOz9tpoZZAzl4 .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-ATRKOz9tpoZZAzl4 .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-ATRKOz9tpoZZAzl4 .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-ATRKOz9tpoZZAzl4 .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-ATRKOz9tpoZZAzl4 .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-ATRKOz9tpoZZAzl4 .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-ATRKOz9tpoZZAzl4 .cluster text{fill:#333;}#mermaid-svg-ATRKOz9tpoZZAzl4 .cluster span{color:#333;}#mermaid-svg-ATRKOz9tpoZZAzl4 div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-ATRKOz9tpoZZAzl4 .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-ATRKOz9tpoZZAzl4 rect.text{fill:none;stroke-width:0;}#mermaid-svg-ATRKOz9tpoZZAzl4 .icon-shape,#mermaid-svg-ATRKOz9tpoZZAzl4 .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-ATRKOz9tpoZZAzl4 .icon-shape p,#mermaid-svg-ATRKOz9tpoZZAzl4 .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-ATRKOz9tpoZZAzl4 .icon-shape .label rect,#mermaid-svg-ATRKOz9tpoZZAzl4 .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-ATRKOz9tpoZZAzl4 .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-ATRKOz9tpoZZAzl4 .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-ATRKOz9tpoZZAzl4 :root{–mermaid-font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;}<\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      L2<\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      f * w = g<\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      g + b_g = h<\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      ….<\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      L1<\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      a * b = c<\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      c + d = e<\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      max(0,e) = f<\/p>\n

      <\/span><\/p>\n

      ->\u8868\u793a\u6a21\u578b\u63a8\u7406\u65f6\u7684\u65b9\u5411;—>\u8868\u793a\u6a21\u578b\u6839\u636e\u7ed3\u679c\u53cd\u9988\u8c03\u6574\u7684\u65b9\u5411<\/p>\n

      \u56fe\u793a\u7684\u56fa\u5b9a\u6a21\u5f0f\u5c31\u662f\u6211\u4eec\u73b0\u9636\u6bb5\u6240\u505a\u7684\u5de5\u4f5c，\u5b83\u9075\u5faa\u4ee5\u4e0b\u51e0\u70b9\u7ea6\u675f<\/p>\n

    • \u4e00\u5207\u64cd\u4f5c\u90fd\u6309\u7167\u56fa\u5b9a\u6d41\u7a0b\u8fdb\u884c<\/li>\n
    • \u4e0d\u80fd\u6253\u5370\u4e2d\u95f4\u503c\u8c03\u8bd5<\/li>\n
    • \u6ca1\u6709\u5206\u652f\u548c\u5faa\u73af\u6839\u636e\u6761\u4ef6\u7075\u6d3b\u8c03\u6574<\/li>\n

      \u8fd9\u6837\u7684\u5de5\u4f5c\u65b9\u5f0f\u975e\u5e38\u9002\u5408\u5728\u5df2\u77e5\u786e\u5207\u8ba1\u7b97\u56fe \/ \u8ba1\u7b97\u516c\u5f0f\u7684\u65f6\u5019\u7cbe\u786e\u63a7\u5236\u53c2\u6570\u548c\u6d41\u7a0b，\u4ee5\u8fbe\u5230\u7406\u8bba\u4e0a\u7684\u901f\u5ea6\u548c\u5185\u5b58\u6548\u7387，\u4f46\u4ee3\u4ef7\u662f\u4f1a\u653e\u5f03\u7075\u6d3b\u6027\u3010\u56e0\u4e3a\u4ee5\u56fa\u5b9a\u6d41\u7a0b\u8fd0\u884c\u3011<\/p>\n

      \u6709\u6ca1\u6709\u4e00\u79cd\u65b9\u5f0f\u662f\u80fd\u591f\u8ba9\u6211\u4eec\u53ef\u4ee5\u50cf\u5199python\u7a0b\u5e8f\u90a3\u6837\u6784\u5efa\u6211\u4eec\u7684\u6a21\u578b\u5462，\u6bd4\u5982\u6839\u636e\u6761\u4ef6\u8fd4\u56de<\/p>\n

      def<\/span> forward<\/span>(<\/span>self,<\/span>x)<\/span>:<\/span>
      \n if<\/span> self.<\/span>is_flower(<\/span>x)<\/span>:<\/span>
      \n return<\/span> self.<\/span>flower_layers(<\/span>x)<\/span>
      \n else<\/span>:<\/span>
      \n return<\/span> self.<\/span>butterfly_layers(<\/span>x)<\/span><\/p>\n

      \u6211\u5e0c\u671b\u4e00\u4e2a\u6a21\u578b\u53ef\u4ee5\u540c\u65f6\u8bad\u7ec3\u82b1\u5349\u548c\u8774\u8776，\u6216\u8005\u5f53\u4e00\u4e2a\u6570\u636e\u96c6\u540c\u65f6\u5305\u542b\u82b1\u5349\u548c\u8774\u8776\u7684\u65f6\u5019，\u6a21\u578b\u4f9d\u7136\u53ef\u4ee5\u51c6\u786e\u7684\u8bc6\u522b\u5b83\u4eec\u5462，\u6211\u4eec\u4f1a\u4e3a\u4e0d\u540c\u7684\u54c1\u79cd\u6784\u5efa\u4e0d\u540c\u7684\u8bad\u7ec3\u5c42\u3010\u6bd4\u5982\u82b1\u5349\u7684\u56fe\u7247\u5f62\u72b6\u662f(3,224,224)，\u800c\u8774\u8776\u7684\u56fe\u7247\u5f62\u72b6\u662f(3,256,256)\u3011，pytorch\u7684\u52a8\u6001\u8ba1\u7b97\u56fe\u53ef\u4ee5\u5e2e\u6211\u4eec\u505a\u5230\u8fd9\u4e00\u70b9<\/p>\n

      \u548c\u9759\u6001\u6784\u5efa\u4e00\u6837，\u52a8\u6001\u8ba1\u7b97\u56fe\u4e5f\u662f\u4e00\u79cd\u884c\u4e3a\u6a21\u5f0f，\u60a8\u65e0\u9700\u8fdb\u884c\u4efb\u4f55\u989d\u5916\u8bbe\u7f6e，\u65e0\u9700\u5bf9\u73b0\u6709\u7684\u8bbe\u8ba1\u6a21\u5f0f\u5927\u52a8\u5e72\u6208，\u53ea\u9700\u50cf\u5199\u65e5\u5e38\u5de5\u5177\u90a3\u6837\u8fdb\u884c\u5224\u65ad<\/p>\n

      \u4e0a\u8ff0\u6839\u636e\u5bf9\u8c61\u65b9\u6cd5\u5224\u65ad\u8f93\u5165\u56fe\u50cf\u7684\u7c7b\u522b\u4ece\u800c\u8c03\u7528\u4e0d\u540c\u7684\u6784\u5efa\u5c42\u5e76\u8fd4\u56de\u7ed3\u679c，\u8fd9\u5728pytorch\u7684\u8bad\u7ec3\u6a21\u5f0f\u4e2d\u662f\u8fd9\u6837\u4f53\u73b0\u7684<\/p>\n

      #mermaid-svg-WZnUZd0vdRWLHx5W{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-WZnUZd0vdRWLHx5W .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-WZnUZd0vdRWLHx5W .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-WZnUZd0vdRWLHx5W .error-icon{fill:#552222;}#mermaid-svg-WZnUZd0vdRWLHx5W .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-WZnUZd0vdRWLHx5W .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-WZnUZd0vdRWLHx5W .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-WZnUZd0vdRWLHx5W .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-WZnUZd0vdRWLHx5W .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-WZnUZd0vdRWLHx5W .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-WZnUZd0vdRWLHx5W .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-WZnUZd0vdRWLHx5W .marker{fill:#333333;stroke:#333333;}#mermaid-svg-WZnUZd0vdRWLHx5W .marker.cross{stroke:#333333;}#mermaid-svg-WZnUZd0vdRWLHx5W svg{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-WZnUZd0vdRWLHx5W p{margin:0;}#mermaid-svg-WZnUZd0vdRWLHx5W .label{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-WZnUZd0vdRWLHx5W .cluster-label text{fill:#333;}#mermaid-svg-WZnUZd0vdRWLHx5W .cluster-label span{color:#333;}#mermaid-svg-WZnUZd0vdRWLHx5W .cluster-label span p{background-color:transparent;}#mermaid-svg-WZnUZd0vdRWLHx5W .label text,#mermaid-svg-WZnUZd0vdRWLHx5W span{fill:#333;color:#333;}#mermaid-svg-WZnUZd0vdRWLHx5W .node rect,#mermaid-svg-WZnUZd0vdRWLHx5W .node circle,#mermaid-svg-WZnUZd0vdRWLHx5W .node ellipse,#mermaid-svg-WZnUZd0vdRWLHx5W .node polygon,#mermaid-svg-WZnUZd0vdRWLHx5W .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-WZnUZd0vdRWLHx5W .rough-node .label text,#mermaid-svg-WZnUZd0vdRWLHx5W .node .label text,#mermaid-svg-WZnUZd0vdRWLHx5W .image-shape .label,#mermaid-svg-WZnUZd0vdRWLHx5W .icon-shape .label{text-anchor:middle;}#mermaid-svg-WZnUZd0vdRWLHx5W .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-WZnUZd0vdRWLHx5W .rough-node .label,#mermaid-svg-WZnUZd0vdRWLHx5W .node .label,#mermaid-svg-WZnUZd0vdRWLHx5W .image-shape .label,#mermaid-svg-WZnUZd0vdRWLHx5W .icon-shape .label{text-align:center;}#mermaid-svg-WZnUZd0vdRWLHx5W .node.clickable{cursor:pointer;}#mermaid-svg-WZnUZd0vdRWLHx5W .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-WZnUZd0vdRWLHx5W .arrowheadPath{fill:#333333;}#mermaid-svg-WZnUZd0vdRWLHx5W .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-WZnUZd0vdRWLHx5W .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-WZnUZd0vdRWLHx5W .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-WZnUZd0vdRWLHx5W .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-WZnUZd0vdRWLHx5W .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-WZnUZd0vdRWLHx5W .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-WZnUZd0vdRWLHx5W .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-WZnUZd0vdRWLHx5W .cluster text{fill:#333;}#mermaid-svg-WZnUZd0vdRWLHx5W .cluster span{color:#333;}#mermaid-svg-WZnUZd0vdRWLHx5W div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-WZnUZd0vdRWLHx5W .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-WZnUZd0vdRWLHx5W rect.text{fill:none;stroke-width:0;}#mermaid-svg-WZnUZd0vdRWLHx5W .icon-shape,#mermaid-svg-WZnUZd0vdRWLHx5W .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-WZnUZd0vdRWLHx5W .icon-shape p,#mermaid-svg-WZnUZd0vdRWLHx5W .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-WZnUZd0vdRWLHx5W .icon-shape .label rect,#mermaid-svg-WZnUZd0vdRWLHx5W .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-WZnUZd0vdRWLHx5W .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-WZnUZd0vdRWLHx5W .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-WZnUZd0vdRWLHx5W :root{–mermaid-font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;}<\/p>\n

      <\/p>\n

      is_flower<\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      flower_layer<\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      exe_lay1<\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      exe_lay2<\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      …<\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      model<\/p>\n

      <\/span><\/p>\n

      is_flower:true<\/p>\n

      \u8fd9\u4e2a\u56fe\u4e0e\u672c\u7ae0\u6807\u9898\u201c\u52a8\u6001\u8ba1\u7b97\u56fe\u201d\u5b8c\u7f8e\u547c\u5e94，\u4e0d\u540c\u4e8e\u9759\u6001\u6784\u5efa\u7684\u628a\u6240\u6709\u8def\u5f84\u987a\u5e8f\u8fde\u63a5\u6210\u4e3a\u9759\u6001\u56fe，\u52a8\u6001\u8ba1\u7b97\u56fe\u4f1a\u6839\u636e\u6761\u4ef6\u5b9e\u65f6\u8ba1\u7b97，\u4ece\u800c\u53ea\u6784\u5efa\u4e0e\u6761\u4ef6\u7b26\u5408\u7684\u8bad\u7ec3\u5c42，\u5f53\u524d\u8bad\u7ec3\u7ed3\u675f\u6a21\u578b\u6839\u636e\u5404\u5c42\u53cd\u9988\u8c03\u6574\u53c2\u6570\u540e，\u4f1a\u62b9\u9664\u5f53\u524d\u8bad\u7ec3\u5c42，\u91cd\u65b0\u6839\u636e\u4e0b\u4e00\u8f6e\u8bad\u7ec3\u7684\u65b0\u6761\u4ef6\u6784\u5efa\u7b26\u5408\u6761\u4ef6\u7684\u8bad\u7ec3\u5c42<\/p>\n

      #mermaid-svg-q2hK2ZGXILZ4e1gS{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-q2hK2ZGXILZ4e1gS .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-q2hK2ZGXILZ4e1gS .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-q2hK2ZGXILZ4e1gS .error-icon{fill:#552222;}#mermaid-svg-q2hK2ZGXILZ4e1gS .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-q2hK2ZGXILZ4e1gS .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-q2hK2ZGXILZ4e1gS .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-q2hK2ZGXILZ4e1gS .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-q2hK2ZGXILZ4e1gS .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-q2hK2ZGXILZ4e1gS .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-q2hK2ZGXILZ4e1gS .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-q2hK2ZGXILZ4e1gS .marker{fill:#333333;stroke:#333333;}#mermaid-svg-q2hK2ZGXILZ4e1gS .marker.cross{stroke:#333333;}#mermaid-svg-q2hK2ZGXILZ4e1gS svg{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-q2hK2ZGXILZ4e1gS p{margin:0;}#mermaid-svg-q2hK2ZGXILZ4e1gS .label{font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-q2hK2ZGXILZ4e1gS .cluster-label text{fill:#333;}#mermaid-svg-q2hK2ZGXILZ4e1gS .cluster-label span{color:#333;}#mermaid-svg-q2hK2ZGXILZ4e1gS .cluster-label span p{background-color:transparent;}#mermaid-svg-q2hK2ZGXILZ4e1gS .label text,#mermaid-svg-q2hK2ZGXILZ4e1gS span{fill:#333;color:#333;}#mermaid-svg-q2hK2ZGXILZ4e1gS .node rect,#mermaid-svg-q2hK2ZGXILZ4e1gS .node circle,#mermaid-svg-q2hK2ZGXILZ4e1gS .node ellipse,#mermaid-svg-q2hK2ZGXILZ4e1gS .node polygon,#mermaid-svg-q2hK2ZGXILZ4e1gS .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-q2hK2ZGXILZ4e1gS .rough-node .label text,#mermaid-svg-q2hK2ZGXILZ4e1gS .node .label text,#mermaid-svg-q2hK2ZGXILZ4e1gS .image-shape .label,#mermaid-svg-q2hK2ZGXILZ4e1gS .icon-shape .label{text-anchor:middle;}#mermaid-svg-q2hK2ZGXILZ4e1gS .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-q2hK2ZGXILZ4e1gS .rough-node .label,#mermaid-svg-q2hK2ZGXILZ4e1gS .node .label,#mermaid-svg-q2hK2ZGXILZ4e1gS .image-shape .label,#mermaid-svg-q2hK2ZGXILZ4e1gS .icon-shape .label{text-align:center;}#mermaid-svg-q2hK2ZGXILZ4e1gS .node.clickable{cursor:pointer;}#mermaid-svg-q2hK2ZGXILZ4e1gS .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-q2hK2ZGXILZ4e1gS .arrowheadPath{fill:#333333;}#mermaid-svg-q2hK2ZGXILZ4e1gS .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-q2hK2ZGXILZ4e1gS .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-q2hK2ZGXILZ4e1gS .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-q2hK2ZGXILZ4e1gS .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-q2hK2ZGXILZ4e1gS .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-q2hK2ZGXILZ4e1gS .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-q2hK2ZGXILZ4e1gS .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-q2hK2ZGXILZ4e1gS .cluster text{fill:#333;}#mermaid-svg-q2hK2ZGXILZ4e1gS .cluster span{color:#333;}#mermaid-svg-q2hK2ZGXILZ4e1gS div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-q2hK2ZGXILZ4e1gS .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-q2hK2ZGXILZ4e1gS rect.text{fill:none;stroke-width:0;}#mermaid-svg-q2hK2ZGXILZ4e1gS .icon-shape,#mermaid-svg-q2hK2ZGXILZ4e1gS .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-q2hK2ZGXILZ4e1gS .icon-shape p,#mermaid-svg-q2hK2ZGXILZ4e1gS .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-q2hK2ZGXILZ4e1gS .icon-shape .label rect,#mermaid-svg-q2hK2ZGXILZ4e1gS .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-q2hK2ZGXILZ4e1gS .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-q2hK2ZGXILZ4e1gS .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-q2hK2ZGXILZ4e1gS :root{–mermaid-font-family:\\”trebuchet ms\\”,verdana,arial,sans-serif;}<\/p>\n

      <\/p>\n

      ! is_flower<\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      butterfly_layer<\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      exe_lay1<\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      exe_lay2<\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      …<\/p>\n

      <\/span><\/p>\n

      <\/p>\n

      model<\/p>\n

      <\/span><\/p>\n

      is_flower:false<\/p>\n

      \u60a8\u8fd8\u53ef\u4ee5\u5728\u52a8\u6001\u6a21\u5f0f\u4e2d\u8fdb\u884c\u8f93\u51fa\u8c03\u8bd5，\u4ee5\u6355\u6349\u6a21\u578b\u5728\u8bad\u7ec3\u9636\u6bb5\u7684\u5f02\u5e38\u8868\u73b0<\/p>\n

      def<\/span> forward<\/span>(<\/span>self,<\/span>x)<\/span>:<\/span>
      \n x =<\/span> self.<\/span>conv1(<\/span>x)<\/span>
      \n if<\/span> x.<\/span>std(<\/span>)<\/span> <<\/span> 0.1<\/span>
      \n print<\/span>(<\/span>f"Problem detected! Variance: <\/span>{<\/span>x.<\/span>std(<\/span>)<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span>
      \n x =<\/span> self.<\/span>conv2(<\/span>x)<\/span>
      \n return<\/span> x<\/p>\n

      \u5982\u679c\u60a8\u9700\u8981\u4f7f\u7528\u52a8\u6001\u8ba1\u7b97\u56fe\u6a21\u5f0f，\u60a8\u9700\u8981\u4f7f\u7528nn\u6a21\u5757（\u7ee7\u627fnn.Module）\u6765\u6784\u5efa\u6a21\u578b，nn.Sequential(\u2026)\u66f4\u50cf\u662f\u4e00\u4e2a\u65b9\u6cd5，\u60a8\u4e0d\u4f1a\u5728python\u65b9\u6cd5\u7684\u53c2\u6570\u5217\u8868\u91cc\u9762\u5199\u5206\u652f\u548c\u5faa\u73af\u7b49\u903b\u8f91，\u4f46\u662f\u5728\u51fd\u6570\u4f53def __forward__(self,x):\u4e2d\u60a8\u53ef\u4ee5\u8fd9\u4e48\u505a<\/p>\n

      \u5c3d\u7ba1\u52a8\u6001\u56fe\u7684\u7075\u6d3b\u6027\u80fd\u5e2e\u52a9\u60a8\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u505a\u8bb8\u591a\u60a8\u60f3\u505a\u7684\u4e8b\u60c5，\u4f46\u5b83\u5e76\u975e\u5b8c\u7f8e\u65e0\u7455，\u5b83\u6709\u6027\u80fd\u4e0a\u7684\u5f00\u9500，pytorch\u6bcf\u6b21\u9700\u8981\u6839\u636e\u6761\u4ef6\u521b\u5efa\u65b0\u7684\u52a8\u6001\u56fe，\u4e14\u8f93\u51fa\u4f1a\u6839\u636e\u6761\u4ef6\u7684\u4e0d\u540c\u800c\u53d8\u5316，\u96be\u4ee5\u505a\u5230\u50cf\u9759\u6001\u56fe\u90a3\u6837\u7cbe\u786e\u4f18\u5316，\u4f46\u662f\u5b83\u63d0\u4f9b\u7684\u7075\u6d3b\u6027\u5bf9\u4e8e\u7814\u7a76\u4eba\u5458\u548c\u5f00\u53d1\u4eba\u5458\u6765\u8bf4\u610f\u5473\u7740\u53ef\u4ee5\u8bad\u7ec3\u51fa\u66f4\u5feb\u7684\u8fed\u4ee3\u3001\u66f4\u5bb9\u6613\u8c03\u8bd5\u548c\u66f4\u5177\u8868\u73b0\u529b\u7684\u6a21\u578b<\/p>\n

      \u6a21\u5757\u5316\u67b6\u6784<\/h3>\n

      \u5982\u679c\u60a8\u64c5\u7528\u590d\u5236\u7c98\u8d34\u64cd\u4f5c，\u6a21\u578b\u6784\u5efa\u662f\u4e00\u4ef6\u80fd\u591f\u5feb\u901f\u5b8c\u6210\u7684\u4e8b\u60c5，\u53ef\u662f\u5982\u679c\u6211\u4eec\u670950\u5c42，100\u5c42\u751a\u81f3\u66f4\u591a，\u6211\u4eec\u9700\u8981\u628a\u5377\u79ef->\u6fc0\u6d3b->\u6c60\u5316\u8fd9\u4e9b\u8bad\u7ec3\u5c42\u590d\u5236\u6210\u767e\u4e0a\u5343\u904d，\u7136\u540e\u4e00\u4e2a\u4e00\u4e2a\u7684\u6539\u53c2\u6570\u561b，\u8fd9\u6837\u6548\u7387\u5e76\u4e0d\u9ad8，\u66f4\u4ee4\u4eba\u7cdf\u5fc3\u7684\u662f，\u5b83\u4f1a\u628a\u4e00\u4e2a\u597d\u6d88\u606f\u53d8\u6210\u574f\u6d88\u606f<\/p>\n

      \u60a8\u7684\u8774\u8776\u8bc6\u522b\u6a21\u578b\u6784\u5efa\u5b8c\u6210，\u4ea4\u4ed8\u7ed9\u5ba2\u6237\u4f7f\u7528\u65f6，\u5ba2\u6237\u5f88\u6ee1\u610f<\/p>\n

      \u60a8\u7684\u6a21\u578b\u51c6\u786e\u7387\u5f88\u4e0d\u9519，\u4e0d\u8fc7\u6211\u4eec\u65b0\u5f15\u5165\u4e86\u4e00\u7c7b\u8774\u8776，\u5b83\u4eec\u770b\u6765\u548c\u5176\u4ed6\u8774\u8776\u4e00\u6837，\u4f46\u662f\u5728\u7ec6\u8282\u4e0a\u6709\u6240\u533a\u5206，\u4e5f\u8bb8\u53ea\u662f\u7fc5\u8180\u7684\u6591\u70b9\u6392\u5e03\u4e0d\u540c，\u4e5f\u8bb8\u53ea\u662f\u7fc5\u8180\u7a0d\u5fae\u957f\u4e86\u4e00\u4e9b，\u60a8\u80fd\u5e2e\u6211\u4eec\u6539\u5584\u6a21\u578b，\u65b0\u589e\u8fd9\u7c7b\u8774\u8776\u7684\u8bc6\u522b\u561b<\/p>\n

      \u60a8\u60f3\u5230\u4e86\u53ef\u4ee5\u65b0\u589e\u4e00\u4e2a\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u8ba9\u6a21\u578b\u589e\u5f3a\u7279\u5f81\u8bc6\u522b，\u4e8e\u662f\u60a8\u5728\u5377\u79ef\u5c42\u7684\u6fc0\u6d3b\u51fd\u6570\u5199\u4e0brelu4 = nn.ReLU()，\u5047\u8bbe\u60a8\u67093\u5c42\u5377\u79ef\u5c42，\u4e00\u5c42\u5168\u8fde\u63a5\u9690\u85cf\u5c42，\u4e00\u5c42\u8f93\u51fa\u5c42，\u60a8\u9a6c\u4e0a\u610f\u8bc6\u5230relu4\u88ab\u5168\u8fde\u63a5\u5c42\u58f0\u660e\u5e76\u4f7f\u7528\u4e86，\u4e8e\u662f\u60a8\u5c06\u540d\u79f0\u6539\u540d\u4e3arelu5，\u8fd9\u5f88\u5947\u602a，\u660e\u660e\u5b83\u5728\u7b2c\u56db\u5c42\u5377\u79ef\u795e\u7ecf\u7f51\u7edc，\u4e3a\u4ec0\u4e48\u8981\u5c06\u5b83\u79f0\u4e3arelu5\u2026\u2026，\u8fd9\u5f88\u5bb9\u6613\u4ee4\u4eba\u4ea7\u751f\u8bef\u89e3<\/p>\n

      \u6211\u4eec\u53ef\u4ee5\u8fd0\u7528nn.Sequential\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898，\u5c06\u6240\u6709\u7684\u6d41\u7a0b\u5199\u5728\u4e00\u4e2ann.Sequential\u5f53\u4e2d，\u7136\u540e\u5728__forward__(self,x)\u8c03\u7528\u8fd9\u4e2a\u6d41\u7a0b，\u6211\u4eec\u65e0\u9700\u5728\u6d41\u7a0b\u4e2d\u5bf9\u6bcf\u4e2a\u8bad\u7ec3\u5c42\u547d\u540d，\u800c\u662f\u5bf9\u6574\u4e2a\u6d41\u7a0b\u547d\u540d，\u8fd9\u53ef\u4ee5\u89e3\u51b3\u6211\u4eec\u5728\u4e0a\u8ff0\u8868\u8ff0\u4e2d\u63d0\u5230\u7684\u5355\u8bad\u7ec3\u5c42\u547d\u540d\u51b2\u7a81\u95ee\u9898<\/p>\n

      def<\/span> __init__<\/span>(<\/span>self,<\/span>x)<\/span>:<\/span>
      \n super<\/span>.<\/span>__init__(<\/span>)<\/span><\/p>\n

      block_1 =<\/span> nn.<\/span>Sequential(<\/span>
      \n nn.<\/span>Conv2d(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span>
      \n nn.<\/span>ReLU(<\/span>)<\/span>,<\/span>
      \n nn.<\/span>MaxPool2d(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span><\/p>\n

      nn.<\/span>Conv2d(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span>
      \n nn.<\/span>ReLU(<\/span>)<\/span>,<\/span>
      \n nn.<\/span>MaxPool2d(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span><\/p>\n

      nn.<\/span>Conv2d(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span>
      \n nn.<\/span>ReLU(<\/span>)<\/span>,<\/span>
      \n nn.<\/span>MaxPool2d(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span><\/p>\n

      nn.<\/span>Conv2d(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span>
      \n nn.<\/span>ReLU(<\/span>)<\/span>,<\/span>
      \n nn.<\/span>MaxPool2d(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span>
      \n .<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>
      \n )<\/span><\/p>\n

      block2 =<\/span> nn.<\/span>Sequential(<\/span>
      \n nn.<\/span>Flatten(<\/span>)<\/span>
      \n nn.<\/span>Linear(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>
      \n nn.<\/span>ReLU(<\/span>)<\/span>,<\/span>
      \n nn.<\/span>DropOut(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>
      \n nn.<\/span>Linear(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>
      \n )<\/span><\/p>\n

      def<\/span> forward<\/span>(<\/span>self,<\/span>x)<\/span>:<\/span>
      \n x =<\/span> block_1(<\/span>x)<\/span>
      \n x =<\/span> block_2(<\/span>x)<\/span>
      \n return<\/span> x<\/p>\n

      \u6211\u4eec\u89e3\u51b3\u4e86\u547d\u540d\u95ee\u9898，\u4f46\u662f\u6548\u7387\u95ee\u9898\u4f9d\u7136\u5b58\u5728，\u5728\u4e0a\u8ff0\u4ee3\u7801\u4e2d，\u6211\u4eec\u4ecd\u7136\u9700\u8981\u5199\u5197\u957f\u7684\u6a21\u578b\u5b9a\u4e49，\u4f46\u662f\u597d\u5728\u6211\u4eec\u5e76\u4e0d\u9700\u8981\u5199\u5197\u957f\u7684\u6a21\u578b\u8c03\u7528<\/p>\n

      \u80fd\u5426\u5c06\u8fd9\u4e9b\u91cd\u590d\u7684nn.Conv2d(\u2026)，nn.ReLU()，nn.MaxPool2d(……)\u5199\u6210\u4e00\u4e2a\u901a\u7528\u7684\u5f62\u5f0f\u5462？\u5176\u5b9e\u60a8\u5728\u6570\u636e\u5904\u7406\u90a3\u4e00\u7ae0\u8282\u4e2d\u5df2\u7ecf\u505a\u8fc7\u7c7b\u4f3c\u7684\u4e8b\u60c5\u4e86<\/p>\n

      \u60a8\u53ef\u4ee5\u67e5\u9605\u7b2c\u4e09\u7ae0\u2014\u2014\u6570\u636e\u589e\u5f3a<\/p>\n

      \u4e3a\u4e86\u7ed9\u4e0d\u540c\u7684\u6570\u636e\u96c6\u5e94\u7528\u4e0d\u540c\u7684\u53d8\u6362，\u60a8\u5199\u4e86\u4e00\u4e2a\u5305\u88c5\u7c7bSubSet，\u5b83\u540c\u6837\u7ee7\u627f\u81eaDataSet\u7c7b，\u5b9e\u73b0\u4e86\u83b7\u53d6\u65b9\u5f0f\u7b49\u65b9\u6cd5，\u6240\u4ee5\u53ef\u4ee5\u76f4\u63a5\u4f20\u5165train_loader\u53c2\u6570\u88ab\u8c03\u7528<\/p>\n

      \u4e0e\u4e4b\u76f8\u4f3c\u7684\u60c5\u51b5\u662f\u4e0b\u4e00\u5c42\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u5c42\u7684\u8f93\u5165\u5c31\u662f\u4e0a\u4e00\u5c42\u795e\u7ecf\u7f51\u7edc\u5c42\u7684\u8f93\u51fa，\u66f4\u5de7\u5999\u7684\u662f\u60a8\u4f9d\u7136\u53ef\u4ee5\u7528nn.Module\u89c4\u8303\u3010\u5b9a\u4e49\u7ee7\u627fnn.Module\u3011\u7684\u7c7b，\u5b83\u7684\u8c03\u7528\u65b9\u5f0f\u548c\u8f93\u51fa\u90fd\u662f\u60a8\u6240\u719f\u6089\u4e14\u5728pytorch\u89c4\u8303\u4e0b\u7684\u5199\u6cd5，\u60a8\u4e0d\u5fc5\u62c5\u5fc3\u4e0d\u9002\u914d\u7684\u95ee\u9898\u3002\u6216\u8005\u6211\u4eec\u53ef\u4ee5\u8fd9\u6837\u5b50\u53bb\u505a<\/p>\n

      \u6211\u4eec\u5c06nn.Conv2d(\u2026)，nn.ReLU()，nn.MaxPool2d(……)\u5c01\u88c5\u6210\u4e00\u4e2a\u9ed1\u76d2\u5b50，\u4e0d\u53bb\u7ba1\u91cc\u9762\u505a\u4e86\u4ec0\u4e48，\u53ea\u5173\u6ce8\u8f93\u5165\u6570\u636e\u548c\u8f93\u51fa\u6570\u636e，\u5bf9\u4e8e\u8f93\u5165\u6765\u8bf4，\u5b83\u662f\u4e0a\u4e00\u8f6e\u7684\u8f93\u51fa\u6216\u8005\u539f\u59cb\u56fe\u50cf，\u5bf9\u4e8e\u8f93\u51fa\u6765\u8bf4，\u5b83\u4f1a\u4f5c\u7528\u4e8e\u4e0b\u4e00\u8f6e\u7684\u8f93\u5165\u3002<\/p>\n

      class<\/span> SimpleCNN<\/span>(<\/span>nn.<\/span>Module)<\/span>:<\/span>
      \n def<\/span> __init__<\/span>(<\/span>self)<\/span>:<\/span>
      \n super<\/span>(<\/span>)<\/span>.<\/span>__init__(<\/span>)<\/span>
      \n block_1 =<\/span> nn.<\/span>Sequential(<\/span>
      \n # chain1 [a black box]<\/span>
      \n nn.<\/span>Conv2d(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span>
      \n nn.<\/span>ReLU(<\/span>)<\/span>,<\/span>
      \n nn.<\/span>MaxPool2d(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span><\/p>\n

      # chain2 [a black box] (last output as the level input)<\/span>
      \n nn.<\/span>Conv2d(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span>
      \n nn.<\/span>ReLU(<\/span>)<\/span>,<\/span>
      \n nn.<\/span>MaxPool2d(<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span>
      \n .<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span>
      \n )<\/span>
      \n .<\/span>.<\/span>.<\/span>.<\/span>.<\/span>.<\/span><\/p>\n

      \u8fd9\u4e0d\u6b63\u662f\u6211\u4eec\u7684\u94fe\u5f0f\u8c03\u7528\u561b,\u81f3\u4e8e\u94fe\u5f0f\u8c03\u7528\u7684\u7ec6\u8282\u3010\u6bcf\u4e00\u4e2a\u8bad\u7ec3\u5c42\u3011，\u5916\u90e8\u8c03\u7528\u8005\u5e76\u4e0d\u5173\u5fc3，\u5c31\u50cfforward\u8c03\u7528block1(x)\u5c31\u53ef\u4ee5\u505a\u597d\u4e00\u5207\u8bad\u7ec3\u5de5\u4f5c，\u5c01\u88c5\u7c7b\u4e5f\u7ee7\u627f\u4e86nn.Module，\u8fd9\u610f\u5473\u7740\u8c03\u7528\u8005\u53ea\u9700\u8981\u8c03\u7528block1(x)， block(x)\u5c31\u53ef\u4ee5\u5c06x\u4f20\u5165\u7c7b\u5185\u90e8\u6784\u9020\u5bf9\u8c61\u8fdb\u884c\u8c03\u7528，\u4e3a\u4ec0\u4e48\u53ef\u4ee5\u8fd9\u4e48\u505a？\u5176\u5b9enn.Conv2d\u4e5f\u662f\u7ee7\u627fnn.Module，\u73b0\u5728\u60a8\u7684\u7c7b\u4e5f\u7ee7\u627f\u81eann.Module\u4e86，\u60a8\u4f20\u9012x\u548c\u8981\u6c42x\u8fd4\u56de\u7ed3\u679c\u7684\u8fc7\u7a0b\u90fd\u662fnn.Module\u8fd9\u4e2a\u7c7b\u89c4\u5b9a\u7684，\u6240\u4ee5\u60a8\u7684\u505a\u6cd5\u548cnn.Conv2d\u4e00\u6837\u662f\u7b26\u5408\u89c4\u8303\u7684，\u53ea\u4e0d\u8fc7\u60a8\u5728\u51fd\u6570\u5185\u90e8\u7684\u903b\u8f91\u4f1a\u4e0eConv2d\u8fd9\u4e9b\u51fd\u6570\u4e0d\u540c，\u4f46\u662f\u4ece\u4f20\u9012\u7684\u8f93\u5165\u7c7b\u578b\u548c\u8f93\u51fa\u7c7b\u578b\u6765\u770b，\u4e24\u8005\u90fd\u662f\u4e00\u6837\u7684；\u5173\u4e8e\u5e95\u5c42\u7ec6\u8282，\u5982\u679c\u60a8\u611f\u5174\u8da3\u53ef\u4ee5\u67e5\u9605\u66f4\u591a\u8d44\u6599\u8fdb\u884c\u63a2\u7d22，\u6a21\u578b\u5185\u90e8\u9700\u8981\u5b9a\u4e49\u6d41\u7a0b\u7ec6\u8282\u4f5c\u4e3a\u8c03\u7528\u903b\u8f91，\u4e0e\u6b64\u7c7b\u4f3c，\u6211\u4eec\u7684\u7ee7\u627f\u7c7b\u4e5f\u9700\u8981\u5b9a\u4e49\u6211\u4eec\u7684\u5355\u8bad\u7ec3\u5c42\u4f5c\u4e3a\u8c03\u7528\u903b\u8f91<\/p>\n

      \u8ba9\u6211\u4eec\u5c1d\u8bd5\u8fd9\u6837\u7684\u6a21\u5757\u5316\u6784\u5efa，\u8fd8\u8bb0\u5f97\u6211\u4eec\u8bad\u7ec3oxford\u6570\u636e\u96c6\u7684\u8bad\u7ec3\u7c7b\u561b，\u6211\u4eec\u53ef\u4ee5\u7528\u8fd9\u79cd\u65b9\u5f0f\u91cd\u6784，\u8ba9\u5b83\u66f4\u52a0\u7cbe\u7b80<\/p>\n

      from<\/span> torch import<\/span> nn<\/p>\n

      class<\/span> CNNLevel<\/span>(<\/span>nn.<\/span>Module)<\/span>:<\/span>
      \n def<\/span> __init__<\/span>(<\/span>self,<\/span>in_channel,<\/span>out_channel)<\/span>:<\/span>
      \n super<\/span>(<\/span>)<\/span>.<\/span>__init__(<\/span>)<\/span>
      \n self.<\/span>cnn_general_step =<\/span> nn.<\/span>Sequential(<\/span>
      \n nn.<\/span>Conv2d(<\/span>in_channel,<\/span> out_channel,<\/span> kernel_size=<\/span>3<\/span>,<\/span> padding=<\/span>1<\/span>)<\/span>,<\/span>
      \n nn.<\/span>ReLU(<\/span>)<\/span>,<\/span>
      \n nn.<\/span>MaxPool2d(<\/span>kernel_size=<\/span>2<\/span>,<\/span> stride=<\/span>2<\/span>)<\/span>,<\/span>
      \n )<\/span>
      \n def<\/span> forward<\/span>(<\/span>self,<\/span>x)<\/span>:<\/span>
      \n x =<\/span> self.<\/span>cnn_general_step(<\/span>x)<\/span>
      \n return<\/span> x<\/p>\n

      class<\/span> SimpleCNN<\/span>(<\/span>nn.<\/span>Module)<\/span>:<\/span>
      \n def<\/span> __init__<\/span>(<\/span>self)<\/span>:<\/span>
      \n super<\/span>(<\/span>)<\/span>.<\/span>__init__(<\/span>)<\/span>
      \n self.<\/span>cnn_step =<\/span> nn.<\/span>Sequential(<\/span>
      \n CNNLevel(<\/span>in_channel=<\/span>3<\/span>,<\/span> out_channel=<\/span>32<\/span>)<\/span>,<\/span>
      \n CNNLevel(<\/span>in_channel=<\/span>32<\/span>,<\/span> out_channel=<\/span>64<\/span>)<\/span>,<\/span>
      \n CNNLevel(<\/span>in_channel=<\/span>64<\/span>,<\/span> out_channel=<\/span>128<\/span>)<\/span>,<\/span>
      \n )<\/span>
      \n self.<\/span>flatten =<\/span> nn.<\/span>Flatten(<\/span>)<\/span>
      \n self.<\/span>dropout1 =<\/span> nn.<\/span>Dropout(<\/span>p=<\/span>0.2<\/span>)<\/span><\/p>\n

      self.<\/span>fc1 =<\/span> nn.<\/span>Linear(<\/span>128<\/span> *<\/span> 28<\/span> *<\/span> 28<\/span>,<\/span> 256<\/span>)<\/span>
      \n self.<\/span>relu4 =<\/span> nn.<\/span>ReLU(<\/span>)<\/span>
      \n self.<\/span>dropout2 =<\/span> nn.<\/span>Dropout(<\/span>0.5<\/span>)<\/span>
      \n self.<\/span>fc2 =<\/span> nn.<\/span>Linear(<\/span>256<\/span>,<\/span> 102<\/span>)<\/span>
      \n def<\/span> forward<\/span>(<\/span>self,<\/span> x)<\/span>:<\/span>
      \n x =<\/span> self.<\/span>cnn_step(<\/span>x)<\/span><\/p>\n

      x =<\/span> self.<\/span>flatten(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>dropout1(<\/span>x)<\/span><\/p>\n

      x =<\/span> self.<\/span>fc1(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>relu4(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>dropout2(<\/span>x)<\/span>
      \n x =<\/span> self.<\/span>fc2(<\/span>x)<\/span>
      \n return<\/span> x<\/p>\n

      \u6a21\u5757\u5316\u6784\u5efa\u4f1a\u51cf\u5c11\u5197\u4f59\u8bbe\u8ba1，\u8fd9\u4f1a\u6781\u5927\u7684\u63d0\u5347\u5f00\u53d1\u6548\u7387，\u7279\u522b\u662f\u5728\u8bad\u7ec3\u5c42\u7684\u6570\u91cf\u504f\u591a\u7684\u65f6\u5019<\/p>\n

      \u65e2\u7136\u6a21\u5757\u5316\u6784\u5efa\u5982\u6b64\u9ad8\u6548，\u90a3\u4e3a\u4ec0\u4e48\u6211\u4eec\u4e00\u5f00\u59cb\u8fd8\u9700\u8981\u4ece\u91cd\u590d\u5b9a\u4e49\u8bad\u7ec3\u5c42\u5f00\u59cb\u5b66\u4e60\u5462，\u8fd9\u4ecd\u662f\u6709\u5fc5\u8981\u7684\u3002\u5bf9\u4e8e\u521a\u63a5\u89e6pytorch\u7684\u5f00\u53d1\u8005\u6765\u8bf4，\u5199\u51fa\u6240\u6709\u7684\u91cd\u590d\u7684\u5c42\u80fd\u591f\u5e2e\u52a9\u6211\u4eec\u7406\u89e3\u6784\u5efa\u539f\u7406，\u77e5\u9053\u54ea\u4e9b\u8bad\u7ec3\u5c42\u662f\u91cd\u590d\u7684，\u5e94\u8be5\u5c06\u5b83\u6784\u5efa\u6210\u4e00\u4e2a\u5355\u72ec\u7684\u6a21\u5757，\u54ea\u4e9b\u8bad\u7ec3\u5c42\u7684\u6a21\u5f0f\u662f\u5355\u4e00\u7684，\u4e0d\u5fc5\u5c06\u5b83\u6784\u5efa\u6210\u4e00\u4e2a\u6a21\u5757，\u5c31\u50cf\u6211\u4eec\u5e76\u6ca1\u6709\u5c06\u540e\u9762\u7684\u5168\u8fde\u63a5\u5c42\u6784\u5efa\u6210\u4e00\u4e2a\u6a21\u5757，\u56e0\u4e3a\u6211\u4eec\u53d1\u73b0，\u4e24\u4e2a\u5168\u8fde\u63a5\u5c42\u6240\u505a\u7684\u5de5\u4f5c\u5e76\u4e0d\u4e00\u6837，\u7b2c\u4e00\u4e2a\u8fde\u63a5\u5c42\u4f5c\u4e3a\u9690\u85cf\u5c42，\u4f7f\u7528ReLU()\u6fc0\u6d3b\u62df\u5408\u590d\u6742\u7279\u5f81，\u540c\u65f6\u5e94\u7528dropout\u907f\u514d\u795e\u7ecf\u5143\u4f9d\u8d56\u5355\u4e00\u6a21\u5f0f\u8fdb\u884c\u8bad\u7ec3，\u4f46\u662f\u7b2c\u4e8c\u6b21\u5168\u8fde\u63a5\u5c42\u4f5c\u4e3a\u8f93\u51fa\u5c42，\u8f93\u51fa\u7ed3\u679c\u7684\u8282\u70b9\u6570\u76f4\u63a5\u5c31\u662f\u8f93\u51fa\u7279\u5f81，\u4e0d\u80fd\u5ffd\u7565\u4efb\u4f55\u4e00\u4e2a\u795e\u7ecf\u5143\u3010\u8fd9\u91cc\u662f\u8f93\u51fa\u7279\u5f81\u3011\u7684\u8f93\u51fa，\u4e5f\u4e0d\u80fd\u5173\u95ed\u67d0\u4e00\u795e\u7ecf\u5143，\u6240\u4ee5\u4e0d\u80fd\u5728\u8f93\u51fa\u5c42\u540e\u9762\u4f7f\u7528\u6fc0\u6d3b\u51fd\u6570\u548cdropout\u3002\u8fd9\u5e76\u4e0d\u662f\u4e00\u4e2a\u56fa\u5b9a\u7684\u6a21\u5f0f，\u60a8\u9700\u8981\u5b9a\u4e49\u4e24\u4e2a\u7c7b\u6216\u8bb8\u5728\u4e00\u4e2a\u7c7b\u91cc\u9762\u540c\u65f6\u505a\u8fd9\u4e24\u4e2a\u4e8b\u60c5，\u8fd9\u76f8\u8f83\u4e8e\u76f4\u63a5\u5728\u8bad\u7ec3\u6a21\u578b\u4e2d\u5b9a\u4e49\u90fd\u770b\u8d77\u6765\u66f4\u52a0\u590d\u6742，\u800c\u4e14\u5728\u4e00\u4e2a\u7c7b\u91cc\u9762\u540c\u65f6\u505a\u8fd9\u4e24\u4e2a\u4e8b\u60c5\u8fd9\u79cd\u60c5\u51b5\u5e76\u4e0d\u901a\u7528，\u9664\u4e86\u6a21\u578b\u5728\u6700\u540e\u8f93\u51fa\u7684\u65f6\u5019，\u5176\u4ed6\u5c42\u90fd\u4e0d\u4f1a\u4f5c\u4e3a\u8f93\u51fa\u5c42，\u6240\u4ee5\u8fd9\u5e76\u4e0d\u901a\u7528\u3002<\/p>\n

      \u6a21\u578b\u68c0\u67e5\u4e0e\u8c03\u8bd5<\/h3>\n

      \u6784\u5efa\u6a21\u578b\u7684\u8fc7\u7a0b\u5c31\u50cf\u57f9\u80b2\u4e00\u682a\u76c6\u683d，\u8bad\u7ec3\u5c42\u50cf\u662f\u571f\u58e4，\u5b83\u4eec\u6e90\u6e90\u4e0d\u65ad\u5730\u4e3a\u690d\u7269\u63d0\u4f9b\u517b\u5206，\u4f46\u662f\u571f\u58e4\u91cc\u9762\u6709\u4ec0\u4e48，\u5f88\u5927\u7a0b\u5ea6\u4e0a\u53d6\u51b3\u4e8e\u60a8\u4e3a\u5b83\u63d0\u4f9b\u4ec0\u4e48，\u5982\u679c\u7ed9\u5b83\u5408\u9002\u5730\u6c34\u4e0e\u65e0\u673a\u76d0，\u5b83\u5927\u6982\u7387\u4f1a\u6210\u957f，\u5982\u679c\u7ed9\u5b83\u6492\u4e0a\u996e\u6599，\u5b83\u5927\u6982\u7387\u4f1a\u67af\u7aed，\u5728\u6a21\u578b\u6784\u5efa\u4e2d，\u6211\u4eec\u4e5f\u4f1a\u63d0\u4f9b\u8bad\u7ec3\u96c6\u548c\u8bad\u7ec3\u53c2\u6570，\u8fd9\u662f\u6a21\u578b\u7684\u201c\u517b\u6599\u201d，\u5982\u679c\u5b83\u8db3\u591f\u597d，\u6a21\u578b\u7684\u5b66\u4e60\u6548\u679c\u53ef\u80fd\u5c31\u66f4\u597d，\u5982\u679c\u5b83\u4e0d\u80fd\u88ab\u6a21\u578b\u63a5\u6536\u3010\u5f62\u72b6\u4e0d\u5339\u914d\u3011，\u4ed6\u751a\u81f3\u62d2\u7edd\u8bad\u7ec3\u3010\u62a5\u9519\u3011，\u5728\u6a21\u578b\u6784\u5efa\u4e4b\u524d\u68c0\u67e5\u5b83\u6216\u8005\u5728\u8fd0\u884c\u8fc7\u7a0b\u4e2d\u8c03\u8bd5，\u662f\u786e\u4fdd\u6a21\u578b\u80fd\u7a33\u5b9a\u8fd0\u884c\u7684\u5173\u952e<\/p>\n

      \u5982\u679c\u67e5\u770b\u6a21\u578b\u7684\u4fe1\u606f\u5462，\u6211\u4eec\u53ef\u80fd\u4f1a\u60f3\u5230\u76f4\u63a5\u6253\u5370,print(model),\u8fd9\u662f\u4e00\u4e2a\u4e0d\u9519\u7684\u65b9\u5f0f，\u5b83\u4f1a\u6253\u5370\u51fa\u6211\u4eec\u5728__init__\u4e2d\u5b9a\u4e49\u7684\u8bad\u7ec3\u5c42，\u60a8\u53ef\u4ee5\u68c0\u67e5\u6a21\u578b\u662f\u5426\u6309\u7167\u60a8\u9884\u5b9a\u7684\u65b9\u5f0f\u6267\u884c，\u7ed3\u6784\u662f\u5426\u6b63\u786e<\/p>\n

      from<\/span> cnn import<\/span> *<\/span>
      \nmodel =<\/span> SimpleCNN(<\/span>)<\/span>
      \nprint<\/span>(<\/span>model)<\/span><\/p>\n

      E:\\\\Users\\\\ASUS\\\\anaconda3\\\\envs\\\\torch_env\\\\python.<\/span>exe E:\\\\pythonProject\\\\PythonProject\\\\modelPrint.<\/span>py
      \nSimpleCNN(<\/span>
      \n (<\/span>cnn_step)<\/span>: Sequential(<\/span>
      \n (<\/span>0)<\/span>: CNNLevel(<\/span>
      \n (<\/span>cnn_general_step)<\/span>: Sequential(<\/span>
      \n (<\/span>0)<\/span>: Conv2d(<\/span>3,<\/span> 32,<\/span> kernel_size=(<\/span>3,<\/span> 3)<\/span>,<\/span> stride=(<\/span>1,<\/span> 1)<\/span>,<\/span> padding=(<\/span>1,<\/span> 1)<\/span>)<\/span>
      \n (<\/span>1)<\/span>: ReLU(<\/span>)<\/span>
      \n (<\/span>2)<\/span>: MaxPool2d(<\/span>kernel_size=2,<\/span> stride=2,<\/span> padding=0,<\/span> dilation=1,<\/span> ceil_mode=False)<\/span>
      \n )<\/span>
      \n )<\/span>
      \n (<\/span>1)<\/span>: CNNLevel(<\/span>
      \n (<\/span>cnn_general_step)<\/span>: Sequential(<\/span>
      \n (<\/span>0)<\/span>: Conv2d(<\/span>32,<\/span> 64,<\/span> kernel_size=(<\/span>3,<\/span> 3)<\/span>,<\/span> stride=(<\/span>1,<\/span> 1)<\/span>,<\/span> padding=(<\/span>1,<\/span> 1)<\/span>)<\/span>
      \n (<\/span>1)<\/span>: ReLU(<\/span>)<\/span>
      \n (<\/span>2)<\/span>: MaxPool2d(<\/span>kernel_size=2,<\/span> stride=2,<\/span> padding=0,<\/span> dilation=1,<\/span> ceil_mode=False)<\/span>
      \n )<\/span>
      \n )<\/span>
      \n (<\/span>2)<\/span>: CNNLevel(<\/span>
      \n (<\/span>cnn_general_step)<\/span>: Sequential(<\/span>
      \n (<\/span>0)<\/span>: Conv2d(<\/span>64,<\/span> 128,<\/span> kernel_size=(<\/span>3,<\/span> 3)<\/span>,<\/span> stride=(<\/span>1,<\/span> 1)<\/span>,<\/span> padding=(<\/span>1,<\/span> 1)<\/span>)<\/span>
      \n (<\/span>1)<\/span>: ReLU(<\/span>)<\/span>
      \n (<\/span>2)<\/span>: MaxPool2d(<\/span>kernel_size=2,<\/span> stride=2,<\/span> padding=0,<\/span> dilation=1,<\/span> ceil_mode=False)<\/span>
      \n )<\/span>
      \n )<\/span>
      \n )<\/span>
      \n (<\/span>flatten)<\/span>: Flatten(<\/span>start_dim=1,<\/span> end_dim=–<\/span>1)<\/span>
      \n (<\/span>dropout1)<\/span>: Dropout(<\/span>p=0.<\/span>2,<\/span> inplace=False)<\/span>
      \n (<\/span>fc1)<\/span>: Linear(<\/span>in_features=100352,<\/span> out_features=256,<\/span> bias=True)<\/span>
      \n (<\/span>relu4)<\/span>: ReLU(<\/span>)<\/span>
      \n (<\/span>dropout2)<\/span>: Dropout(<\/span>p=0.<\/span>5,<\/span> inplace=False)<\/span>
      \n (<\/span>fc2)<\/span>: Linear(<\/span>in_features=256,<\/span> out_features=102,<\/span> bias=True)<\/span>
      \n)<\/span><\/p>\n

      \u8fdb\u7a0b\u5df2\u7ed3\u675f，\u9000\u51fa\u4ee3\u7801\u4e3a 0<\/p>\n

      \u4f46\u662f，\u6bcf\u4e00\u5c42\u6709\u591a\u5c11\u53c2\u6570，\u5f20\u91cf\u7684\u5f62\u72b6\u662f\u4ec0\u4e48，\u6211\u4eec\u53ef\u4ee5\u4f7f\u7528model.paraments()\u8ffd\u8e2a，\u60a8\u662f\u5426\u719f\u6089\u5b83\u5462，\u4ed6\u66fe\u4f5c\u4e3a\u5b9e\u53c2\u3010\u4ee5\u5f15\u7528\u7684\u5f62\u5f0f\u3011\u653e\u5165\u4f18\u5316\u5668\u5f53\u4e2d，\u8fd9\u610f\u5473\u7740\u4f18\u5316\u5668\u53ef\u4ee5\u8ffd\u8e2a\u6a21\u578b\u7684\u8f93\u51fa\u5e76\u6839\u636e\u7ed3\u679c\u8fdb\u884c\u4f18\u5316，\u5f53\u6211\u4eec\u62ff\u5230\u5b83\u4e4b\u540e\u4e5f\u53ef\u4ee5\u8ffd\u8e2a\u6a21\u578b\u7684\u8f93\u51fa，\u73b0\u5728\u6211\u4eec\u5e76\u4e0d\u805a\u7126\u4e8e\u8f93\u51fa\u7ed3\u679c，\u90a3\u662f\u8bad\u7ec3\u5c42\u9762\u505a\u7684\u4e8b\u60c5，\u5728\u8c03\u8bd5\u548c\u68c0\u67e5\u7684\u65f6\u5019，\u6211\u4eec\u4e3b\u8981\u89c2\u5bdf\u5f20\u91cf\u5927\u5c0f\u548c\u53c2\u6570\u4e2a\u6570<\/p>\n

      \u5982\u679c\u6211\u4eec\u4f7f\u7528print(model.paraments())，\u5e76\u4e0d\u4f1a\u5f97\u5230\u6240\u6709\u7684\u53c2\u6570，\u800c\u662f\u5f97\u5230\u4e00\u4e2a\u751f\u6210\u5668<\/p>\n

      <generator object Module.<\/span>parameters at 0x00000231C1997290><\/p>\n

      \u5b83\u91c7\u53d6\u548cDataLoader\u4e00\u6837\u7684\u7b56\u7565，\u4e0d\u4f1a\u4e00\u6b21\u6027\u5c06\u6240\u6709\u5185\u5bb9\u52a0\u8f7d\u5230\u5185\u5b58\u5f53\u4e2d，\u4e5f\u662f\u5728\u60a8\u9700\u8981\u7684\u65f6\u5019（\u5faa\u73af\u83b7\u53d6\u7684\u65f6\u5019）\u4e00\u6b21\u7ed9\u60a8\u4e00\u4e2a\u53c2\u6570，\u8fd9\u6837\u5bf9\u5185\u5b58\u66f4\u53cb\u597d<\/p>\n

      #……<\/span>
      \nfor<\/span> item in<\/span> model.<\/span>parameters(<\/span>)<\/span>:<\/span>
      \n print<\/span>(<\/span>item.<\/span>shape)<\/span><\/p>\n

      torch.<\/span>Size(<\/span>[<\/span>32,<\/span> 3,<\/span> 3,<\/span> 3]<\/span>)<\/span>
      \ntorch.<\/span>Size(<\/span>[<\/span>32]<\/span>)<\/span>
      \ntorch.<\/span>Size(<\/span>[<\/span>64,<\/span> 32,<\/span> 3,<\/span> 3]<\/span>)<\/span>
      \ntorch.<\/span>Size(<\/span>[<\/span>64]<\/span>)<\/span>
      \ntorch.<\/span>Size(<\/span>[<\/span>128,<\/span> 64,<\/span> 3,<\/span> 3]<\/span>)<\/span>
      \ntorch.<\/span>Size(<\/span>[<\/span>128]<\/span>)<\/span>
      \ntorch.<\/span>Size(<\/span>[<\/span>256,<\/span> 100352]<\/span>)<\/span>
      \ntorch.<\/span>Size(<\/span>[<\/span>256]<\/span>)<\/span>
      \ntorch.<\/span>Size(<\/span>[<\/span>102,<\/span> 256]<\/span>)<\/span>
      \ntorch.<\/span>Size(<\/span>[<\/span>102]<\/span>)<\/span><\/p>\n

      \u8fd9\u57fa\u672c\u4e0a\u5c31\u5305\u542b\u4e86\u6bcf\u4e00\u5c42\u7684\u5f20\u91cf\u5927\u5c0f，\u4f46\u662f\u8fd9\u4e9b\u6570\u636e\u7a76\u7adf\u4ee3\u8868\u4ec0\u4e48\u542b\u4e49，\u6211\u4eec\u9700\u8981\u4ee5\u7c7b\u4f3c\u952e\u503c\u7684\u65b9\u5f0f\u53bb\u7f57\u5217\u51fa\u6765，model\u5bf9\u8c61\u4e2d\u6709\u4e00\u4e2a\u540d\u4e3anamed_parameters\u7684\u65b9\u6cd5\u53ef\u4ee5\u5e2e\u52a9\u6211\u4eec\u505a\u5230\u8fd9\u4ef6\u4e8b\u60c5<\/p>\n

      for<\/span> name,<\/span>data in<\/span> model.<\/span>named_parameters(<\/span>)<\/span>:<\/span>
      \n print<\/span>(<\/span>f'<\/span>{<\/span>name}<\/span><\/span>: <\/span>{<\/span>data.<\/span>shape}<\/span><\/span>'<\/span><\/span>)<\/span><\/p>\n

      cnn_step.<\/span>0.<\/span>cnn_general_step.<\/span>0.<\/span>weight: torch.<\/span>Size(<\/span>[<\/span>32,<\/span> 3,<\/span> 3,<\/span> 3]<\/span>)<\/span>
      \ncnn_step.<\/span>0.<\/span>cnn_general_step.<\/span>0.<\/span>bias: torch.<\/span>Size(<\/span>[<\/span>32]<\/span>)<\/span>
      \ncnn_step.<\/span>1.<\/span>cnn_general_step.<\/span>0.<\/span>weight: torch.<\/span>Size(<\/span>[<\/span>64,<\/span> 32,<\/span> 3,<\/span> 3]<\/span>)<\/span>
      \ncnn_step.<\/span>1.<\/span>cnn_general_step.<\/span>0.<\/span>bias: torch.<\/span>Size(<\/span>[<\/span>64]<\/span>)<\/span>
      \ncnn_step.<\/span>2.<\/span>cnn_general_step.<\/span>0.<\/span>weight: torch.<\/span>Size(<\/span>[<\/span>128,<\/span> 64,<\/span> 3,<\/span> 3]<\/span>)<\/span>
      \ncnn_step.<\/span>2.<\/span>cnn_general_step.<\/span>0.<\/span>bias: torch.<\/span>Size(<\/span>[<\/span>128]<\/span>)<\/span>
      \nfc1.<\/span>weight: torch.<\/span>Size(<\/span>[<\/span>256,<\/span> 100352]<\/span>)<\/span>
      \nfc1.<\/span>bias: torch.<\/span>Size(<\/span>[<\/span>256]<\/span>)<\/span>
      \nfc2.<\/span>weight: torch.<\/span>Size(<\/span>[<\/span>102,<\/span> 256]<\/span>)<\/span>
      \nfc2.<\/span>bias: torch.<\/span>Size(<\/span>[<\/span>102]<\/span>)<\/span><\/p>\n

      \u6211\u4eec\u53d1\u73b0，\u8fd9\u4e9b\u6570\u636e\u4e0e\u6211\u4eec\u5728__init__\u51fd\u6570\u586b\u5145\u7684\u53c2\u6570\u76f8\u4f3c，\u4f46\u662f\u5448\u73b0\u65b9\u5f0f\u5374\u6709\u6240\u4e0d\u540c，\u4ee5\u5377\u79ef\u5c42\u4e3a\u4f8b，\u5f53\u6211\u4eec\u5199\u5165\u53c2\u6570\u7684\u65f6，\u6211\u4eec\u7684\u53c2\u6570\u586b\u5145\u65b9\u5f0f\u662fnn.Conv2d(\u8f93\u5165\u901a\u9053\u6570，\u8f93\u51fa\u901a\u9053\u6570，\u5377\u79ef\u6838\u5927\u5c0f，padding)，\u4f46\u662f\u5728\u8fd9\u91cc\u6bcf\u4e00\u4e2a\u5377\u79ef\u5c42\u7684weight\u7684\u7b2c\u4e00\u5c42\u5927\u5c0f\u662f\u8f93\u51fa\u901a\u9053\u6570，\u7b2c\u4e8c\u5c42\u5927\u5c0f\u662f\u8f93\u5165\u901a\u9053\u6570，\u5b83\u5e76\u4e0d\u662f\u4ee5\u5f00\u53d1\u8005\u7684\u89c6\u89d2\u63cf\u8ff0\u5e94\u8be5\u600e\u4e48\u505a\u3010\u5148\u8f93\u5165\u540e\u8f93\u51fa\u3011，\u800c\u662f\u4ee5pytorch\u80fd\u591f\u5904\u7406\u7684\u65b9\u5f0f\u5b9a\u4e49\u4e86\u6240\u6709\u7684\u5377\u79ef\u6838，\u4e00\u4e2a\u903b\u8f91\u5377\u79ef\u6838\u7684\u5927\u5c0f\u5c31\u662f\u8f93\u5165\u901a\u9053 * (\u5377\u79ef\u6838\u5927\u5c0f)，\u800c\u8fd9\u6837\u7684\u903b\u8f91\u5377\u79ef\u6838\u670932\u4e2a\u3010\u5bf9\u5e94\u8f93\u51fa\u901a\u905332，\u4ee5\u4e0a\u8ff0\u4ee3\u7801\u7b2c\u4e00\u5c42\u4e3a\u4f8b\u3011，\u5bf9\u5e94\u5927\u5c0f[32,3,3,3]，\u5377\u79ef\u5c42\u7684bias\u5219\u5bf9\u6bcf\u4e00\u4e2a\u8f93\u51fa\u901a\u9053\u8fdb\u884c\u504f\u7f6e\u5904\u7406，32\u4e2a\u903b\u8f91\u5377\u79ef\u6838\u4ea7\u751f32\u4e2a\u8f93\u51fa\u901a\u9053，\u5b83\u4f1a\u5bf932\u4e2a\u8f93\u51fa\u901a\u9053\u8fdb\u884c\u5904\u7406，\u6240\u4ee5\u5b83\u51c6\u5907\u4e86\u5927\u5c0f\u4e3a[32]\u7684\u504f\u7f6e\u9879\u6570\u636e\u3002<\/p>\n

      fc1\u548cfc2\u662f\u6211\u4eec\u5b9a\u4e49\u7684\u5168\u8fde\u63a5\u5c42，\u6309\u7167\u548c\u5377\u79ef\u5c42\u76f8\u4f3c\u7684\u5206\u6790，\u5c06weight\u770b\u6210\u6a21\u578b\u9700\u8981\u51c6\u5907\u7684\u6240\u6709\u53c2\u6570\u5927\u5c0f，\u6839\u636e\u516c\u5f0f<\/p>\n

      <\/p>\n

      y<\/p>\n

      1<\/p>\n

      \u2005\u200a<\/p>\n

      =
      \n \u2005\u200a<\/p>\n

      (<\/p>\n

      a<\/p>\n

      1<\/p>\n

      \u2005\u200a<\/p>\n

      \u2217
      \n \u2005\u200a<\/p>\n

      x<\/p>\n

      1<\/p>\n

      )
      \n \u2005\u200a<\/p>\n

      +
      \n \u2005\u200a<\/p>\n

      (<\/p>\n

      a<\/p>\n

      2<\/p>\n

      \u2005\u200a<\/p>\n

      \u2217
      \n \u2005\u200a<\/p>\n

      x<\/p>\n

      2<\/p>\n

      )
      \n \u2005\u200a<\/p>\n

      +
      \n \u2005\u200a<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .
      \n \u2005\u200a<\/p>\n

      +
      \n \u2005\u200a<\/p>\n

      (<\/p>\n

      a<\/p>\n

      100352<\/p>\n

      \u2005\u200a<\/p>\n

      \u2217
      \n \u2005\u200a<\/p>\n

      x<\/p>\n

      100352<\/p>\n

      )
      \n \u2005\u200a<\/p>\n

      +<\/p>\n

      b<\/p>\n

      1<\/p>\n

      y<\/p>\n

      2<\/p>\n

      \u2005\u200a<\/p>\n

      =
      \n \u2005\u200a<\/p>\n

      (<\/p>\n

      a<\/p>\n

      100353<\/p>\n

      \u2005\u200a<\/p>\n

      \u2217
      \n \u2005\u200a<\/p>\n

      x<\/p>\n

      1<\/p>\n

      )
      \n \u2005\u200a<\/p>\n

      +
      \n \u2005\u200a<\/p>\n

      (<\/p>\n

      a<\/p>\n

      100354<\/p>\n

      \u2005\u200a<\/p>\n

      \u2217
      \n \u2005\u200a<\/p>\n

      x<\/p>\n

      2<\/p>\n

      )
      \n \u2005\u200a<\/p>\n

      +
      \n \u2005\u200a<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .
      \n \u2005\u200a<\/p>\n

      +
      \n \u2005\u200a<\/p>\n

      (<\/p>\n

      a<\/p>\n

      (<\/p>\n

      2<\/p>\n

      \u2217<\/p>\n

      100352<\/p>\n

      )<\/p>\n

      \u2005\u200a<\/p>\n

      \u2217
      \n \u2005\u200a<\/p>\n

      x<\/p>\n

      100352<\/p>\n

      )
      \n \u2005\u200a<\/p>\n

      +<\/p>\n

      b<\/p>\n

      2<\/p>\n

      )<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      y<\/p>\n

      256<\/p>\n

      \u2005\u200a<\/p>\n

      =
      \n \u2005\u200a<\/p>\n

      (<\/p>\n

      a<\/p>\n

      (<\/p>\n

      255<\/p>\n

      \u2217<\/p>\n

      100353<\/p>\n

      )<\/p>\n

      \u2005\u200a<\/p>\n

      \u2217
      \n \u2005\u200a<\/p>\n

      x<\/p>\n

      1<\/p>\n

      )
      \n \u2005\u200a<\/p>\n

      +
      \n \u2005\u200a<\/p>\n

      (<\/p>\n

      a<\/p>\n

      (<\/p>\n

      (<\/p>\n

      255<\/p>\n

      \u2217<\/p>\n

      100353<\/p>\n

      )<\/p>\n

      +<\/p>\n

      1<\/p>\n

      )<\/p>\n

      \u2005\u200a<\/p>\n

      \u2217
      \n \u2005\u200a<\/p>\n

      x<\/p>\n

      2<\/p>\n

      )
      \n \u2005\u200a<\/p>\n

      +
      \n \u2005\u200a<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .<\/p>\n

      .
      \n \u2005\u200a<\/p>\n

      +
      \n \u2005\u200a<\/p>\n

      (<\/p>\n

      a<\/p>\n

      256
      \n \u2005\u200a<\/p>\n

      \u2217
      \n \u2005\u200a<\/p>\n

      100352<\/p>\n

      \u2005\u200a<\/p>\n

      \u2217
      \n \u2005\u200a<\/p>\n

      x<\/p>\n

      100352<\/p>\n

      )
      \n \u2005\u200a<\/p>\n

      +
      \n \u2005\u200a<\/p>\n

      b<\/p>\n

      256<\/p>\n

      \\\\begin{align} y_1\\\\;&=\\\\;(a_{1}\\\\;*\\\\;x_1)\\\\;+\\\\;(a_2\\\\;*\\\\;x_2)\\\\;+\\\\;……\\\\;+\\\\;(a_{100352}\\\\;*\\\\;x_{100352})\\\\;+b_1\\\\\\\\ y_2\\\\;&=\\\\;(a_{100353}\\\\;*\\\\;x_1)\\\\;+\\\\;(a_{100354}\\\\;*\\\\;x_2)\\\\;+\\\\;……\\\\;+\\\\;(a_{(2*100352)}\\\\;*\\\\;x_{100352})\\\\;+b_2)\\\\\\\\ &……\\\\\\\\ y_{256}\\\\;&=\\\\;(a_{(255*100353)}\\\\;*\\\\;x_1)\\\\;+\\\\;(a_{((255*100353)+1)}\\\\;*\\\\;x_2)\\\\;+\\\\;……\\\\;+\\\\;(a_{256\\\\;*\\\\;100352}\\\\;*\\\\;x_{100352})\\\\;+\\\\;b_{256} \\\\end{align} <\/p>\n

      <\/span><\/span><\/span>y<\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>y<\/span><\/span>2<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>y<\/span><\/span>256<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span>(<\/span>a<\/span><\/span>1<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span>x<\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span>+<\/span><\/span><\/span>(<\/span>a<\/span><\/span>2<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span>x<\/span><\/span>2<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span>+<\/span><\/span><\/span>……<\/span><\/span><\/span>+<\/span><\/span><\/span>(<\/span>a<\/span><\/span>100352<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span>x<\/span><\/span>100352<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span>+<\/span><\/span>b<\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span>(<\/span>a<\/span><\/span>100353<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span>x<\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span>+<\/span><\/span><\/span>(<\/span>a<\/span><\/span>100354<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span>x<\/span><\/span>2<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span>+<\/span><\/span><\/span>……<\/span><\/span><\/span>+<\/span><\/span><\/span>(<\/span>a<\/span><\/span>(<\/span>2<\/span>\u2217<\/span>100352<\/span>)<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span>x<\/span><\/span>100352<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span>+<\/span><\/span>b<\/span><\/span>2<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span><\/span><\/span>……<\/span><\/span><\/span><\/span><\/span><\/span>=<\/span><\/span><\/span>(<\/span>a<\/span><\/span>(<\/span>255<\/span>\u2217<\/span>100353<\/span>)<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span>x<\/span><\/span>1<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span>+<\/span><\/span><\/span>(<\/span>a<\/span><\/span>((<\/span>255<\/span>\u2217<\/span>100353<\/span>)<\/span>+<\/span>1<\/span>)<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span>x<\/span><\/span>2<\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span>+<\/span><\/span><\/span>……<\/span><\/span><\/span>+<\/span><\/span><\/span>(<\/span>a<\/span><\/span>256<\/span><\/span>\u2217<\/span><\/span>100352<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u2217<\/span><\/span><\/span>x<\/span><\/span>100352<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>)<\/span><\/span><\/span>+<\/span><\/span><\/span>b<\/span><\/span>256<\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>\u200b<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u6211\u4eec\u53d1\u73b0weight\u603b\u5171\u9700\u8981256 * 100352\u4e2a\u53c2\u6570，\u5176\u53c2\u6570\u5927\u5c0f\u4e3a[256,100352]，\u5bf9\u4e8e\u504f\u7f6e\u9879b\u6765\u8bf4，\u5728256\u4e2a\u8f93\u51fa\u8282\u70b9\u4e2d，\u603b\u5171\u6709256\u4e2a\u504f\u7f6e\u9879，\u6240\u4ee5\u8f93\u51fa\u662f[256]<\/p>\n

      \u77e5\u9053\u4e86\u6bcf\u4e00\u5c42\u7684\u8bad\u7ec3\u5c42\u53c2\u6570\u4e4b\u540e，\u60a8\u53ef\u4ee5\u6240\u6709\u8bad\u7ec3\u5c42\u7684\u53c2\u6570\u4e2a\u6570，\u4e0d\u8fc7\u8fd9\u5e76\u4e0d\u9700\u8981\u60a8\u624b\u52a8\u5b8c\u6210，\u4ec5\u9700\u8c03\u7528\u6bcf\u4e00\u4e2aitem.numel()\u5c31\u53ef\u4ee5\u8ba1\u7b97\u6bcf\u4e00\u5c42\u7684\u5143\u7d20\u4e2a\u6570\u3010\u5bf9\u5e94\u53c2\u6570\u4e2a\u6570\u3011，\u901a\u8fc7for\u5faa\u73af\u548csum\u6c42\u548c，\u60a8\u53ef\u4ee5\u8ba1\u7b97\u51fa\u6240\u6709\u7684\u53c2\u6570\u4e2a\u6570<\/p>\n

      total_parament =<\/span> sum<\/span>(<\/span>p.<\/span>numel(<\/span>)<\/span> for<\/span> p in<\/span> model.<\/span>parameters(<\/span>)<\/span>)<\/span>
      \nprint<\/span>(<\/span>total_parament)<\/span><\/p>\n

      25809830<\/p>\n

      \u5982\u679c\u60a8\u9700\u8981\u6a21\u578b\u8f93\u51fa\u66f4\u590d\u6742\u7684\u4fe1\u606f，\u6216\u8005\u5c06\u6240\u6709\u5d4c\u5957\u7ed3\u6784\u5c55\u793a\u51fa\u6765，\u60a8\u53ef\u4ee5\u4f7f\u7528model.named_children()\u548cmodel.named_modules()，\u5b83\u4eec\u5c06\u5c55\u793a\u66f4\u8be6\u7ec6\u7684\u4fe1\u606f,<\/p>\n

      for<\/span> name,<\/span>module in<\/span> model.<\/span>named_chidren(<\/span>)<\/span>:<\/span>
      \n print<\/span>(<\/span>name,<\/span>module)<\/span>
      \nfor<\/span> name,<\/span>module in<\/span> model.<\/span>named_modules(<\/span>)<\/span>:<\/span>
      \n if<\/span> name:<\/span>
      \n print<\/span>(<\/span>name)<\/span><\/p>\n

      \u5bf9\u4e8enamed_children\u5b83\u53ea\u663e\u793a\u9876\u5c42\u7684\u6a21\u5757，\u5982\u679c\u6a21\u5757\u4e2d\u5305\u542b\u5176\u4ed6\u5c42，\u60a8\u9700\u8981\u4f7f\u7528named_modules\u83b7\u53d6\u5168\u90e8\u4fe1\u606f<\/p>\n

      \u4ee5\u4e0b\u4ee3\u7801\u7684\u8fd0\u884c\u7ed3\u679c\u4ec5\u4e3a\u6a21\u62df，\u5b83\u4e0e\u4e0a\u8ff0\u4ee3\u7801\u65e0\u5173，\u4ec5\u7528\u4e8e\u63cf\u8ff0\u6b65\u9aa4<\/p>\n

      \u77e5\u9053\u4e86\u8fd9\u4e9b\u53c2\u6570\u4fe1\u606f\u4e4b\u540e，\u6211\u4eec\u53ef\u4ee5\u505a\u4ec0\u4e48\u5462\u3002\u5047\u8bbe\u60a8\u5728\u5904\u7406\u8bad\u7ec3\u96c6\u7684\u65f6\u5019\u4e0d\u5c0f\u5fc3\u8ba9\u4e00\u4e2a\u4e0d\u89c4\u6574\u7684\u56fe\u50cf\u4e58\u865a\u800c\u5165，\u6a21\u578b\u53ef\u80fd\u4f1a\u62a5\u8fd9\u6837\u7684\u9519\u8bef，\u60a8\u5b9a\u4f4d\u5230\u8fd9\u4e2a\u9519\u8bef\u7684\u6765\u6e90\u662f\u6765\u81eacnn_x<\/p>\n

      RuntimeError: mat1 and mat2 shapes cannot be multiplied(32×2048 and 1024×512)<\/p>\n

      \u60a8\u4e0d\u786e\u5b9a\u662f\u54ea\u4e00\u7ec4\u56fe\u7247\u5c3a\u5bf8\u53c2\u6570\u5bfc\u81f4\u7684\u9519\u8bef，\u6700\u76f4\u63a5\u7684\u529e\u6cd5\u662f\u6253\u5370\u51fa\u9519\u62a5\u9519\u8bad\u7ec3\u5c42\u671f\u671b\u7684\u53c2\u6570\u5927\u5c0f，\u5982\u679c\u60a8\u5df2\u786e\u4fdd\u6a21\u578b\u6784\u5efa\u9636\u6bb5\u53c2\u6570\u8bbe\u7f6e\u51c6\u786e\u65e0\u8bef<\/p>\n

      print<\/span>(<\/span>model.<\/span>cnn_x.<\/span>weight.<\/span>shape)<\/span><\/p>\n

      for<\/span> name,<\/span>param in<\/span> model.<\/span>named_parameters(<\/span>)<\/span>:<\/span>
      \nif<\/span> 'cnn_x.weight'<\/span> in<\/span> name:<\/span>
      \nprint<\/span>(<\/span>f"<\/span>{<\/span>name}<\/span><\/span>: <\/span>{<\/span>param.<\/span>shape}<\/span><\/span>"<\/span><\/span>)<\/span><\/p>\n

      torch.<\/span>Size(<\/span>[<\/span>512,<\/span>1024]<\/span>)<\/span><\/p>\n

      \u60a8\u786e\u5b9a\u662f32 x 2048\u8fd9\u7ec4\u53c2\u6570\u5e26\u6765\u7684\u9519\u8bef，\u5b83\u662f\u8bad\u7ec3\u96c6\u4e2d\u7684\u54ea\u4e00\u7ec4\u6570\u636e\u5462，\u662f\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u7684\u54ea\u4e00\u4e2a\u9636\u6bb5\u51fa\u9519\u7684，\u6211\u4eec\u53ef\u4ee5\u76f4\u63a5\u8c03\u8bd5def forward(self,x)\u6839\u636e\u8f93\u51fa\u7ed3\u679c\u6216\u8005\u5199\u4e00\u4e2a\u6761\u4ef6\u5224\u65ad\u8fc7\u6ee4\u9519\u8bef\u9879\u6765\u8fdb\u884c\u4e00\u6b21\u6df1\u5165\u7684\u6392\u67e5<\/p>\n

      def<\/span> forward<\/span>(<\/span>self,<\/span>x)<\/span>:<\/span>
      \n print<\/span>(<\/span>f"Input: <\/span>{<\/span>x.<\/span>shape}<\/span><\/span>"<\/span><\/span>)<\/span>
      \n x =<\/span> self.<\/span>features(<\/span>x)<\/span>
      \n print<\/span>(<\/span>f"After features: <\/span>{<\/span>x.<\/span>shape}<\/span><\/span>"<\/span><\/span>)<\/span>
      \n x =<\/span> x.<\/span>flatten(<\/span>1<\/span>)<\/span>
      \n print<\/span>(<\/span>f"Flattened: <\/span>{<\/span>x.<\/span>shape}<\/span><\/span>"<\/span><\/span>)<\/span>
      \n x =<\/span> self.<\/span>classifier(<\/span>x)<\/span>
      \n return<\/span> x<\/p>\n

      Congratulations!!!<\/h3>\n

      \u901a\u8fc7\u7b2c\u56db\u7ae0\u7684\u5b66\u4e60，\u60a8\u638c\u63e1\u4e86<\/p>\n

        \n
      • CNN\u795e\u7ecf\u7f51\u7edc\u7684\u5de5\u4f5c\u539f\u7406<\/li>\n
      • \u5982\u4f55\u5728pytorch\u4e2d\u5b9e\u73b0CNN\u795e\u7ecf\u7f51\u7edc<\/li>\n
      • \u4e86\u89e3\u52a8\u6001\u8ba1\u7b97\u56fe\u8bbe\u8ba1\u548c\u9759\u6001\u6784\u5efa\u7684\u5de5\u4f5c\u539f\u7406\u548c\u5177\u4f53\u5b9e\u73b0<\/li>\n
      • \u4f7f\u7528\u6a21\u5757\u5316\u67b6\u6784\u907f\u514d\u5197\u4f59\u8bbe\u8ba1，\u63d0\u5347\u6a21\u578b\u7684\u53ef\u7ef4\u62a4\u6027\u548c\u53ef\u6269\u5c55\u6027<\/li>\n
      • \u4e86\u89e3\u68c0\u67e5\u6a21\u578b\u53c2\u6570\u548c\u8c03\u8bd5\u6a21\u578b\u7684\u65b9\u6cd5<\/li>\n<\/ul>\n

        Success<\/h2>\n

        \u60a8\u5df2\u7ecf\u6210\u529f\u5b8c\u6210pytorch\u57fa\u7840\u9636\u6bb5\u7684\u8bfe\u7a0b，\u5728\u8fd9\u4e00\u9636\u6bb5，\u60a8\u6784\u5efa\u4e86\u5feb\u9012\u65f6\u95f4\u9884\u6d4b\u6a21\u578b，MNIST\u624b\u5199\u6570\u5b57\u8bc6\u522b\u6a21\u578b，oxford\u82b1\u5349\u8bc6\u522b\u6a21\u578b，\u638c\u63e1\u4e86pytorch\u7684\u57fa\u672c\u4f7f\u7528\u548c\u6784\u5efa\u539f\u7406，\u65e0\u8bba\u63a5\u4e0b\u6765\u60a8\u6253\u7b97\u505a\u4ec0\u4e48，\u795d\u60a8\u987a\u5229😀<\/p>\n","protected":false},"excerpt":{"rendered":"

        \u7b2c\u4e09\u7ae0 \u60a8\u53ef\u4ee5\u5728\u6b64\u5904\u67e5\u770b\u524d\u4e8c\u7ae0\u7684\u5185\u5bb9 \u6570\u636e\u7ba1\u9053 \u6211\u4eec\u4f7f\u7528MNIST\u7684\u6570\u636e\u96c6，\u8fd9\u662f\u6807\u51c6\u7684\u6df1\u5ea6\u5b66\u4e60\u5165\u95e8\u65b9\u5f0f\u3002\u53ef\u662f\u5982\u679c\u6211\u4eec\u5e0c\u671b\u6784\u5efa\u7279\u5b9a\u529f\u80fd\u7684\u6a21\u578b，\u6bd4\u5982\u4e0d\u53ea\u662f\u8bc6\u522b\u6570\u5b57，\u800c\u662f\u8bc6\u522b\u82b1\u5349\u3001\u8bc6\u522b\u884c\u4eba\u3002\u6211\u4eec\u6216\u8bb8\u5c31\u8981\u7528\u5230\u81ea\u5df1\u51c6\u5907\u7684\u8bad\u7ec3\u96c6\u4e86，\u8fd9\u4e2a\u65f6\u5019，\u8bad\u7ec3\u96c6\u53ef\u80fd\u5e76\u4e0d\u6807\u51c6，\u6216\u8bb8\u5305\u542b\u7a7a\u503c\u6216\u8005\u6ca1\u6709\u6ca1\u6709\u610f\u4e49\u7684\u503c，\u800c\u8fd9\u4e9b\u503c\u4e00\u65e6\u4ea4\u7ed9\u6a21\u578b\u8bad\u7ec3，\u5c31\u4f1a\u5bfc\u81f4\u6a21\u578b\u65e0\u6cd5\u8bfb\u53d6\u6709\u6548\u6570\u636e\u800c\u9020\u6210\u6700\u7ec8\u7684\u8bad\u7ec3\u6548\u679c\u4e0d\u5c3d\u4eba\u610f\u7684<\/p>\n","protected":false},"author":2,"featured_media":85447,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[152,371,305],"topic":[],"class_list":["post-85453","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-pytorch","tag-371","tag-305"],"yoast_head":"\npytorch\u3010\u7b2c\u4e8c\u90e8\u5206\u3011 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/85453.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"pytorch\u3010\u7b2c\u4e8c\u90e8\u5206\u3011 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u7b2c\u4e09\u7ae0 \u60a8\u53ef\u4ee5\u5728\u6b64\u5904\u67e5\u770b\u524d\u4e8c\u7ae0\u7684\u5185\u5bb9 \u6570\u636e\u7ba1\u9053 \u6211\u4eec\u4f7f\u7528MNIST\u7684\u6570\u636e\u96c6，\u8fd9\u662f\u6807\u51c6\u7684\u6df1\u5ea6\u5b66\u4e60\u5165\u95e8\u65b9\u5f0f\u3002\u53ef\u662f\u5982\u679c\u6211\u4eec\u5e0c\u671b\u6784\u5efa\u7279\u5b9a\u529f\u80fd\u7684\u6a21\u578b，\u6bd4\u5982\u4e0d\u53ea\u662f\u8bc6\u522b\u6570\u5b57，\u800c\u662f\u8bc6\u522b\u82b1\u5349\u3001\u8bc6\u522b\u884c\u4eba\u3002\u6211\u4eec\u6216\u8bb8\u5c31\u8981\u7528\u5230\u81ea\u5df1\u51c6\u5907\u7684\u8bad\u7ec3\u96c6\u4e86，\u8fd9\u4e2a\u65f6\u5019，\u8bad\u7ec3\u96c6\u53ef\u80fd\u5e76\u4e0d\u6807\u51c6，\u6216\u8bb8\u5305\u542b\u7a7a\u503c\u6216\u8005\u6ca1\u6709\u6ca1\u6709\u610f\u4e49\u7684\u503c，\u800c\u8fd9\u4e9b\u503c\u4e00\u65e6\u4ea4\u7ed9\u6a21\u578b\u8bad\u7ec3，\u5c31\u4f1a\u5bfc\u81f4\u6a21\u578b\u65e0\u6cd5\u8bfb\u53d6\u6709\u6548\u6570\u636e\u800c\u9020\u6210\u6700\u7ec8\u7684\u8bad\u7ec3\u6548\u679c\u4e0d\u5c3d\u4eba\u610f\u7684\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/85453.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-07-27T07:41:17+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/07\/20260727074113-6a670b99d0d45.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"45 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/85453.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/85453.html\",\"name\":\"pytorch\u3010\u7b2c\u4e8c\u90e8\u5206\u3011 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-07-27T07:41:17+00:00\",\"dateModified\":\"2026-07-27T07:41:17+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/85453.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/85453.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/85453.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"pytorch\u3010\u7b2c\u4e8c\u90e8\u5206\u3011\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"pytorch\u3010\u7b2c\u4e8c\u90e8\u5206\u3011 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/85453.html","og_locale":"zh_CN","og_type":"article","og_title":"pytorch\u3010\u7b2c\u4e8c\u90e8\u5206\u3011 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u7b2c\u4e09\u7ae0 \u60a8\u53ef\u4ee5\u5728\u6b64\u5904\u67e5\u770b\u524d\u4e8c\u7ae0\u7684\u5185\u5bb9 \u6570\u636e\u7ba1\u9053 \u6211\u4eec\u4f7f\u7528MNIST\u7684\u6570\u636e\u96c6，\u8fd9\u662f\u6807\u51c6\u7684\u6df1\u5ea6\u5b66\u4e60\u5165\u95e8\u65b9\u5f0f\u3002\u53ef\u662f\u5982\u679c\u6211\u4eec\u5e0c\u671b\u6784\u5efa\u7279\u5b9a\u529f\u80fd\u7684\u6a21\u578b，\u6bd4\u5982\u4e0d\u53ea\u662f\u8bc6\u522b\u6570\u5b57，\u800c\u662f\u8bc6\u522b\u82b1\u5349\u3001\u8bc6\u522b\u884c\u4eba\u3002\u6211\u4eec\u6216\u8bb8\u5c31\u8981\u7528\u5230\u81ea\u5df1\u51c6\u5907\u7684\u8bad\u7ec3\u96c6\u4e86，\u8fd9\u4e2a\u65f6\u5019，\u8bad\u7ec3\u96c6\u53ef\u80fd\u5e76\u4e0d\u6807\u51c6，\u6216\u8bb8\u5305\u542b\u7a7a\u503c\u6216\u8005\u6ca1\u6709\u6ca1\u6709\u610f\u4e49\u7684\u503c，\u800c\u8fd9\u4e9b\u503c\u4e00\u65e6\u4ea4\u7ed9\u6a21\u578b\u8bad\u7ec3，\u5c31\u4f1a\u5bfc\u81f4\u6a21\u578b\u65e0\u6cd5\u8bfb\u53d6\u6709\u6548\u6570\u636e\u800c\u9020\u6210\u6700\u7ec8\u7684\u8bad\u7ec3\u6548\u679c\u4e0d\u5c3d\u4eba\u610f\u7684","og_url":"https:\/\/www.wsisp.com\/helps\/85453.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-07-27T07:41:17+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/07\/20260727074113-6a670b99d0d45.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"45 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/85453.html","url":"https:\/\/www.wsisp.com\/helps\/85453.html","name":"pytorch\u3010\u7b2c\u4e8c\u90e8\u5206\u3011 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-07-27T07:41:17+00:00","dateModified":"2026-07-27T07:41:17+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/85453.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/85453.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/85453.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"pytorch\u3010\u7b2c\u4e8c\u90e8\u5206\u3011"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/85453","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=85453"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/85453\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/85447"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=85453"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=85453"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=85453"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=85453"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}