{"id":103329,"date":"2026-09-10T09:41:08","date_gmt":"2026-09-10T01:41:08","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/103329.html"},"modified":"2026-09-10T09:41:08","modified_gmt":"2026-09-10T01:41:08","slug":"%e5%9b%be%e5%83%8f%e9%a2%86%e5%9f%9f%e7%9a%84%e9%a2%84%e8%ae%ad%e7%bb%83","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/103329.html","title":{"rendered":"\u56fe\u50cf\u9886\u57df\u7684\u9884\u8bad\u7ec3"},"content":{"rendered":"<h3>\u4e00\u3001\u4e3a\u4ec0\u4e48\u56fe\u50cf\u9886\u57df\u9700\u8981\u9884\u8bad\u7ec3&#xff1f;<\/h3>\n<h4>1.1 \u6570\u636e\u56f0\u5883&#xff1a;\u6807\u6ce8\u6210\u672c\u9ad8\u6602<\/h4>\n<ul>\n<li>\u4e00\u5f20\u9ad8\u6e05\u56fe\u50cf\u7684\u50cf\u7d20\u6570\u901a\u5e38\u4e3a <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">10610^6<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8141em\"><\/span><span class=\"mord\">1<\/span><span class=\"mord\"><span class=\"mord\">0<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8141em\"><span class=\"\" style=\"top: -3.063em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">6<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span> \u7ea7\u522b&#xff08;\u5982 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">224\u00d7224\u00d73&#061;150,528224 \\\\times 224 \\\\times 3 &#061; 150,528<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.7278em;vertical-align: -0.0833em\"><\/span><span class=\"mord\">224<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u00d7<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.7278em;vertical-align: -0.0833em\"><\/span><span class=\"mord\">224<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u00d7<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.6444em\"><\/span><span class=\"mord\">3<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8389em;vertical-align: -0.1944em\"><\/span><span class=\"mord\">150<\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\">528<\/span><\/span><\/span><\/span><\/span> \u50cf\u7d20&#xff09;<\/li>\n<li>\u9ad8\u8d28\u91cf\u6807\u6ce8\u9700\u8981\u4e13\u4e1a\u9886\u57df\u77e5\u8bc6&#xff1a;\n<ul>\n<li>\u533b\u5b66\u5f71\u50cf&#xff1a;\u9700\u6267\u4e1a\u533b\u5e08\u6807\u6ce8\u75c5\u7076\u533a\u57df<\/li>\n<li>\u9065\u611f\u56fe\u50cf&#xff1a;\u9700\u5730\u7406\u4fe1\u606f\u4e13\u5bb6\u8bc6\u522b\u5730\u7269\u7c7b\u578b<\/li>\n<li>\u5de5\u4e1a\u68c0\u6d4b&#xff1a;\u9700\u5de5\u7a0b\u5e08\u5b9a\u4e49\u7f3a\u9677\u6807\u51c6<\/li>\n<\/ul>\n<\/li>\n<li>\u9884\u8bad\u7ec3\u7684\u6838\u5fc3\u601d\u60f3&#xff1a;\u5148\u5728\u5927\u89c4\u6a21\u65e0\u6807\u6ce8\/\u5f31\u6807\u6ce8\u6570\u636e\u4e0a\u5b66\u4e60\u901a\u7528\u89c6\u89c9\u8868\u5f81&#xff0c;\u518d\u8fc1\u79fb\u5230\u4e0b\u6e38\u4efb\u52a1<\/li>\n<\/ul>\n<h4>1.2 \u9884\u8bad\u7ec3\u7684\u4ef7\u503c\u5bf9\u6bd4<\/h4>\n<table>\n<tr>\u7ef4\u5ea6\u4ece\u5934\u8bad\u7ec3\u9884\u8bad\u7ec3&#043;\u5fae\u8c03<\/tr>\n<tbody>\n<tr>\n<td>\u6570\u636e\u9700\u6c42<\/td>\n<td>\u767e\u4e07\u7ea7\u6807\u6ce8\u6837\u672c<\/td>\n<td>\u5343\u7ea7\u6807\u6ce8\u5373\u53ef<\/td>\n<\/tr>\n<tr>\n<td>\u8bad\u7ec3\u65f6\u95f4<\/td>\n<td>\u6570\u5468&#xff08;\u591aGPU&#xff09;<\/td>\n<td>\u6570\u5c0f\u65f6&#xff08;\u5355\u5361&#xff09;<\/td>\n<\/tr>\n<tr>\n<td>\u6536\u655b\u7a33\u5b9a\u6027<\/td>\n<td>\u9ad8\u5ea6\u4f9d\u8d56\u521d\u59cb\u5316<\/td>\n<td>\u66f4\u7a33\u5065&#xff0c;\u6613\u6536\u655b<\/td>\n<\/tr>\n<tr>\n<td>\u6cdb\u5316\u6027\u80fd<\/td>\n<td>\u6613\u8fc7\u62df\u5408&#xff0c;\u57df\u8fc1\u79fb\u5dee<\/td>\n<td>\u663e\u8457\u63d0\u5347&#xff0c;\u8de8\u57df\u9c81\u68d2<\/td>\n<\/tr>\n<tr>\n<td>\u8ba1\u7b97\u6210\u672c<\/td>\n<td>\u9ad8&#xff08;\u91cd\u590d\u9020\u8f6e\u5b50&#xff09;<\/td>\n<td>\u4f4e&#xff08;\u4e00\u6b21\u9884\u8bad\u7ec3&#xff0c;\u591a\u6b21\u590d\u7528&#xff09;<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h4>1.3 \u9884\u8bad\u7ec3\u7684\u751f\u7269\u5b66\u542f\u53d1<\/h4>\n<p>\u4eba\u7c7b\u89c6\u89c9\u7cfb\u7edf\u7684\u53d1\u80b2\u8fc7\u7a0b&#xff1a;<\/p>\n<ul>\n<li>\u5a74\u513f\u671f&#xff1a;\u5927\u91cf\u89c2\u5bdf\u73af\u5883&#xff0c;\u5b66\u4e60\u8fb9\u7f18\u3001\u989c\u8272\u3001\u7eb9\u7406\u7b49\u4f4e\u7ea7\u7279\u5f81<\/li>\n<li>\u6210\u957f\u671f&#xff1a;\u901a\u8fc7\u5c11\u91cf\u6307\u5bfc&#xff0c;\u5feb\u901f\u8bc6\u522b\u7269\u4f53\u7c7b\u522b\u548c\u5173\u7cfb<\/li>\n<li>\u4e13\u5bb6\u671f&#xff1a;\u7279\u5b9a\u9886\u57df\u6df1\u5ea6\u4f18\u5316<\/li>\n<\/ul>\n<p>\u9884\u8bad\u7ec3\u6a21\u62df\u4e86\u8fd9\u4e00\u8fc7\u7a0b&#xff1a;\u65e0\u76d1\u7763\/\u5f31\u76d1\u7763\u9636\u6bb5\u5bf9\u5e94\u5a74\u513f\u671f\u89c2\u5bdf&#xff0c;\u5fae\u8c03\u9636\u6bb5\u5bf9\u5e94\u6210\u957f\u671f\u6307\u5bfc\u3002<\/p>\n<hr \/>\n<h3>\u4e8c\u3001\u9884\u8bad\u7ec3\u7684\u6280\u672f\u6f14\u8fdb\u8def\u7ebf<\/h3>\n<h4>2.1 \u7b2c\u4e00\u4ee3&#xff1a;\u76d1\u7763\u9884\u8bad\u7ec3&#xff08;2012-2018&#xff09;<\/h4>\n<p>ImageNet \u7684\u8bde\u751f\u4e0e\u5f71\u54cd<\/p>\n<ul>\n<li>2009\u5e74&#xff0c;\u674e\u98de\u98de\u56e2\u961f\u53d1\u5e03 ImageNet&#xff0c;\u5305\u542b 1280 \u4e07\u5f20\u56fe\u50cf&#xff0c;2.2 \u4e07\u4e2a\u7c7b\u522b<\/li>\n<li>2012\u5e74&#xff0c;AlexNet \u5728 ImageNet \u7ade\u8d5b\u4e2d\u4ee5\u663e\u8457\u4f18\u52bf\u593a\u51a0&#xff0c;\u5f00\u542f\u6df1\u5ea6\u5b66\u4e60\u9769\u547d<\/li>\n<\/ul>\n<p>\u4ee3\u8868\u6027\u7f51\u7edc\u67b6\u6784\u6f14\u8fdb&#xff1a;<\/p>\n<p>AlexNet (2012) \u2192 VGGNet (2014) \u2192 ResNet (2015) \u2192 DenseNet (2017) \u2192 EfficientNet (2019)<br \/>\n     \u2191              \u2191                \u2191                \u2191                  \u2191<br \/>\n   8\u5c42           16-19\u5c42          152&#043;\u5c42           \u7279\u5f81\u590d\u7528            \u590d\u5408\u7f29\u653e<\/p>\n<p>\u6838\u5fc3\u8303\u5f0f\u2014\u2014\u76d1\u7763\u5206\u7c7b&#xff1a;<br \/>\n<span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\">LCE&#061;\u2212\u2211i&#061;1N\u2211c&#061;1Cyi,clog\u2061(y^i,c)<br \/>\n\\\\mathcal{L}_{\\\\text{CE}} &#061; -\\\\sum_{i&#061;1}^{N} \\\\sum_{c&#061;1}^{C} y_{i,c} \\\\log(\\\\hat{y}_{i,c})<br \/>\n<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8333em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathcal\">L<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3283em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord text mtight\"><span class=\"mord mtight\">CE<\/span><\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 3.106em;vertical-align: -1.2777em\"><\/span><span class=\"mord\">\u2212<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mop op-limits\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.8283em\"><span class=\"\" style=\"top: -1.8723em;margin-left: 0em\"><span class=\"pstrut\" style=\"height: 3.05em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">i<\/span><span class=\"mrel mtight\">&#061;<\/span><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -3.05em\"><span class=\"pstrut\" style=\"height: 3.05em\"><\/span><span class=\"\"><span class=\"mop op-symbol large-op\">\u2211<\/span><\/span><\/span><span class=\"\" style=\"top: -4.3em;margin-left: 0em\"><span class=\"pstrut\" style=\"height: 3.05em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.109em\">N<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.2777em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mop op-limits\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.8283em\"><span class=\"\" style=\"top: -1.8829em;margin-left: 0em\"><span class=\"pstrut\" style=\"height: 3.05em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">c<\/span><span class=\"mrel mtight\">&#061;<\/span><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -3.05em\"><span class=\"pstrut\" style=\"height: 3.05em\"><\/span><span class=\"\"><span class=\"mop op-symbol large-op\">\u2211<\/span><\/span><\/span><span class=\"\" style=\"top: -4.3em;margin-left: 0em\"><span class=\"pstrut\" style=\"height: 3.05em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0715em\">C<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.2671em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0359em\">y<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0359em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">i<\/span><span class=\"mpunct mtight\">,<\/span><span class=\"mord mathnormal mtight\">c<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2861em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mop\">lo<span style=\"margin-right: 0.0139em\">g<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord accent\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.6944em\"><span class=\"\" style=\"top: -3em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0359em\">y<\/span><\/span><span class=\"\" style=\"top: -3em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"accent-body\" style=\"left: -0.1944em\"><span class=\"mord\">^<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.1944em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0359em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">i<\/span><span class=\"mpunct mtight\">,<\/span><span class=\"mord mathnormal mtight\">c<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2861em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p>\u8fc1\u79fb\u5b66\u4e60\u7684\u53d1\u73b0&#xff1a;<\/p>\n<ul>\n<li>\u6d45\u5c42\u5377\u79ef&#xff08;\u7b2c1-3\u5c42&#xff09;&#xff1a;\u5b66\u4e60\u8fb9\u7f18\u3001\u989c\u8272\u3001\u7eb9\u7406\u7b49\u901a\u7528\u4f4e\u7ea7\u7279\u5f81<\/li>\n<li>\u4e2d\u5c42\u5377\u79ef&#xff08;\u7b2c4-5\u5c42&#xff09;&#xff1a;\u5b66\u4e60\u90e8\u4ef6\u3001\u8f6e\u5ed3\u3001\u7eb9\u7406\u7ec4\u5408\u7b49\u4e2d\u7ea7\u7279\u5f81<\/li>\n<li>\u6df1\u5c42\u5377\u79ef&#xff08;\u7b2c6\u5c42&#043;&#xff09;&#xff1a;\u5b66\u4e60\u7269\u4f53\u90e8\u4ef6\u3001\u7c7b\u522b\u7279\u5f81\u7b49\u9ad8\u7ea7\u8bed\u4e49\u7279\u5f81<\/li>\n<\/ul>\n<h4>2.2 \u7b2c\u4e8c\u4ee3&#xff1a;\u81ea\u76d1\u7763\u9884\u8bad\u7ec3&#xff08;2019-2022&#xff09;<\/h4>\n<p>\u52a8\u673a\u4e0e\u80cc\u666f&#xff1a;<\/p>\n<ul>\n<li>ImageNet \u6807\u6ce8\u6210\u672c&#xff1a;\u6bcf\u5f20\u56fe\u50cf\u7ea6 $0.5-2 \u7f8e\u5143&#xff0c;\u603b\u6210\u672c\u8d85 2500 \u4e07\u7f8e\u5143<\/li>\n<li>\u4e92\u8054\u7f51\u56fe\u50cf&#xff1a;\u6570\u5341\u4ebf\u5f20\u65e0\u6807\u6ce8\u56fe\u50cf\u7b49\u5f85\u5229\u7528<\/li>\n<li>\u6838\u5fc3\u95ee\u9898&#xff1a;\u5982\u4f55\u8bbe\u8ba1\u4ee3\u7406\u4efb\u52a1&#xff0c;\u8ba9\u6a21\u578b\u4ece\u56fe\u50cf\u81ea\u8eab\u7ed3\u6784\u4e2d\u5b66\u4e60&#xff1f;<\/li>\n<\/ul>\n<h5>2.2.1 \u5bf9\u6bd4\u5b66\u4e60\u5bb6\u65cf<\/h5>\n<table>\n<tr>\u65b9\u6cd5\u6838\u5fc3\u521b\u65b0\u6280\u672f\u7ec6\u8282\u6027\u80fd<\/tr>\n<tbody>\n<tr>\n<td>SimCLR (2020)<\/td>\n<td>\u7b80\u5355\u5bf9\u6bd4\u5b66\u4e60<\/td>\n<td>\u540c\u56fe\u50cf\u7684\u4e0d\u540c\u589e\u5f3a\u89c6\u56fe\u4e3a\u6b63\u6837\u672c&#xff0c;\u5176\u4ed6\u4e3a\u8d1f\u6837\u672c&#xff1b;\u5927batch&#xff08;4096&#xff09;&#043; \u6e29\u5ea6\u7cfb\u6570<\/td>\n<td>ResNet-50 \u8fbe\u5230 76.5% Top-1<\/td>\n<\/tr>\n<tr>\n<td>MoCo v1\/v2\/v3 (2019-2021)<\/td>\n<td>\u52a8\u91cf\u7f16\u7801\u5668<\/td>\n<td>\u7ef4\u62a4\u52a8\u6001\u5b57\u5178&#xff0c;\u52a8\u91cf\u66f4\u65b0\u7f16\u7801\u5668&#xff0c;\u65e0\u9700\u5927batch<\/td>\n<td>\u5c0fbatch\u4e5f\u80fd\u5de5\u4f5c&#xff0c;\u66f4\u5b9e\u7528<\/td>\n<\/tr>\n<tr>\n<td>SwAV (2020)<\/td>\n<td>\u805a\u7c7b\u5bf9\u6bd4<\/td>\n<td>\u5728\u7ebf\u805a\u7c7b\u5206\u914d\u4f2a\u6807\u7b7e&#xff0c;\u907f\u514d\u663e\u5f0f\u8d1f\u6837\u672c<\/td>\n<td>\u66f4\u7a33\u5b9a\u7684\u8bad\u7ec3<\/td>\n<\/tr>\n<tr>\n<td>BYOL\/SimSiam (2020-2021)<\/td>\n<td>\u65e0\u8d1f\u6837\u672c<\/td>\n<td>\u5728\u7ebf\u7f51\u7edc&#043;\u52a8\u91cf\u7f51\u7edc\/\u5b6a\u751f\u7f51\u7edc&#xff0c;\u505c\u6b62\u68af\u5ea6\u4f20\u64ad<\/td>\n<td>\u63ed\u793a\u5bf9\u6bd4\u5b66\u4e60\u4e0d\u4e00\u5b9a\u9700\u8981\u8d1f\u6837\u672c<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>SimCLR \u7684\u635f\u5931\u51fd\u6570&#xff08;InfoNCE&#xff09;&#xff1a;<br \/>\n<span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\">LSimCLR&#061;\u2212log\u2061exp\u2061(sim(zi,zj)\/\u03c4)\u2211kexp\u2061(sim(zi,zk)\/\u03c4)<br \/>\n\\\\mathcal{L}_{\\\\text{SimCLR}} &#061; -\\\\log \\\\frac{\\\\exp(\\\\text{sim}(z_i, z_j)\/\\\\tau)}{\\\\sum_{k} \\\\exp(\\\\text{sim}(z_i, z_k)\/\\\\tau)}<br \/>\n<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8333em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathcal\">L<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3283em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord text mtight\"><span class=\"mord mtight\">SimCLR<\/span><\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 2.4127em;vertical-align: -0.9857em\"><\/span><span class=\"mord\">\u2212<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mop\">lo<span style=\"margin-right: 0.0139em\">g<\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mopen nulldelimiter\"><\/span><span class=\"mfrac\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.427em\"><span class=\"\" style=\"top: -2.314em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mop\"><span class=\"mop op-symbol small-op\" style=\"position: relative;top: 0em\">\u2211<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.1864em\"><span class=\"\" style=\"top: -2.4003em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0315em\">k<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2997em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mop\">exp<\/span><span class=\"mopen\">(<\/span><span class=\"mord text\"><span class=\"mord\">sim<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0315em\">k<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><span class=\"mord\">\/<\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1132em\">\u03c4<\/span><span class=\"mclose\">)<\/span><\/span><\/span><span class=\"\" style=\"top: -3.23em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"frac-line\" style=\"border-bottom-width: 0.04em\"><\/span><\/span><span class=\"\" style=\"top: -3.677em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mop\">exp<\/span><span class=\"mopen\">(<\/span><span class=\"mord text\"><span class=\"mord\">sim<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0572em\">j<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2861em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><span class=\"mord\">\/<\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1132em\">\u03c4<\/span><span class=\"mclose\">)<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.9857em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mclose nulldelimiter\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><br \/>\n\u5176\u4e2d <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">sim(u,v)&#061;u\u22a4v\u2225u\u2225\u2225v\u2225\\\\text{sim}(u, v) &#061; \\\\frac{u^\\\\top v}{\\\\|u\\\\| \\\\|v\\\\|}<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 1em;vertical-align: -0.25em\"><\/span><span class=\"mord text\"><span class=\"mord\">sim<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord mathnormal\">u<\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0359em\">v<\/span><span class=\"mclose\">)<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1.5629em;vertical-align: -0.52em\"><\/span><span class=\"mord\"><span class=\"mopen nulldelimiter\"><\/span><span class=\"mfrac\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.0429em\"><span class=\"\" style=\"top: -2.655em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mtight\">\u2225<\/span><span class=\"mord mathnormal mtight\">u<\/span><span class=\"mord mtight\">\u2225\u2225<\/span><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0359em\">v<\/span><span class=\"mord mtight\">\u2225<\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -3.23em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"frac-line\" style=\"border-bottom-width: 0.04em\"><\/span><\/span><span class=\"\" style=\"top: -3.394em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">u<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.927em\"><span class=\"\" style=\"top: -2.931em;margin-right: 0.0714em\"><span class=\"pstrut\" style=\"height: 2.5em\"><\/span><span class=\"sizing reset-size3 size1 mtight\"><span class=\"mord mtight\">\u22a4<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0359em\">v<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.52em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mclose nulldelimiter\"><\/span><\/span><\/span><\/span><\/span><\/span> \u4e3a\u4f59\u5f26\u76f8\u4f3c\u5ea6&#xff0c;<span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">\u03c4\\\\tau<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.4306em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1132em\">\u03c4<\/span><\/span><\/span><\/span><\/span> \u4e3a\u6e29\u5ea6\u7cfb\u6570\u3002<\/p>\n<h5>2.2.2 \u63a9\u7801\u56fe\u50cf\u5efa\u6a21&#xff08;MIM&#xff09;<\/h5>\n<p>MAE&#xff08;Masked Autoencoder&#xff0c;2021&#xff09;&#xff1a;<\/p>\n<ul>\n<li>\u968f\u673a\u63a9\u7801 75% \u7684\u56fe\u50cf\u5757&#xff08;patches&#xff09;<\/li>\n<li>\u7528\u975e\u5bf9\u79f0\u7684\u7f16\u7801\u5668-\u89e3\u7801\u5668\u67b6\u6784\u91cd\u5efa\u539f\u56fe<\/li>\n<li>\u7f16\u7801\u5668\u4ec5\u5904\u7406\u53ef\u89c1 patches&#xff0c;\u8ba1\u7b97\u9ad8\u6548<\/li>\n<\/ul>\n<p>MAE \u7684\u6570\u5b66\u5f62\u5f0f&#xff1a;<br \/>\n<span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\">LMAE&#061;1\u2223M\u2223\u2211i\u2208M\u2225xi\u2212x^i\u22252<br \/>\n\\\\mathcal{L}_{\\\\text{MAE}} &#061; \\\\frac{1}{|\\\\mathcal{M}|} \\\\sum_{i \\\\in \\\\mathcal{M}} \\\\|x_i &#8211; \\\\hat{x}_i\\\\|^2<br \/>\n<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8333em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathcal\">L<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3283em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord text mtight\"><span class=\"mord mtight\">MAE<\/span><\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 2.6431em;vertical-align: -1.3217em\"><\/span><span class=\"mord\"><span class=\"mopen nulldelimiter\"><\/span><span class=\"mfrac\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.3214em\"><span class=\"\" style=\"top: -2.314em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\">\u2223<\/span><span class=\"mord mathcal\">M<\/span><span class=\"mord\">\u2223<\/span><\/span><\/span><span class=\"\" style=\"top: -3.23em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"frac-line\" style=\"border-bottom-width: 0.04em\"><\/span><\/span><span class=\"\" style=\"top: -3.677em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\">1<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.936em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mclose nulldelimiter\"><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mop op-limits\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.05em\"><span class=\"\" style=\"top: -1.8557em;margin-left: 0em\"><span class=\"pstrut\" style=\"height: 3.05em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">i<\/span><span class=\"mrel mtight\">\u2208<\/span><span class=\"mord mathcal mtight\">M<\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -3.05em\"><span class=\"pstrut\" style=\"height: 3.05em\"><\/span><span class=\"\"><span class=\"mop op-symbol large-op\">\u2211<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.3217em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\">\u2225<\/span><span class=\"mord\"><span class=\"mord mathnormal\">x<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u2212<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1.1141em;vertical-align: -0.25em\"><\/span><span class=\"mord\"><span class=\"mord accent\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.6944em\"><span class=\"\" style=\"top: -3em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord mathnormal\">x<\/span><\/span><span class=\"\" style=\"top: -3em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"accent-body\" style=\"left: -0.2222em\"><span class=\"mord\">^<\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mord\"><span class=\"mord\">\u2225<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8641em\"><span class=\"\" style=\"top: -3.113em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><br \/>\n\u5176\u4e2d <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">M\\\\mathcal{M}<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.6833em\"><\/span><span class=\"mord mathcal\">M<\/span><\/span><\/span><\/span><\/span> \u4e3a\u63a9\u7801\u4f4d\u7f6e\u96c6\u5408&#xff0c;<span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">x^i&#061;D(E(xvisible)&#043;mask\u00a0tokens)\\\\hat{x}_i &#061; D(E(x_{\\\\text{visible}}) &#043; \\\\text{mask tokens})<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8444em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord accent\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.6944em\"><span class=\"\" style=\"top: -3em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord mathnormal\">x<\/span><\/span><span class=\"\" style=\"top: -3em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"accent-body\" style=\"left: -0.2222em\"><span class=\"mord\">^<\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1em;vertical-align: -0.25em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0278em\">D<\/span><span class=\"mopen\">(<\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0576em\">E<\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\">x<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord text mtight\"><span class=\"mord mtight\">visible<\/span><\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">&#043;<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1em;vertical-align: -0.25em\"><\/span><span class=\"mord text\"><span class=\"mord\">mask\u00a0tokens<\/span><\/span><span class=\"mclose\">)<\/span><\/span><\/span><\/span><\/span>\u3002<\/p>\n<p>MAE \u7684\u4f18\u52bf&#xff1a;<\/p>\n<ul>\n<li>\u63a9\u7801\u6bd4\u4f8b\u9ad8\u8fbe 75% \u4ecd\u80fd\u91cd\u5efa&#xff0c;\u8bf4\u660e\u6a21\u578b\u5b66\u5230\u4e86\u5f3a\u8bed\u4e49\u8868\u5f81<\/li>\n<li>\u4e0b\u6e38\u5fae\u8c03\u65f6&#xff0c;\u4ec5\u7528 1% \u7684 ImageNet \u6807\u7b7e\u5373\u53ef\u8fbe\u5230 80%&#043; \u51c6\u786e\u7387<\/li>\n<\/ul>\n<p>BEiT&#xff08;2021&#xff09;\u4e0e\u540e\u7eed\u6539\u8fdb&#xff1a;<\/p>\n<ul>\n<li>\u4f7f\u7528\u79bb\u6563 VAE \u5c06\u56fe\u50cf token \u5316&#xff0c;\u9884\u6d4b\u79bb\u6563\u89c6\u89c9 token<\/li>\n<li>\u7ed3\u5408 NLP \u7684 BERT \u601d\u60f3\u5230\u89c6\u89c9\u9886\u57df<\/li>\n<\/ul>\n<h4>2.3 \u7b2c\u4e09\u4ee3&#xff1a;\u591a\u6a21\u6001\u9884\u8bad\u7ec3&#xff08;2021-\u81f3\u4eca&#xff09;<\/h4>\n<p>CLIP&#xff1a;\u8fde\u63a5\u89c6\u89c9\u4e0e\u8bed\u8a00\u7684\u6865\u6881&#xff08;2021&#xff09;<\/p>\n<p>\u6838\u5fc3\u521b\u65b0&#xff1a;<\/p>\n<ul>\n<li>4\u4ebf\u4e92\u8054\u7f51\u56fe\u6587\u5bf9&#xff08;WebImageText&#xff09;\u8fdb\u884c\u5bf9\u6bd4\u5b66\u4e60<\/li>\n<li>\u53cc\u5854\u67b6\u6784&#xff1a;\u56fe\u50cf\u7f16\u7801\u5668 &#043; \u6587\u672c\u7f16\u7801\u5668<\/li>\n<li>\u5bf9\u6bd4\u635f\u5931\u5bf9\u9f50\u89c6\u89c9-\u8bed\u4e49\u7a7a\u95f4<\/li>\n<\/ul>\n<p>CLIP \u7684\u5bf9\u6bd4\u635f\u5931&#xff1a;<br \/>\n<span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\">LCLIP&#061;12[\u2211i\u2212log\u2061exp\u2061(sim(Ii,Ti)\/\u03c4)\u2211jexp\u2061(sim(Ii,Tj)\/\u03c4)&#043;\u2211j\u2212log\u2061exp\u2061(sim(Ij,Tj)\/\u03c4)\u2211iexp\u2061(sim(Ii,Tj)\/\u03c4)]<br \/>\n\\\\mathcal{L}_{\\\\text{CLIP}} &#061; \\\\frac{1}{2} \\\\left[ \\\\sum_{i} -\\\\log \\\\frac{\\\\exp(\\\\text{sim}(I_i, T_i)\/\\\\tau)}{\\\\sum_{j} \\\\exp(\\\\text{sim}(I_i, T_j)\/\\\\tau)} &#043; \\\\sum_{j} -\\\\log \\\\frac{\\\\exp(\\\\text{sim}(I_j, T_j)\/\\\\tau)}{\\\\sum_{i} \\\\exp(\\\\text{sim}(I_i, T_j)\/\\\\tau)} \\\\right]<br \/>\n<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8333em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathcal\">L<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3283em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord text mtight\"><span class=\"mord mtight\">CLIP<\/span><\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 3.1638em;vertical-align: -1.4138em\"><\/span><span class=\"mord\"><span class=\"mopen nulldelimiter\"><\/span><span class=\"mfrac\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.3214em\"><span class=\"\" style=\"top: -2.314em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\">2<\/span><\/span><\/span><span class=\"\" style=\"top: -3.23em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"frac-line\" style=\"border-bottom-width: 0.04em\"><\/span><\/span><span class=\"\" style=\"top: -3.677em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\">1<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.686em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mclose nulldelimiter\"><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"minner\"><span class=\"mopen delimcenter\" style=\"top: 0em\"><span class=\"delimsizing size4\">[<\/span><\/span><span class=\"mop op-limits\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.05em\"><span class=\"\" style=\"top: -1.8723em;margin-left: 0em\"><span class=\"pstrut\" style=\"height: 3.05em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -3.05em\"><span class=\"pstrut\" style=\"height: 3.05em\"><\/span><span class=\"\"><span class=\"mop op-symbol large-op\">\u2211<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.2777em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\">\u2212<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mop\">lo<span style=\"margin-right: 0.0139em\">g<\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mopen nulldelimiter\"><\/span><span class=\"mfrac\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.427em\"><span class=\"\" style=\"top: -2.314em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mop\"><span class=\"mop op-symbol small-op\" style=\"position: relative;top: 0em\">\u2211<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.162em\"><span class=\"\" style=\"top: -2.4003em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0572em\">j<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.4358em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mop\">exp<\/span><span class=\"mopen\">(<\/span><span class=\"mord text\"><span class=\"mord\">sim<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0785em\">I<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0785em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">T<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.1389em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0572em\">j<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2861em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><span class=\"mord\">\/<\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1132em\">\u03c4<\/span><span class=\"mclose\">)<\/span><\/span><\/span><span class=\"\" style=\"top: -3.23em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"frac-line\" style=\"border-bottom-width: 0.04em\"><\/span><\/span><span class=\"\" style=\"top: -3.677em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mop\">exp<\/span><span class=\"mopen\">(<\/span><span class=\"mord text\"><span class=\"mord\">sim<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0785em\">I<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0785em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">T<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.1389em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><span class=\"mord\">\/<\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1132em\">\u03c4<\/span><span class=\"mclose\">)<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.1218em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mclose nulldelimiter\"><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">&#043;<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mop op-limits\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.05em\"><span class=\"\" style=\"top: -1.8723em;margin-left: 0em\"><span class=\"pstrut\" style=\"height: 3.05em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0572em\">j<\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -3.05em\"><span class=\"pstrut\" style=\"height: 3.05em\"><\/span><span class=\"\"><span class=\"mop op-symbol large-op\">\u2211<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.4138em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\">\u2212<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mop\">lo<span style=\"margin-right: 0.0139em\">g<\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mopen nulldelimiter\"><\/span><span class=\"mfrac\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 1.427em\"><span class=\"\" style=\"top: -2.314em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mop\"><span class=\"mop op-symbol small-op\" style=\"position: relative;top: 0em\">\u2211<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.162em\"><span class=\"\" style=\"top: -2.4003em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2997em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mop\">exp<\/span><span class=\"mopen\">(<\/span><span class=\"mord text\"><span class=\"mord\">sim<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0785em\">I<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0785em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">i<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">T<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.1389em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0572em\">j<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2861em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><span class=\"mord\">\/<\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1132em\">\u03c4<\/span><span class=\"mclose\">)<\/span><\/span><\/span><span class=\"\" style=\"top: -3.23em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"frac-line\" style=\"border-bottom-width: 0.04em\"><\/span><\/span><span class=\"\" style=\"top: -3.677em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mop\">exp<\/span><span class=\"mopen\">(<\/span><span class=\"mord text\"><span class=\"mord\">sim<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0785em\">I<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0785em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0572em\">j<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2861em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">T<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3117em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.1389em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0572em\">j<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2861em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><span class=\"mord\">\/<\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1132em\">\u03c4<\/span><span class=\"mclose\">)<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.9857em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"mclose nulldelimiter\"><\/span><\/span><span class=\"mclose delimcenter\" style=\"top: 0em\"><span class=\"delimsizing size4\">]<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<p>CLIP \u7684\u96f6\u6837\u672c\u80fd\u529b&#xff1a;<\/p>\n<ul>\n<li>\u7ed9\u5b9a\u7c7b\u522b\u63cf\u8ff0&#xff08;\u5982 \u201ca photo of a dog\u201d&#xff09;&#xff0c;\u8ba1\u7b97\u56fe\u50cf\u4e0e\u63cf\u8ff0\u7684\u76f8\u4f3c\u5ea6<\/li>\n<li>\u5728 ImageNet \u4e0a\u96f6\u6837\u672c\u8fbe\u5230 76.2% Top-1&#xff0c;\u5ab2\u7f8e\u76d1\u7763 ResNet-50<\/li>\n<\/ul>\n<p>\u540e\u7eed\u591a\u6a21\u6001\u5de5\u4f5c&#xff1a;<\/p>\n<table>\n<tr>\u6a21\u578b\u5e74\u4efd\u89c4\u6a21\u7279\u70b9<\/tr>\n<tbody>\n<tr>\n<td>ALIGN<\/td>\n<td>2021<\/td>\n<td>18\u4ebf\u56fe\u6587\u5bf9<\/td>\n<td>\u566a\u58f0\u9c81\u68d2\u7684\u5bf9\u6bd4\u5b66\u4e60<\/td>\n<\/tr>\n<tr>\n<td>Florence<\/td>\n<td>2021<\/td>\n<td>9\u4ebf\u56fe\u6587\u5bf9<\/td>\n<td>\u7edf\u4e00\u89c6\u89c9-\u8bed\u8a00-\u4efb\u52a1<\/td>\n<\/tr>\n<tr>\n<td>BLIP\/BLIP-2<\/td>\n<td>2022-2023<\/td>\n<td>14\u4ebf\u56fe\u6587\u5bf9<\/td>\n<td>\u5f15\u5bfc\u5f0f\u8fc7\u6ee4 &#043; Q-Former<\/td>\n<\/tr>\n<tr>\n<td>LLaVA<\/td>\n<td>2023<\/td>\n<td>\u89c6\u89c9\u6307\u4ee4\u5fae\u8c03<\/td>\n<td>\u8fde\u63a5 CLIP \u4e0e LLM<\/td>\n<\/tr>\n<tr>\n<td>GPT-4V<\/td>\n<td>2023<\/td>\n<td>\u672a\u516c\u5f00<\/td>\n<td>\u7aef\u5230\u7aef\u591a\u6a21\u6001\u5927\u6a21\u578b<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<hr \/>\n<h3>\u4e09\u3001\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u67b6\u6784\u6f14\u8fdb<\/h3>\n<h4>3.1 \u4ece CNN \u5230 Transformer \u7684\u8303\u5f0f\u8f6c\u79fb<\/h4>\n<p>\u5377\u79ef\u795e\u7ecf\u7f51\u7edc&#xff08;CNN&#xff09;\u7684\u5c40\u9650&#xff1a;<\/p>\n<ul>\n<li>\u5c40\u90e8\u611f\u53d7\u91ce&#xff0c;\u96be\u4ee5\u5efa\u6a21\u957f\u8ddd\u79bb\u4f9d\u8d56<\/li>\n<li>\u5e73\u79fb\u7b49\u53d8\u6027\u662f\u4f18\u52bf&#xff0c;\u4f46\u5bf9\u65cb\u8f6c\u3001\u5c3a\u5ea6\u53d8\u5316\u9700\u989d\u5916\u8bbe\u8ba1<\/li>\n<li>\u6df1\u5c42\u7f51\u7edc\u4f18\u5316\u56f0\u96be&#xff08;\u68af\u5ea6\u6d88\u5931\/\u7206\u70b8&#xff09;<\/li>\n<\/ul>\n<p>Vision Transformer&#xff08;ViT&#xff0c;2020&#xff09;\u7684\u7a81\u7834&#xff1a;<\/p>\n<ul>\n<li>\u5c06\u56fe\u50cf\u5207\u5206\u4e3a <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">16\u00d71616 \\\\times 16<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.7278em;vertical-align: -0.0833em\"><\/span><span class=\"mord\">16<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u00d7<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.6444em\"><\/span><span class=\"mord\">16<\/span><\/span><\/span><\/span><\/span> \u7684 patches&#xff0c;\u89c6\u4e3a \u201c\u89c6\u89c9 token\u201d<\/li>\n<li>\u6807\u51c6 Transformer \u7f16\u7801\u5668\u5904\u7406\u5168\u5c40\u5173\u7cfb<\/li>\n<li>\u5728\u5927\u89c4\u6a21\u6570\u636e&#xff08;JFT-300M&#xff09;\u4e0a\u8d85\u8d8a ResNet<\/li>\n<\/ul>\n<p>ViT \u7684\u67b6\u6784\u516c\u5f0f&#xff1a;<br \/>\n<span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\">Patch\u00a0Embedding:z0&#061;[xclass;xp1E;xp2E;\u22ef\u2009;xpNE]&#043;EposTransformer\u00a0Block:zl\u2032&#061;MSA(LN(zl\u22121))&#043;zl\u22121zl&#061;MLP(LN(zl\u2032))&#043;zl\u2032Classification:y&#061;Head(LN(zL0))<br \/>\n\\\\begin{aligned}<br \/>\n&amp;\\\\text{Patch Embedding:} &amp;&amp; z_0 &#061; [x_{\\\\text{class}}; x_p^1 E; x_p^2 E; \\\\cdots; x_p^N E] &#043; E_{\\\\text{pos}} \\\\\\\\<br \/>\n&amp;\\\\text{Transformer Block:} &amp;&amp; z_l&#039; &#061; \\\\text{MSA}(\\\\text{LN}(z_{l-1})) &#043; z_{l-1} \\\\\\\\<br \/>\n&amp; &amp;&amp; z_l &#061; \\\\text{MLP}(\\\\text{LN}(z_l&#039;)) &#043; z_l&#039; \\\\\\\\<br \/>\n&amp;\\\\text{Classification:} &amp;&amp; y &#061; \\\\text{Head}(\\\\text{LN}(z_L^0))<br \/>\n\\\\end{aligned}<br \/>\n<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 6.0985em;vertical-align: -2.7993em\"><\/span><span class=\"mord\"><span class=\"mtable\"><span class=\"col-align-r\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 3.2993em\"><span class=\"\" style=\"top: -5.2993em\"><span class=\"pstrut\" style=\"height: 2.8913em\"><\/span><span class=\"mord\"><\/span><\/span><span class=\"\" style=\"top: -3.7762em\"><span class=\"pstrut\" style=\"height: 2.8913em\"><\/span><span class=\"mord\"><\/span><\/span><span class=\"\" style=\"top: -2.2762em\"><span class=\"pstrut\" style=\"height: 2.8913em\"><\/span><span class=\"mord\"><\/span><\/span><span class=\"\" style=\"top: -0.7521em\"><span class=\"pstrut\" style=\"height: 2.8913em\"><\/span><span class=\"mord\"><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 2.7993em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"col-align-l\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 3.2993em\"><span class=\"\" style=\"top: -5.4079em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\"><\/span><span class=\"mord text\"><span class=\"mord\">Patch\u00a0Embedding:<\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -3.8848em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\"><\/span><span class=\"mord text\"><span class=\"mord\">Transformer\u00a0Block:<\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -2.3848em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\"><\/span><\/span><\/span><span class=\"\" style=\"top: -0.8607em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\"><\/span><span class=\"mord text\"><span class=\"mord\">Classification:<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 2.7993em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"arraycolsep\" style=\"width: 1em\"><\/span><span class=\"col-align-r\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 3.2993em\"><span class=\"\" style=\"top: -5.2993em\"><span class=\"pstrut\" style=\"height: 2.8913em\"><\/span><span class=\"mord\"><\/span><\/span><span class=\"\" style=\"top: -3.7762em\"><span class=\"pstrut\" style=\"height: 2.8913em\"><\/span><span class=\"mord\"><\/span><\/span><span class=\"\" style=\"top: -2.2762em\"><span class=\"pstrut\" style=\"height: 2.8913em\"><\/span><span class=\"mord\"><\/span><\/span><span class=\"\" style=\"top: -0.7521em\"><span class=\"pstrut\" style=\"height: 2.8913em\"><\/span><span class=\"mord\"><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 2.7993em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><span class=\"col-align-l\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 3.2993em\"><span class=\"\" style=\"top: -5.4079em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">0<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mopen\">[<\/span><span class=\"mord\"><span class=\"mord mathnormal\">x<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord text mtight\"><span class=\"mord mtight\">class<\/span><\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mpunct\">;<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">x<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8641em\"><span class=\"\" style=\"top: -2.453em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">p<\/span><\/span><\/span><span class=\"\" style=\"top: -3.113em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3831em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0576em\">E<\/span><span class=\"mpunct\">;<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">x<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8641em\"><span class=\"\" style=\"top: -2.453em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">p<\/span><\/span><\/span><span class=\"\" style=\"top: -3.113em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">2<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3831em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0576em\">E<\/span><span class=\"mpunct\">;<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"minner\">\u22ef<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mpunct\">;<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\">x<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8913em\"><span class=\"\" style=\"top: -2.453em;margin-left: 0em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">p<\/span><\/span><\/span><span class=\"\" style=\"top: -3.113em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.109em\">N<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3831em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0576em\">E<\/span><span class=\"mclose\">]<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">&#043;<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.0576em\">E<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.1514em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.0576em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord text mtight\"><span class=\"mord mtight\">pos<\/span><\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2861em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -3.8848em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8019em\"><span class=\"\" style=\"top: -2.453em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0197em\">l<\/span><\/span><\/span><span class=\"\" style=\"top: -3.113em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mtight\">\u2032<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.247em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mord text\"><span class=\"mord\">MSA<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord text\"><span class=\"mord\">LN<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0197em\">l<\/span><span class=\"mbin mtight\">\u2212<\/span><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2083em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">))<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">&#043;<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0197em\">l<\/span><span class=\"mbin mtight\">\u2212<\/span><span class=\"mord mtight\">1<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.2083em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -2.3848em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3361em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0197em\">l<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mord text\"><span class=\"mord\">MLP<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord text\"><span class=\"mord\">LN<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8019em\"><span class=\"\" style=\"top: -2.453em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0197em\">l<\/span><\/span><\/span><span class=\"\" style=\"top: -3.113em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mtight\">\u2032<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.247em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">))<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">&#043;<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8019em\"><span class=\"\" style=\"top: -2.453em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0197em\">l<\/span><\/span><\/span><span class=\"\" style=\"top: -3.113em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mtight\">\u2032<\/span><\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.247em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"\" style=\"top: -0.8607em\"><span class=\"pstrut\" style=\"height: 3em\"><\/span><span class=\"mord\"><span class=\"mord\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0359em\">y<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mord text\"><span class=\"mord\">Head<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord text\"><span class=\"mord\">LN<\/span><\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.044em\">z<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8641em\"><span class=\"\" style=\"top: -2.453em;margin-left: -0.044em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mathnormal mtight\">L<\/span><\/span><\/span><span class=\"\" style=\"top: -3.113em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">0<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.247em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">))<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 2.7993em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/p>\n<h4>3.2 \u5c42\u6b21\u5316\u8bbe\u8ba1&#xff1a;\u878d\u5408 CNN \u4e0e Transformer \u4f18\u52bf<\/h4>\n<table>\n<tr>\u6a21\u578b\u6838\u5fc3\u8bbe\u8ba1\u4f18\u52bf\u5178\u578b\u5e94\u7528<\/tr>\n<tbody>\n<tr>\n<td>Swin Transformer (2021)<\/td>\n<td>\u79fb\u4f4d\u7a97\u53e3 &#043; \u5c42\u6b21\u5316\u7279\u5f81\u91d1\u5b57\u5854<\/td>\n<td>\u7ebf\u6027\u590d\u6742\u5ea6&#xff0c;\u591a\u5c3a\u5ea6\u7279\u5f81<\/td>\n<td>\u68c0\u6d4b\u3001\u5206\u5272<\/td>\n<\/tr>\n<tr>\n<td>PVT (2021)<\/td>\n<td>\u91d1\u5b57\u5854\u7ed3\u6784 &#043; \u7a7a\u95f4\u7f29\u51cf\u6ce8\u610f\u529b<\/td>\n<td>\u6e10\u8fdb\u5f0f\u4e0b\u91c7\u6837<\/td>\n<td>\u5bc6\u96c6\u9884\u6d4b<\/td>\n<\/tr>\n<tr>\n<td>Twins-SVT (2021)<\/td>\n<td>\u7a7a\u95f4\u53ef\u5206\u79bb\u6ce8\u610f\u529b<\/td>\n<td>\u5168\u5c40-\u5c40\u90e8\u4fe1\u606f\u4ea4\u4e92<\/td>\n<td>\u5206\u7c7b\u3001\u68c0\u6d4b<\/td>\n<\/tr>\n<tr>\n<td>ConvNeXt (2022)<\/td>\n<td>\u73b0\u4ee3\u5316\u7eaf CNN \u8bbe\u8ba1<\/td>\n<td>\u6311\u6218 Transformer \u6027\u80fd<\/td>\n<td>\u5404\u7c7b\u89c6\u89c9\u4efb\u52a1<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Swin Transformer \u7684\u5173\u952e\u521b\u65b0&#xff1a;<\/p>\n<ul>\n<li>\u7a97\u53e3\u81ea\u6ce8\u610f\u529b&#xff1a;\u5728 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">M\u00d7MM \\\\times M<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.7667em;vertical-align: -0.0833em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.109em\">M<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">\u00d7<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.6833em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.109em\">M<\/span><\/span><\/span><\/span><\/span> \u5c40\u90e8\u7a97\u53e3\u5185\u8ba1\u7b97&#xff0c;\u590d\u6742\u5ea6\u4ece <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">O(N2)O(N^2)<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 1.0641em;vertical-align: -0.25em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0278em\">O<\/span><span class=\"mopen\">(<\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.109em\">N<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8141em\"><span class=\"\" style=\"top: -3.063em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">2<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mclose\">)<\/span><\/span><\/span><\/span><\/span> \u964d\u81f3 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">O(N)O(N)<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 1em;vertical-align: -0.25em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0278em\">O<\/span><span class=\"mopen\">(<\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.109em\">N<\/span><span class=\"mclose\">)<\/span><\/span><\/span><\/span><\/span><\/li>\n<li>\u79fb\u4f4d\u7a97\u53e3&#xff08;Shifted Window&#xff09;&#xff1a;\u7a97\u53e3\u5212\u5206\u504f\u79fb <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">M\/2M\/2<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 1em;vertical-align: -0.25em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.109em\">M<\/span><span class=\"mord\">\/2<\/span><\/span><\/span><\/span><\/span>&#xff0c;\u5b9e\u73b0\u8de8\u7a97\u53e3\u8fde\u63a5<\/li>\n<li>\u5c42\u6b21\u5316\u7279\u5f81&#xff1a;\u6bcf\u9636\u6bb5 patches \u5408\u5e76&#xff0c;\u751f\u6210\u591a\u5c3a\u5ea6\u7279\u5f81\u91d1\u5b57\u5854<\/li>\n<\/ul>\n<hr \/>\n<h3>\u56db\u3001\u4e0b\u6e38\u4efb\u52a1\u7684\u8fc1\u79fb\u7b56\u7565<\/h3>\n<h4>4.1 \u56fe\u50cf\u5206\u7c7b<\/h4>\n<p>\u7b56\u7565\u4e00&#xff1a;\u7ebf\u6027\u63a2\u6d4b&#xff08;Linear Probing&#xff09;<\/p>\n<ul>\n<li>\u51bb\u7ed3\u9884\u8bad\u7ec3\u9aa8\u5e72&#xff0c;\u4ec5\u8bad\u7ec3\u5206\u7c7b\u5934<\/li>\n<li>\u6d4b\u8bd5\u9884\u8bad\u7ec3\u7279\u5f81\u7684\u5224\u522b\u6027<\/li>\n<\/ul>\n<p>\u7b56\u7565\u4e8c&#xff1a;\u90e8\u5206\u5fae\u8c03&#xff08;Partial Fine-tuning&#xff09;<\/p>\n<ul>\n<li>\u5fae\u8c03\u6700\u540e\u51e0\u5c42&#xff0c;\u51bb\u7ed3\u6d45\u5c42<\/li>\n<li>\u5e73\u8861\u6548\u7387\u4e0e\u6027\u80fd<\/li>\n<\/ul>\n<p>\u7b56\u7565\u4e09&#xff1a;\u5168\u91cf\u5fae\u8c03&#xff08;Full Fine-tuning&#xff09;<\/p>\n<ul>\n<li>\u4ee5\u8f83\u5c0f\u5b66\u4e60\u7387&#xff08;\u5982 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">10\u2212510^{-5}<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.8141em\"><\/span><span class=\"mord\">1<\/span><span class=\"mord\"><span class=\"mord\">0<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8141em\"><span class=\"\" style=\"top: -3.063em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mtight\">\u2212<\/span><span class=\"mord mtight\">5<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>&#xff09;\u66f4\u65b0\u5168\u90e8\u53c2\u6570<\/li>\n<li>\u6570\u636e\u5145\u8db3\u65f6\u7684\u6700\u4f73\u9009\u62e9<\/li>\n<\/ul>\n<p><span class=\"token comment\"># PyTorch \u4f2a\u4ee3\u7801&#xff1a;\u5206\u5c42\u5b66\u4e60\u7387\u5fae\u8c03<\/span><br \/>\nmodel <span class=\"token operator\">&#061;<\/span> torchvision<span class=\"token punctuation\">.<\/span>models<span class=\"token punctuation\">.<\/span>resnet50<span class=\"token punctuation\">(<\/span>pretrained<span class=\"token operator\">&#061;<\/span><span class=\"token boolean\">True<\/span><span class=\"token punctuation\">)<\/span><\/p>\n<p><span class=\"token comment\"># \u5206\u5c42\u8bbe\u7f6e\u4f18\u5316\u5668\u53c2\u6570<\/span><br \/>\nparam_groups <span class=\"token operator\">&#061;<\/span> <span class=\"token punctuation\">[<\/span><br \/>\n    <span class=\"token punctuation\">{<\/span><span class=\"token string\">&#039;params&#039;<\/span><span class=\"token punctuation\">:<\/span> model<span class=\"token punctuation\">.<\/span>layer1<span class=\"token punctuation\">.<\/span>parameters<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">&#039;lr&#039;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token number\">1e-5<\/span><span class=\"token punctuation\">}<\/span><span class=\"token punctuation\">,<\/span><br \/>\n    <span class=\"token punctuation\">{<\/span><span class=\"token string\">&#039;params&#039;<\/span><span class=\"token punctuation\">:<\/span> model<span class=\"token punctuation\">.<\/span>layer2<span class=\"token punctuation\">.<\/span>parameters<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">&#039;lr&#039;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token number\">1e-4<\/span><span class=\"token punctuation\">}<\/span><span class=\"token punctuation\">,<\/span><br \/>\n    <span class=\"token punctuation\">{<\/span><span class=\"token string\">&#039;params&#039;<\/span><span class=\"token punctuation\">:<\/span> model<span class=\"token punctuation\">.<\/span>layer3<span class=\"token punctuation\">.<\/span>parameters<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">&#039;lr&#039;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token number\">1e-4<\/span><span class=\"token punctuation\">}<\/span><span class=\"token punctuation\">,<\/span><br \/>\n    <span class=\"token punctuation\">{<\/span><span class=\"token string\">&#039;params&#039;<\/span><span class=\"token punctuation\">:<\/span> model<span class=\"token punctuation\">.<\/span>layer4<span class=\"token punctuation\">.<\/span>parameters<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">&#039;lr&#039;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token number\">1e-3<\/span><span class=\"token punctuation\">}<\/span><span class=\"token punctuation\">,<\/span><br \/>\n    <span class=\"token punctuation\">{<\/span><span class=\"token string\">&#039;params&#039;<\/span><span class=\"token punctuation\">:<\/span> model<span class=\"token punctuation\">.<\/span>fc<span class=\"token punctuation\">.<\/span>parameters<span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token string\">&#039;lr&#039;<\/span><span class=\"token punctuation\">:<\/span> <span class=\"token number\">1e-2<\/span><span class=\"token punctuation\">}<\/span><span class=\"token punctuation\">,<\/span>  <span class=\"token comment\"># \u65b0\u5934&#xff0c;\u8f83\u5927\u5b66\u4e60\u7387<\/span><br \/>\n<span class=\"token punctuation\">]<\/span><\/p>\n<p>optimizer <span class=\"token operator\">&#061;<\/span> torch<span class=\"token punctuation\">.<\/span>optim<span class=\"token punctuation\">.<\/span>Adam<span class=\"token punctuation\">(<\/span>param_groups<span class=\"token punctuation\">)<\/span><\/p>\n<h4>4.2 \u76ee\u6807\u68c0\u6d4b\u4e0e\u5b9e\u4f8b\u5206\u5272<\/h4>\n<p>\u4e24\u9636\u6bb5\u65b9\u6cd5&#xff08;Faster R-CNN \u7cfb\u5217&#xff09;&#xff1a;<\/p>\n<ul>\n<li>\u9884\u8bad\u7ec3\u9aa8\u5e72 \u2192 RPN \u63d0\u53d6\u5019\u9009\u6846 \u2192 RoI \u5206\u7c7b\u4e0e\u56de\u5f52<\/li>\n<li>\u5e38\u7528\u9884\u8bad\u7ec3&#xff1a;ImageNet \u76d1\u7763\u3001MoCo\u3001Detectron2 \u63d0\u4f9b\u7684 COCO \u9884\u8bad\u7ec3<\/li>\n<\/ul>\n<p>\u5355\u9636\u6bb5\u65b9\u6cd5&#xff08;YOLO\u3001RetinaNet&#xff09;&#xff1a;<\/p>\n<ul>\n<li>\u7aef\u5230\u7aef\u8bad\u7ec3&#xff0c;\u9884\u8bad\u7ec3\u52a0\u901f\u6536\u655b<\/li>\n<li>YOLOv8 \u652f\u6301\u591a\u79cd\u9884\u8bad\u7ec3\u6743\u91cd\u5207\u6362<\/li>\n<\/ul>\n<p>Transformer-based&#xff08;DETR\u3001Deformable DETR&#xff09;&#xff1a;<\/p>\n<ul>\n<li>\u7f16\u7801\u5668\u7528\u9884\u8bad\u7ec3 ViT\/Swin \u521d\u59cb\u5316<\/li>\n<li>\u89e3\u7801\u5668\u968f\u673a\u521d\u59cb\u5316&#xff0c;\u7aef\u5230\u7aef\u8bad\u7ec3<\/li>\n<\/ul>\n<h4>4.3 \u8bed\u4e49\u5206\u5272\u4e0e\u5168\u666f\u5206\u5272<\/h4>\n<table>\n<tr>\u65b9\u6cd5\u9884\u8bad\u7ec3\u7b56\u7565\u4ee3\u8868\u6a21\u578b<\/tr>\n<tbody>\n<tr>\n<td>FCN\/DeepLab<\/td>\n<td>ImageNet \u9aa8\u5e72 &#043; ASPP<\/td>\n<td>DeepLabV3&#043;<\/td>\n<\/tr>\n<tr>\n<td>U-Net \u7cfb\u5217<\/td>\n<td>\u533b\u5b66\u9886\u57df\u7279\u5b9a\u9884\u8bad\u7ec3<\/td>\n<td>nnU-Net<\/td>\n<\/tr>\n<tr>\n<td>Mask2Former<\/td>\n<td>\u9884\u8bad\u7ec3 Transformer \u9aa8\u5e72<\/td>\n<td>Swin &#043; Mask2Former<\/td>\n<\/tr>\n<tr>\n<td>SAM<\/td>\n<td>1100\u4e07\u56fe\u50cf &#043; 10\u4ebf\u63a9\u7801<\/td>\n<td>\u63d0\u793a\u5206\u5272\u57fa\u7840\u6a21\u578b<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h4>4.4 \u7279\u6b8a\u9886\u57df\u9002\u914d<\/h4>\n<p>\u533b\u5b66\u5f71\u50cf&#xff1a;<\/p>\n<ul>\n<li>\u6311\u6218&#xff1a;\u6570\u636e\u9690\u79c1\u3001\u6807\u6ce8\u7a00\u7f3a\u3001\u57df\u5dee\u5f02\u5927<\/li>\n<li>\u7b56\u7565&#xff1a;ImageNet \u9884\u8bad\u7ec3 \u2192 \u9886\u57df\u81ea\u76d1\u7763\u9884\u8bad\u7ec3&#xff08;SimCLR\u3001MAE&#xff09;\u2192 \u4efb\u52a1\u5fae\u8c03<\/li>\n<li>\u4ee3\u8868&#xff1a;TransUNet\u3001Swin-UNETR\u3001MedSAM<\/li>\n<\/ul>\n<p>\u9065\u611f\u56fe\u50cf&#xff1a;<\/p>\n<ul>\n<li>\u6311\u6218&#xff1a;\u591a\u5c3a\u5ea6\u3001\u591a\u5149\u8c31\u3001\u5927\u573a\u666f<\/li>\n<li>\u7b56\u7565&#xff1a;\u5730\u7406\u7a7a\u95f4\u9884\u8bad\u7ec3&#xff08;RingMo\u3001SkySense&#xff09;\u3001\u65f6\u5e8f\u5efa\u6a21<\/li>\n<li>\u5e94\u7528&#xff1a;\u571f\u5730\u5229\u7528\u5206\u7c7b\u3001\u53d8\u5316\u68c0\u6d4b\u3001\u76ee\u6807\u68c0\u6d4b<\/li>\n<\/ul>\n<p>\u5de5\u4e1a\u68c0\u6d4b&#xff1a;<\/p>\n<ul>\n<li>\u6311\u6218&#xff1a;\u7f3a\u9677\u6837\u672c\u6781\u5c11\u3001\u6b63\u5e38\u6837\u672c\u6d77\u91cf<\/li>\n<li>\u7b56\u7565&#xff1a;\u81ea\u76d1\u7763\u9884\u8bad\u7ec3\u6b63\u5e38\u6837\u672c &#043; \u5f02\u5e38\u68c0\u6d4b\/\u5c11\u6837\u672c\u5b66\u4e60<\/li>\n<li>\u4ee3\u8868&#xff1a;PatchCore\u3001CutPaste\u3001SimpleNet<\/li>\n<\/ul>\n<hr \/>\n<h3>\u4e94\u3001\u6700\u65b0\u8d8b\u52bf&#xff1a;\u57fa\u7840\u6a21\u578b\u4e0e\u751f\u6210\u5f0f\u9884\u8bad\u7ec3<\/h3>\n<h4>5.1 \u89c6\u89c9\u57fa\u7840\u6a21\u578b&#xff08;Visual Foundation Models&#xff09;<\/h4>\n<p>SAM&#xff08;Segment Anything Model&#xff0c;2023&#xff09;&#xff1a;<\/p>\n<ul>\n<li>\u6570\u636e\u89c4\u6a21&#xff1a;1100\u4e07\u56fe\u50cf&#xff0c;10\u4ebf\u9ad8\u8d28\u91cf\u63a9\u7801<\/li>\n<li>\u67b6\u6784&#xff1a;ViT \u56fe\u50cf\u7f16\u7801\u5668 &#043; \u63d0\u793a\u7f16\u7801\u5668 &#043; \u8f7b\u91cf\u63a9\u7801\u89e3\u7801\u5668<\/li>\n<li>\u80fd\u529b&#xff1a;\u96f6\u6837\u672c\u5206\u5272\u3001\u4ea4\u4e92\u5f0f\u63d0\u793a&#xff08;\u70b9\u3001\u6846\u3001\u6587\u672c\u3001\u63a9\u7801&#xff09;<\/li>\n<\/ul>\n<p>DINOv2&#xff08;2023&#xff09;&#xff1a;<\/p>\n<ul>\n<li>\u81ea\u76d1\u7763\u9884\u8bad\u7ec3\u5f3a\u89c6\u89c9\u7279\u5f81<\/li>\n<li>\u65e0\u9700\u5fae\u8c03\u5373\u53ef\u5728\u5206\u7c7b\u3001\u5206\u5272\u3001\u6df1\u5ea6\u4f30\u8ba1\u7b49\u4efb\u52a1\u4e0a\u8868\u73b0\u4f18\u5f02<\/li>\n<li>\u652f\u6301\u591a\u79cd\u5c3a\u5bf8&#xff08;ViT-S\/B\/L\/g&#xff09;&#xff0c;\u5f00\u6e90\u53ef\u5546\u7528<\/li>\n<\/ul>\n<p>EVA\/EVA-02&#xff08;2022-2023&#xff09;&#xff1a;<\/p>\n<ul>\n<li>\u4ee5 CLIP \u89c6\u89c9\u7279\u5f81\u4e3a\u91cd\u5efa\u76ee\u6807&#xff0c;\u7edf\u4e00\u8868\u793a\u5b66\u4e60<\/li>\n<li>\u5728\u591a\u4e2a\u57fa\u51c6\u4e0a\u5237\u65b0\u8bb0\u5f55<\/li>\n<\/ul>\n<h4>5.2 \u751f\u6210\u5f0f\u9884\u8bad\u7ec3<\/h4>\n<p>\u6f5c\u5728\u6269\u6563\u6a21\u578b&#xff08;Latent Diffusion Models&#xff09;&#xff1a;<\/p>\n<table>\n<tr>\u6a21\u578b\u80fd\u529b\u5e94\u7528\u573a\u666f<\/tr>\n<tbody>\n<tr>\n<td>Stable Diffusion<\/td>\n<td>\u6587\u672c\u5230\u56fe\u50cf\u751f\u6210<\/td>\n<td>\u827a\u672f\u521b\u4f5c\u3001\u8bbe\u8ba1\u8f85\u52a9<\/td>\n<\/tr>\n<tr>\n<td>ControlNet<\/td>\n<td>\u6761\u4ef6\u63a7\u5236\u751f\u6210<\/td>\n<td>\u59ff\u6001\u3001\u8fb9\u7f18\u3001\u6df1\u5ea6\u5f15\u5bfc<\/td>\n<\/tr>\n<tr>\n<td>InstructPix2Pix<\/td>\n<td>\u6307\u4ee4\u5f0f\u56fe\u50cf\u7f16\u8f91<\/td>\n<td>\u81ea\u7136\u8bed\u8a00\u4fee\u56fe<\/td>\n<\/tr>\n<tr>\n<td>SDXL\/Turbo<\/td>\n<td>\u9ad8\u5206\u8fa8\u7387\u3001\u5b9e\u65f6\u751f\u6210<\/td>\n<td>\u751f\u4ea7\u73af\u5883\u90e8\u7f72<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u591a\u6a21\u6001\u5927\u6a21\u578b&#xff1a;<\/p>\n<ul>\n<li>GPT-4V\/GPT-4o&#xff1a;\u7aef\u5230\u7aef\u89c6\u89c9-\u8bed\u8a00\u7406\u89e3\u4e0e\u751f\u6210\u529f\u80fd<\/li>\n<li>Gemini&#xff1a;\u539f\u751f\u591a\u6a21\u6001\u67b6\u6784&#xff0c;\u8de8\u6a21\u6001\u63a8\u7406<\/li>\n<li>Qwen-VL\/LLaVA&#xff1a;\u5f00\u6e90\u89c6\u89c9\u6307\u4ee4\u5fae\u8c03\u6a21\u578b<\/li>\n<\/ul>\n<h4>5.3 \u9ad8\u6548\u5fae\u8c03\u6280\u672f&#xff08;Parameter-Efficient Fine-Tuning&#xff09;<\/h4>\n<table>\n<tr>\u6280\u672f\u6838\u5fc3\u601d\u60f3\u53ef\u8bad\u7ec3\u53c2\u6570\u91cf\u9002\u7528\u573a\u666f<\/tr>\n<tbody>\n<tr>\n<td>LoRA (2021)<\/td>\n<td>\u4f4e\u79e9\u9002\u914d&#xff1a;<span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">W&#061;W0&#043;BAW &#061; W_0 &#043; BA<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.6833em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8333em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.1389em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">0<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">&#043;<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.6833em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0502em\">B<\/span><span class=\"mord mathnormal\">A<\/span><\/span><\/span><\/span><\/span><\/td>\n<td>&lt;1%<\/td>\n<td>\u5927\u6a21\u578b\u5feb\u901f\u9002\u914d<\/td>\n<\/tr>\n<tr>\n<td>Adapter (2019)<\/td>\n<td>\u63d2\u5165\u5c0f\u578b\u74f6\u9888\u7f51\u7edc<\/td>\n<td>~5%<\/td>\n<td>\u591a\u4efb\u52a1\u6301\u7eed\u5b66\u4e60<\/td>\n<\/tr>\n<tr>\n<td>Prompt Tuning (2021)<\/td>\n<td>\u5b66\u4e60\u8f6f\u63d0\u793a\u5d4c\u5165<\/td>\n<td>~0.1%<\/td>\n<td>\u5927\u8bed\u8a00\u6a21\u578b<\/td>\n<\/tr>\n<tr>\n<td>VPT (2022)<\/td>\n<td>\u89c6\u89c9\u63d0\u793a\u5fae\u8c03<\/td>\n<td>~0.1%<\/td>\n<td>ViT \u8fc1\u79fb\u5b66\u4e60<\/td>\n<\/tr>\n<tr>\n<td>SSF (2022)<\/td>\n<td>\u7f29\u653e\u4e0e\u79fb\u4f4d\u56e0\u5b50<\/td>\n<td>~0.3%<\/td>\n<td>\u7edf\u4e00\u9ad8\u6548\u5fae\u8c03<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>LoRA \u7684\u6570\u5b66\u5f62\u5f0f&#xff1a;<br \/>\n<span class=\"katex--display\"><span class=\"katex-display\"><span class=\"katex\"><span class=\"katex-mathml\">h&#061;W0x&#043;\u0394Wx&#061;W0x&#043;BAx<br \/>\nh &#061; W_0 x &#043; \\\\Delta W x &#061; W_0 x &#043; BAx<br \/>\n<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.6944em\"><\/span><span class=\"mord mathnormal\">h<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8333em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.1389em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">0<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mord mathnormal\">x<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">&#043;<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.6833em\"><\/span><span class=\"mord\">\u0394<\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"mord mathnormal\">x<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">&#061;<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8333em;vertical-align: -0.15em\"><\/span><span class=\"mord\"><span class=\"mord mathnormal\" style=\"margin-right: 0.1389em\">W<\/span><span class=\"msupsub\"><span class=\"vlist-t vlist-t2\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.3011em\"><span class=\"\" style=\"top: -2.55em;margin-left: -0.1389em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">0<\/span><\/span><\/span><\/span><span class=\"vlist-s\">\u200b<\/span><\/span><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.15em\"><span class=\"\"><\/span><\/span><\/span><\/span><\/span><\/span><span class=\"mord mathnormal\">x<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><span class=\"mbin\">&#043;<\/span><span class=\"mspace\" style=\"margin-right: 0.2222em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.6833em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0502em\">B<\/span><span class=\"mord mathnormal\">A<\/span><span class=\"mord mathnormal\">x<\/span><\/span><\/span><\/span><\/span><\/span><br \/>\n\u5176\u4e2d <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">B\u2208Rd\u00d7rB \\\\in \\\\mathbb{R}^{d \\\\times r}<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.7224em;vertical-align: -0.0391em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0502em\">B<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">\u2208<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8491em\"><\/span><span class=\"mord\"><span class=\"mord mathbb\">R<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8491em\"><span class=\"\" style=\"top: -3.063em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\">d<\/span><span class=\"mbin mtight\">\u00d7<\/span><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0278em\">r<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>&#xff0c;<span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">A\u2208Rr\u00d7kA \\\\in \\\\mathbb{R}^{r \\\\times k}<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.7224em;vertical-align: -0.0391em\"><\/span><span class=\"mord mathnormal\">A<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">\u2208<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 0.8491em\"><\/span><span class=\"mord\"><span class=\"mord mathbb\">R<\/span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height: 0.8491em\"><span class=\"\" style=\"top: -3.063em;margin-right: 0.05em\"><span class=\"pstrut\" style=\"height: 2.7em\"><\/span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\"><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0278em\">r<\/span><span class=\"mbin mtight\">\u00d7<\/span><span class=\"mord mathnormal mtight\" style=\"margin-right: 0.0315em\">k<\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span><\/span>&#xff0c;\u79e9 <span class=\"katex--inline\"><span class=\"katex\"><span class=\"katex-mathml\">r\u226amin\u2061(d,k)r \\\\ll \\\\min(d, k)<\/span><span class=\"katex-html\"><span class=\"base\"><span class=\"strut\" style=\"height: 0.5782em;vertical-align: -0.0391em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0278em\">r<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><span class=\"mrel\">\u226a<\/span><span class=\"mspace\" style=\"margin-right: 0.2778em\"><\/span><\/span><span class=\"base\"><span class=\"strut\" style=\"height: 1em;vertical-align: -0.25em\"><\/span><span class=\"mop\">min<\/span><span class=\"mopen\">(<\/span><span class=\"mord mathnormal\">d<\/span><span class=\"mpunct\">,<\/span><span class=\"mspace\" style=\"margin-right: 0.1667em\"><\/span><span class=\"mord mathnormal\" style=\"margin-right: 0.0315em\">k<\/span><span class=\"mclose\">)<\/span><\/span><\/span><\/span><\/span>&#xff0c;\u53ef\u8bad\u7ec3\u53c2\u6570\u91cf\u5927\u5e45\u51cf\u5c11\u3002<\/p>\n<hr \/>\n<h3>\u516d\u3001\u5b9e\u8df5\u5efa\u8bae&#xff1a;\u5982\u4f55\u9009\u62e9\u9884\u8bad\u7ec3\u6a21\u578b&#xff1f;<\/h3>\n<h4>6.1 \u51b3\u7b56\u6d41\u7a0b\u56fe<\/h4>\n<p>\u5f00\u59cb<br \/>\n  \u2502<br \/>\n  \u251c\u2500\u2500 \u6570\u636e\u89c4\u6a21 &lt; 1K?<br \/>\n  \u2502     \u2514\u2500\u2500 \u662f \u2192 \u9009\u62e9\u5927\u6a21\u578b &#043; \u9ad8\u6548\u5fae\u8c03&#xff08;LoRA\/Adapter&#xff09;<br \/>\n  \u2502     \u2514\u2500\u2500 \u5426 \u2192 \u7ee7\u7eed<br \/>\n  \u2502<br \/>\n  \u251c\u2500\u2500 \u6570\u636e\u89c4\u6a21 1K-10K?<br \/>\n  \u2502     \u2514\u2500\u2500 \u662f \u2192 \u4e2d\u7b49\u6a21\u578b &#043; \u90e8\u5206\u5fae\u8c03&#xff08;\u6700\u540e\u51e0\u5c42&#xff09;<br \/>\n  \u2502     \u2514\u2500\u2500 \u5426 \u2192 \u7ee7\u7eed<br \/>\n  \u2502<br \/>\n  \u251c\u2500\u2500 \u6570\u636e\u89c4\u6a21 &gt; 10K?<br \/>\n  \u2502     \u2514\u2500\u2500 \u662f \u2192 \u5927\u6a21\u578b\u5168\u91cf\u5fae\u8c03\u6216\u4ece\u5934\u8bad\u7ec3<br \/>\n  \u2502<br \/>\n  \u251c\u2500\u2500 \u4efb\u52a1\u7c7b\u578b?<br \/>\n  \u2502     \u251c\u2500\u2500 \u5206\u7c7b \u2192 ImageNet \u9884\u8bad\u7ec3\u6743\u91cd<br \/>\n  \u2502     \u251c\u2500\u2500 \u68c0\u6d4b\/\u5206\u5272 \u2192 COCO \u9884\u8bad\u7ec3\u6216\u68c0\u6d4b\u4e13\u7528\u9884\u8bad\u7ec3<br \/>\n  \u2502     \u251c\u2500\u2500 \u751f\u6210 \u2192 Stable Diffusion \u6216\u81ea\u7814\u6269\u6563\u6a21\u578b<br \/>\n  \u2502     \u2514\u2500\u2500 \u591a\u6a21\u6001 \u2192 CLIP\/BLIP\/GPT-4V API<br \/>\n  \u2502<br \/>\n  \u2514\u2500\u2500 \u9886\u57df\u5dee\u5f02?<br \/>\n        \u251c\u2500\u2500 \u901a\u7528\u89c6\u89c9 \u2192 \u6807\u51c6 ImageNet\/CLIP \u9884\u8bad\u7ec3<br \/>\n        \u251c\u2500\u2500 \u533b\u5b66 \u2192 Medical MAE\/SAM\/MedSAM<br \/>\n        \u251c\u2500\u2500 \u9065\u611f \u2192 RingMo\/SkySense\/\u9886\u57df MAE<br \/>\n        \u2514\u2500\u2500 \u5de5\u4e1a \u2192 \u81ea\u76d1\u7763\u9884\u8bad\u7ec3 &#043; \u5f02\u5e38\u68c0\u6d4b<\/p>\n<h4>6.2 \u63a8\u8350\u8d44\u6e90\u4e0e\u5de5\u5177<\/h4>\n<p>\u6a21\u578b\u5e93\u4e0e\u6846\u67b6&#xff1a;<\/p>\n<ul>\n<li>timm (PyTorch Image Models)&#xff1a;600&#043; \u9884\u8bad\u7ec3\u6a21\u578b&#xff0c;\u7edf\u4e00\u63a5\u53e3<\/li>\n<li>Hugging Face Transformers&#xff1a;\u591a\u6a21\u6001\u6a21\u578b\u4e2d\u5fc3<\/li>\n<li>OpenMMLab (MMPreTrain, MMDetection, MMSegmentation)&#xff1a;\u5b8c\u6574\u89c6\u89c9\u5de5\u5177\u94fe<\/li>\n<li>Detectron2&#xff1a;Facebook \u68c0\u6d4b\/\u5206\u5272\u7814\u7a76\u5e73\u53f0<\/li>\n<\/ul>\n<p>\u6570\u636e\u96c6\u8d44\u6e90&#xff1a;<\/p>\n<ul>\n<li>ImageNet-1K\/21K&#xff1a;\u7ecf\u5178\u5206\u7c7b\u57fa\u51c6<\/li>\n<li>LAION-5B\/400M&#xff1a;\u5927\u89c4\u6a21\u56fe\u6587\u5bf9&#xff0c;CLIP \u8bad\u7ec3\u6570\u636e<\/li>\n<li>SA-1B&#xff1a;SAM \u7684 10 \u4ebf\u63a9\u7801\u6570\u636e\u96c6<\/li>\n<li>\u9886\u57df\u6570\u636e\u96c6&#xff1a;CheXpert&#xff08;\u533b\u5b66&#xff09;\u3001DOTA&#xff08;\u9065\u611f&#xff09;\u3001MVTec AD&#xff08;\u5de5\u4e1a&#xff09;<\/li>\n<\/ul>\n<p>\u5f00\u53d1\u5de5\u5177&#xff1a;<\/p>\n<ul>\n<li>PyTorch Lightning&#xff1a;\u7b80\u5316\u8bad\u7ec3\u6d41\u7a0b<\/li>\n<li>Weights &amp; Biases&#xff1a;\u5b9e\u9a8c\u8ffd\u8e2a\u4e0e\u53ef\u89c6\u5316<\/li>\n<li>Hugging Face Accelerate&#xff1a;\u5206\u5e03\u5f0f\u8bad\u7ec3\u7b80\u5316<\/li>\n<\/ul>\n<h4>6.3 \u6027\u80fd\u5bf9\u6bd4\u53c2\u8003<\/h4>\n<table>\n<tr>\u9884\u8bad\u7ec3\u65b9\u6cd5ImageNet-1K \u5fae\u8c03COCO \u68c0\u6d4b AP\u4e0b\u6e38\u6837\u672c\u6548\u7387<\/tr>\n<tbody>\n<tr>\n<td>\u968f\u673a\u521d\u59cb\u5316<\/td>\n<td>1.0\u00d7 \u57fa\u7ebf<\/td>\n<td>\u57fa\u7ebf<\/td>\n<td>100% \u6570\u636e<\/td>\n<\/tr>\n<tr>\n<td>ImageNet \u76d1\u7763<\/td>\n<td>1.5-2.0\u00d7<\/td>\n<td>&#043;10-15 AP<\/td>\n<td>50% \u6570\u636e<\/td>\n<\/tr>\n<tr>\n<td>SimCLR\/MoCo<\/td>\n<td>1.8-2.2\u00d7<\/td>\n<td>&#043;12-18 AP<\/td>\n<td>30% \u6570\u636e<\/td>\n<\/tr>\n<tr>\n<td>MAE\/BEiT<\/td>\n<td>2.0-2.5\u00d7<\/td>\n<td>&#043;15-20 AP<\/td>\n<td>10% \u6570\u636e<\/td>\n<\/tr>\n<tr>\n<td>CLIP \u96f6\u6837\u672c<\/td>\n<td>0.8-1.2\u00d7<\/td>\n<td>&#8211;<\/td>\n<td>0% \u6570\u636e&#xff08;\u96f6\u6837\u672c&#xff09;<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<hr \/>\n<h3>\u4e03\u3001\u672a\u6765\u5c55\u671b<\/h3>\n<h4>7.1 \u6280\u672f\u8d8b\u52bf<\/h4>\n<li>\u7edf\u4e00\u67b6\u6784&#xff1a;\u89c6\u89c9-\u8bed\u8a00-\u52a8\u4f5c\u591a\u6a21\u6001\u7edf\u4e00\u5efa\u6a21&#xff08;\u5982 GPT-4o\u3001Gemini&#xff09;<\/li>\n<li>\u4e16\u754c\u6a21\u578b&#xff1a;\u4ece\u9884\u6d4b\u4e0b\u4e00\u4e2a token \u5230\u9884\u6d4b\u4e0b\u4e00\u5e27\u89c6\u9891<\/li>\n<li>\u795e\u7ecf\u6e32\u67d3&#xff1a;3D \u573a\u666f\u7406\u89e3\u4e0e\u751f\u6210&#xff08;NeRF\u3001Gaussian Splatting&#xff09;<\/li>\n<li>\u6301\u7eed\u5b66\u4e60&#xff1a;\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u7ec8\u8eab\u5b66\u4e60\u4e0e\u77e5\u8bc6\u66f4\u65b0<\/li>\n<h4>7.2 \u5f00\u653e\u95ee\u9898<\/h4>\n<ul>\n<li>\u5982\u4f55\u8bc4\u4f30\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u901a\u7528\u89c6\u89c9\u80fd\u529b&#xff1f;<\/li>\n<li>\u5982\u4f55\u5b9e\u73b0\u771f\u6b63\u7684\u5c0f\u6837\u672c\u5b66\u4e60&#xff08;1-shot \u751a\u81f3 0-shot&#xff09;&#xff1f;<\/li>\n<li>\u5982\u4f55\u5e73\u8861\u6a21\u578b\u89c4\u6a21\u4e0e\u63a8\u7406\u6548\u7387&#xff1f;<\/li>\n<li>\u5982\u4f55\u5efa\u7acb\u53ef\u4fe1\u7684\u3001\u53ef\u89e3\u91ca\u7684\u89c6\u89c9\u9884\u8bad\u7ec3\u6a21\u578b&#xff1f;<\/li>\n<\/ul>\n<hr \/>\n<h3>\u4e00\u53e5\u8bdd\u603b\u7ed3<\/h3>\n<p>\u56fe\u50cf\u9884\u8bad\u7ec3\u7ecf\u5386\u4e86 \u201c\u76d1\u7763\u5b66\u4e60&#xff08;ImageNet \u65f6\u4ee3&#xff09;\u2192 \u81ea\u76d1\u7763\u5b66\u4e60&#xff08;\u5bf9\u6bd4\u5b66\u4e60\u4e0e\u63a9\u7801\u5efa\u6a21\u65f6\u4ee3&#xff09;\u2192 \u591a\u6a21\u6001\u7edf\u4e00&#xff08;CLIP \u4e0e\u5927\u6a21\u578b\u65f6\u4ee3&#xff09;\u201d \u7684\u4e09\u4ee3\u6f14\u8fdb&#xff0c;\u6838\u5fc3\u76ee\u6807\u59cb\u7ec8\u662f&#xff1a;\u7528\u66f4\u5c11\u6807\u6ce8\u3001\u66f4\u4f4e\u6210\u672c&#xff0c;\u83b7\u5f97\u66f4\u5f3a\u7684\u89c6\u89c9\u7406\u89e3\u3001\u6cdb\u5316\u4e0e\u751f\u6210\u80fd\u529b\u3002\u672a\u6765&#xff0c;\u89c6\u89c9\u9884\u8bad\u7ec3\u5c06\u4e0e\u8bed\u8a00\u3001\u97f3\u9891\u3001\u52a8\u4f5c\u7b49\u591a\u6a21\u6001\u6df1\u5ea6\u878d\u5408&#xff0c;\u8fc8\u5411\u771f\u6b63\u7684\u901a\u7528\u4eba\u5de5\u667a\u80fd\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u4e00\u3001\u4e3a\u4ec0\u4e48\u56fe\u50cf\u9886\u57df\u9700\u8981\u9884\u8bad\u7ec3&#xff1f;<br \/>\n1.1 \u6570\u636e\u56f0\u5883&#xff1a;\u6807\u6ce8\u6210\u672c\u9ad8\u6602<br \/>\n\u4e00\u5f20\u9ad8\u6e05\u56fe\u50cf\u7684\u50cf\u7d20\u6570\u901a\u5e38\u4e3a 10610^6106 \u7ea7\u522b&#xff08;\u5982 224\u00d7224\u00d73150,528224 \\\\times 224 \\\\times 3  150,528224\u00d7224\u00d73150,528 \u50cf\u7d20&#xff09;\u9ad8\u8d28\u91cf\u6807\u6ce8\u9700\u8981\u4e13\u4e1a\u9886\u57df\u77e5\u8bc6&#xff1a;<br \/>\n\u533b\u5b66\u5f71\u50cf&#xff1a;\u9700\u6267\u4e1a\u533b\u5e08\u6807\u6ce8\u75c5\u7076\u533a\u57df\u9065\u611f\u56fe\u50cf&#xff1a;\u9700\u5730\u7406\u4fe1\u606f\u4e13\u5bb6\u8bc6\u522b\u5730\u7269\u7c7b\u578b\u5de5\u4e1a\u68c0\u6d4b&#xff1a;<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[50],"topic":[],"class_list":["post-103329","post","type-post","status-publish","format-standard","hentry","category-server","tag-50"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u56fe\u50cf\u9886\u57df\u7684\u9884\u8bad\u7ec3 - \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\/103329.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u56fe\u50cf\u9886\u57df\u7684\u9884\u8bad\u7ec3 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u4e00\u3001\u4e3a\u4ec0\u4e48\u56fe\u50cf\u9886\u57df\u9700\u8981\u9884\u8bad\u7ec3&#xff1f; 1.1 \u6570\u636e\u56f0\u5883&#xff1a;\u6807\u6ce8\u6210\u672c\u9ad8\u6602 \u4e00\u5f20\u9ad8\u6e05\u56fe\u50cf\u7684\u50cf\u7d20\u6570\u901a\u5e38\u4e3a 10610^6106 \u7ea7\u522b&#xff08;\u5982 224\u00d7224\u00d73150,528224 \\times 224 \\times 3 150,528224\u00d7224\u00d73150,528 \u50cf\u7d20&#xff09;\u9ad8\u8d28\u91cf\u6807\u6ce8\u9700\u8981\u4e13\u4e1a\u9886\u57df\u77e5\u8bc6&#xff1a; \u533b\u5b66\u5f71\u50cf&#xff1a;\u9700\u6267\u4e1a\u533b\u5e08\u6807\u6ce8\u75c5\u7076\u533a\u57df\u9065\u611f\u56fe\u50cf&#xff1a;\u9700\u5730\u7406\u4fe1\u606f\u4e13\u5bb6\u8bc6\u522b\u5730\u7269\u7c7b\u578b\u5de5\u4e1a\u68c0\u6d4b&#xff1a;\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/103329.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-09-10T01:41:08+00:00\" \/>\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=\"6 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/103329.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/103329.html\",\"name\":\"\u56fe\u50cf\u9886\u57df\u7684\u9884\u8bad\u7ec3 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-09-10T01:41:08+00:00\",\"dateModified\":\"2026-09-10T01:41:08+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/103329.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/103329.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/103329.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u56fe\u50cf\u9886\u57df\u7684\u9884\u8bad\u7ec3\"}]},{\"@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":"\u56fe\u50cf\u9886\u57df\u7684\u9884\u8bad\u7ec3 - \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\/103329.html","og_locale":"zh_CN","og_type":"article","og_title":"\u56fe\u50cf\u9886\u57df\u7684\u9884\u8bad\u7ec3 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u4e00\u3001\u4e3a\u4ec0\u4e48\u56fe\u50cf\u9886\u57df\u9700\u8981\u9884\u8bad\u7ec3&#xff1f; 1.1 \u6570\u636e\u56f0\u5883&#xff1a;\u6807\u6ce8\u6210\u672c\u9ad8\u6602 \u4e00\u5f20\u9ad8\u6e05\u56fe\u50cf\u7684\u50cf\u7d20\u6570\u901a\u5e38\u4e3a 10610^6106 \u7ea7\u522b&#xff08;\u5982 224\u00d7224\u00d73150,528224 \\times 224 \\times 3 150,528224\u00d7224\u00d73150,528 \u50cf\u7d20&#xff09;\u9ad8\u8d28\u91cf\u6807\u6ce8\u9700\u8981\u4e13\u4e1a\u9886\u57df\u77e5\u8bc6&#xff1a; \u533b\u5b66\u5f71\u50cf&#xff1a;\u9700\u6267\u4e1a\u533b\u5e08\u6807\u6ce8\u75c5\u7076\u533a\u57df\u9065\u611f\u56fe\u50cf&#xff1a;\u9700\u5730\u7406\u4fe1\u606f\u4e13\u5bb6\u8bc6\u522b\u5730\u7269\u7c7b\u578b\u5de5\u4e1a\u68c0\u6d4b&#xff1a;","og_url":"https:\/\/www.wsisp.com\/helps\/103329.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-09-10T01:41:08+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"6 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/103329.html","url":"https:\/\/www.wsisp.com\/helps\/103329.html","name":"\u56fe\u50cf\u9886\u57df\u7684\u9884\u8bad\u7ec3 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-09-10T01:41:08+00:00","dateModified":"2026-09-10T01:41:08+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/103329.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/103329.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/103329.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u56fe\u50cf\u9886\u57df\u7684\u9884\u8bad\u7ec3"}]},{"@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\/103329","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=103329"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/103329\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=103329"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=103329"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=103329"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=103329"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}