{"id":113763,"date":"2026-10-07T11:57:54","date_gmt":"2026-10-07T03:57:54","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/113763.html"},"modified":"2026-10-07T11:57:54","modified_gmt":"2026-10-07T03:57:54","slug":"%ef%bc%88%e8%ae%ba%e6%96%87%e9%80%9f%e8%af%bb%ef%bc%89fcdm%ef%bc%9aconvnext-%e4%b9%9f%e8%83%bd%e5%81%9a%e9%ab%98%e6%95%88%e6%89%a9%e6%95%a3%e6%a8%a1%e5%9e%8b%ef%bc%8c%e5%8d%b7%e7%a7%af%e7%bd%91","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/113763.html","title":{"rendered":"\uff08\u8bba\u6587\u901f\u8bfb\uff09FCDM\uff1aConvNeXt \u4e5f\u80fd\u505a\u9ad8\u6548\u6269\u6563\u6a21\u578b\uff0c\u5377\u79ef\u7f51\u7edc\u91cd\u65b0\u6311\u6218 DiT"},"content":{"rendered":"<p style=\"text-align:justify\">\u8bba\u6587\u9898\u76ee&#xff1a; Reviving ConvNeXt for Efficient Convolutional Diffusion Models<br \/>\n\u4e2d\u6587\u7ffb\u8bd1&#xff1a; \u91cd\u632f ConvNeXt&#xff1a;\u9762\u5411\u9ad8\u6548\u5377\u79ef\u6269\u6563\u6a21\u578b\u7684\u67b6\u6784\u8bbe\u8ba1<br \/>\n\u4f1a\u8bae&#xff1a; CVPR 2026<\/p>\n<p style=\"text-align:justify\">\u6458\u8981&#xff1a; \u8fd1\u5e74\u6765\u6269\u6563\u6a21\u578b\u8d8a\u6765\u8d8a\u504f\u5411 Transformer \u4e3b\u5e72&#xff0c;\u56e0\u4e3a\u5168\u6ce8\u610f\u529b\u67b6\u6784\u8868\u73b0\u51fa\u5f88\u5f3a\u7684\u53ef\u6269\u5c55\u6027\u3002\u4f46\u5377\u79ef\u7f51\u7edc\u5177\u5907\u7684\u5c40\u90e8\u5f52\u7eb3\u504f\u7f6e\u3001\u53c2\u6570\u6548\u7387\u548c\u786c\u4ef6\u53cb\u597d\u6027&#xff0c;\u5728\u73b0\u4ee3\u751f\u6210\u6a21\u578b\u4e2d\u53cd\u800c\u6ca1\u6709\u88ab\u5145\u5206\u7814\u7a76\u3002\u8bba\u6587\u63d0\u51fa Fully Convolutional Diffusion Model&#xff08;FCDM&#xff09;&#xff0c;\u4ee5 ConvNeXt \u4e3a\u6838\u5fc3\u91cd\u65b0\u8bbe\u8ba1\u6761\u4ef6\u6269\u6563\u6a21\u578b\u3002FCDM-XL \u5728\u53c2\u6570\u89c4\u6a21\u63a5\u8fd1 DiT-XL\/2 \u7684\u60c5\u51b5\u4e0b&#xff0c;\u53ea\u9700\u8981\u7ea6\u4e00\u534a FLOPs&#xff0c;\u5e76\u5728 ImageNet 256\u00d7256 \u548c 512\u00d7512 \u4e0a\u4ee5\u663e\u8457\u66f4\u5c11\u7684\u8bad\u7ec3\u6b65\u6570\u8fbe\u5230\u66f4\u4f18\u6216\u6709\u7ade\u4e89\u529b\u7684\u751f\u6210\u8d28\u91cf\u3002\u66f4\u91cd\u8981\u7684\u662f&#xff0c;\u6700\u5927\u6a21\u578b\u53ef\u4ee5\u5728 4 \u5f20\u6d88\u8d39\u7ea7 GPU \u4e0a\u8bad\u7ec3&#xff0c;\u8bf4\u660e\u73b0\u4ee3 ConvNet \u4ecd\u7136\u662f\u9ad8\u6548\u6269\u6563\u5efa\u6a21\u7684\u91cd\u8981\u8def\u7ebf\u3002<\/p>\n<hr \/>\n<h3 style=\"text-align:justify\">\u4e00\u3001\u4e3a\u4ec0\u4e48\u6269\u6563\u6a21\u578b\u53c8\u8981\u91cd\u65b0\u770b\u5377\u79ef&#xff1f;<\/h3>\n<p style=\"text-align:justify\">DiT \u4e4b\u540e&#xff0c;\u6269\u6563\u6a21\u578b\u4e3b\u5e72\u5feb\u901f\u5411 Transformer \u9760\u62e2\u3002\u539f\u56e0\u5f88\u76f4\u63a5&#xff1a;Transformer \u7ed3\u6784\u7edf\u4e00\u3001\u6269\u5c55\u65b9\u4fbf&#xff0c;\u800c\u4e14\u6a21\u578b\u89c4\u6a21\u589e\u5927\u540e\u6027\u80fd\u901a\u5e38\u6301\u7eed\u63d0\u5347\u3002FLUX\u3001MM-DiT \u7b49\u6a21\u578b\u8fdb\u4e00\u6b65\u5f3a\u5316\u4e86\u8fd9\u79cd\u8d8b\u52bf\u3002<\/p>\n<p style=\"text-align:justify\">\u4f46\u4ee3\u4ef7\u540c\u6837\u660e\u663e&#xff1a;\u6ce8\u610f\u529b\u8ba1\u7b97\u968f Token \u6570\u589e\u52a0\u800c\u53d8\u91cd&#xff0c;\u9ad8\u5206\u8fa8\u7387\u4e0b\u663e\u5b58\u3001\u8ba1\u7b97\u91cf\u548c\u8bad\u7ec3\u6210\u672c\u8fc5\u901f\u4e0a\u5347\u3002\u8bba\u6587\u56e0\u6b64\u63d0\u51fa\u4e00\u4e2a\u95ee\u9898&#xff1a;<\/p>\n<p style=\"text-align:justify\">\u201c\u53ef\u6269\u5c55\u6027\u771f\u7684\u662f Transformer \u72ec\u6709\u7684\u5417&#xff1f;\u201d<\/p>\n<p style=\"text-align:justify\">\u4f5c\u8005\u91cd\u65b0\u5ba1\u89c6 ConvNeXt\u3002\u76f8\u6bd4\u4f20\u7edf\u5377\u79ef\u7f51\u7edc&#xff0c;ConvNeXt \u5df2\u7ecf\u5438\u6536\u4e86\u4e0d\u5c11\u73b0\u4ee3 Transformer \u7684\u8bbe\u8ba1\u601d\u60f3&#xff0c;\u540c\u65f6\u4fdd\u7559\u5377\u79ef\u7684\u5c40\u90e8\u6027\u548c\u786c\u4ef6\u6548\u7387\u3002\u5982\u679c\u628a\u5b83\u4ece\u5206\u7c7b\u7f51\u7edc\u91cd\u65b0\u6539\u9020\u6210\u6761\u4ef6\u6269\u6563\u6a21\u578b&#xff0c;\u4e5f\u8bb8\u80fd\u591f\u83b7\u5f97\u66f4\u597d\u7684\u6027\u80fd\u2014\u6548\u7387\u5e73\u8861\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"645\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035750-6ac5c33ec25d8.png\" width=\"960\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Figure 1&#xff1a;FCDM-XL \u5728 ImageNet 256\u00d7256 \u548c 512\u00d7512 \u4e0a\u7684\u751f\u6210\u6837\u4f8b<\/p>\n<p class=\"img-center\"><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"381\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035751-6ac5c33f48175.png\" width=\"526\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Figure 2&#xff1a;\u4e0d\u540c\u6a21\u578b\u89c4\u6a21\u4e0b FCDM \u4e0e DiT \u7684 FLOPs\u3001FID \u548c\u53ef\u6269\u5c55\u6027\u5bf9\u6bd4<\/p>\n<p style=\"text-align:justify\">Figure 2 \u7ed9\u51fa\u7684\u7ed3\u8bba\u5f88\u76f4\u89c2&#xff1a;\u4ece Small \u5230 XL&#xff0c;FCDM \u968f\u6a21\u578b\u89c4\u6a21\u589e\u5927\u6301\u7eed\u964d\u4f4e FID&#xff0c;\u800c\u4e14\u5728\u76f8\u8fd1\u53c2\u6570\u91cf\u4e0b\u59cb\u7ec8\u6bd4 DiT \u4f7f\u7528\u66f4\u5c11 FLOPs\u3001\u6536\u655b\u66f4\u5feb\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"486\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035751-6ac5c33f5e4c6.png\" width=\"882\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Table 1&#xff1a;\u76f8\u540c\u8bad\u7ec3\u6b65\u6570\u4e0e\u4e0d\u540c\u6a21\u578b\u89c4\u6a21\u4e0b FCDM \u548c DiT \u7684\u6548\u7387\u3001\u541e\u5410\u91cf\u4e0e FID \u5bf9\u6bd4<\/p>\n<p style=\"text-align:justify\">\u4f8b\u5982 400K iterations \u65f6&#xff0c;FCDM-XL \u7684\u8ba1\u7b97\u91cf\u7ea6\u4e3a 65 GFLOPs\u3001\u541e\u5410\u91cf 272.7 it\/s\u3001FID 10.7&#xff1b;DiT-XL\/2 \u5219\u7ea6\u4e3a 119 GFLOPs\u300180.5 it\/s\u3001FID 19.5\u3002\u7ee7\u7eed\u8bad\u7ec3\u5230 1M steps \u540e&#xff0c;FCDM-XL \u7684 FID \u8fbe\u5230 7.9&#xff0c;\u751a\u81f3\u4f18\u4e8e\u8bad\u7ec3 7M steps \u7684 DiT-XL\/2&#xff08;FID 9.6&#xff09;\u3002<\/p>\n<hr \/>\n<h3 style=\"text-align:justify\">\u4e8c\u3001FCDM&#xff1a;\u600e\u4e48\u628a ConvNeXt \u6539\u9020\u6210\u6269\u6563\u6a21\u578b&#xff1f;<\/h3>\n<p style=\"text-align:justify\">FCDM \u5e76\u4e0d\u662f\u7b80\u5355\u628a DiT \u7684 Transformer Block \u6362\u6210\u5377\u79ef&#xff0c;\u800c\u662f\u9488\u5bf9\u6269\u6563\u6a21\u578b\u91cd\u65b0\u7ec4\u7ec7 ConvNeXt\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"612\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035751-6ac5c33f881df.png\" width=\"1302\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Figure 3&#xff1a;ConvNeXt Block\u3001FCDM Block \u4e0e\u5b8c\u6574 U-shaped FCDM \u67b6\u6784<\/p>\n<h4 style=\"text-align:justify\">2.1 \u4fdd\u7559 ConvNeXt \u7684\u6838\u5fc3 Block<\/h4>\n<p style=\"text-align:justify\">\u539f\u59cb ConvNeXt Block \u5305\u542b&#xff1a;<\/p>\n<ul>\n<li>\n<p style=\"text-align:justify\">7\u00d77 Depthwise Convolution&#xff1b;<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">LayerNorm&#xff1b;<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">\u4e24\u4e2a 1\u00d71 Pointwise Convolution&#xff1b;<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">Inverted Bottleneck \u5f0f\u901a\u9053\u6269\u5f20\u4e0e\u538b\u7f29&#xff1b;<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">Global Response Normalization&#xff08;GRN&#xff09;\u3002<\/p>\n<\/li>\n<\/ul>\n<p style=\"text-align:justify\">\u5176\u4e2d\u5927\u5377\u79ef\u6838\u8d1f\u8d23\u6269\u5927\u6709\u6548\u611f\u53d7\u91ce&#xff0c;Depthwise Conv \u63a7\u5236\u8ba1\u7b97\u91cf&#xff0c;\u800c\u5012\u7f6e\u74f6\u9888\u8ba9\u9ad8\u7ef4\u901a\u9053\u7a7a\u95f4\u627f\u62c5\u66f4\u591a\u7279\u5f81\u53d8\u6362\u3002<\/p>\n<h4 style=\"text-align:justify\">2.2 \u7528 AdaLN \u6ce8\u5165\u65f6\u95f4\u548c\u7c7b\u522b\u6761\u4ef6<\/h4>\n<p style=\"text-align:justify\">\u5206\u7c7b\u7248 ConvNeXt \u6ca1\u6709\u6269\u6563\u6a21\u578b\u9700\u8981\u7684\u65f6\u95f4\u6b65\u548c\u7c7b\u522b\u6761\u4ef6&#xff0c;\u56e0\u6b64 FCDM \u5c06 LayerNorm \u66ff\u6362\u4e3a Adaptive LayerNorm&#xff08;AdaLN&#xff09;\u3002<\/p>\n<p style=\"text-align:justify\">\u7c7b\u522b Embedding \u548c Timestep Embedding \u5148\u7ec4\u5408\u6210\u6761\u4ef6\u5411\u91cf&#xff0c;\u518d\u901a\u8fc7\u8f7b\u91cf MLP \u4ea7\u751f \u03b3\u3001\u03b2\u3001\u03b1&#xff0c;\u7528\u4e8e\u8c03\u5236\u5f52\u4e00\u5316\u540e\u7684\u7279\u5f81\u3002\u4e0e DiT \u7c7b\u4f3c&#xff0c;\u6700\u7ec8\u8c03\u5236\u5c3a\u5ea6 \u03b1 \u91c7\u7528\u96f6\u521d\u59cb\u5316&#xff0c;\u4ece\u800c\u63d0\u9ad8\u6df1\u5c42\u7f51\u7edc\u8bad\u7ec3\u7a33\u5b9a\u6027\u3002<\/p>\n<p style=\"text-align:justify\">\u6240\u4ee5 FCDM \u4fdd\u7559\u7684\u662f ConvNeXt \u7684\u5c40\u90e8\u5377\u79ef\u7ed3\u6784&#xff0c;\u4f46\u6761\u4ef6\u6ce8\u5165\u65b9\u5f0f\u501f\u9274\u4e86\u73b0\u4ee3\u6269\u6563 Transformer\u3002<\/p>\n<h4 style=\"text-align:justify\">2.3 U-shaped \u67b6\u6784&#xff0c;\u4f46\u53ea\u7528\u4e24\u4e2a\u8d85\u53c2\u6570\u6269\u5c55<\/h4>\n<p style=\"text-align:justify\">\u5377\u79ef\u7f51\u7edc\u5929\u7136\u9002\u5408 U-Net \u5f0f\u5c42\u7ea7\u7ed3\u6784\u3002FCDM \u5c06 ConvNeXt Block \u653e\u5165 Encoder\u2014Decoder \u7ed3\u6784&#xff0c;\u5e76\u901a\u8fc7 Skip Connection \u5c06\u6d45\u5c42\u9ad8\u5206\u8fa8\u7387\u7ec6\u8282\u9001\u5230\u89e3\u7801\u9636\u6bb5\u3002<\/p>\n<p style=\"text-align:justify\">\u5b83\u523b\u610f\u907f\u514d\u590d\u6742\u7684\u201c\u6bcf\u4e00\u5c42\u5206\u522b\u914d\u7f6e\u591a\u5c11 Block\u3001\u591a\u5c11 Channel\u201d\u7684\u624b\u5de5\u8bbe\u8ba1&#xff0c;\u53ea\u4fdd\u7559\u4e24\u4e2a\u4e3b\u8981\u7f29\u653e\u53c2\u6570&#xff1a;Block \u6570\u91cf\u548c\u9690\u85cf\u901a\u9053\u6570<\/p>\n<p style=\"text-align:justify\">\u6bcf\u7ecf\u8fc7\u4e00\u6b21 2\u00d7 Downsampling&#xff0c;L \u548c C \u90fd\u6309\u89c4\u5219\u6269\u5c55\u3002\u56e0\u6b64\u4ece FCDM-S\u3001B\u3001L \u5230 XL&#xff0c;\u53ea\u9700\u8981\u8c03\u6574\u5c11\u6570\u53c2\u6570\u5373\u53ef\u5b8c\u6210\u89c4\u6a21\u653e\u5927\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"140\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035751-6ac5c33fcbb93.png\" width=\"713\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Table 2&#xff1a;FCDM-S\/B\/L\/XL \u7684\u53c2\u6570\u91cf\u3001Block\u3001Channel \u4e0e FLOPs<\/p>\n<p style=\"text-align:justify\">\u5728\u53c2\u6570\u91cf\u4e0e DiT \u5bf9\u9f50\u540e&#xff0c;FCDM \u6574\u4f53\u53ea\u4f7f\u7528\u5927\u7ea6 50% \u7684 DiT FLOPs&#xff0c;\u540c\u65f6\u7ea6\u4e3a DiCo FLOPs \u7684 75%\u3002\u4f8b\u5982 FCDM-XL \u4e3a 698.8M \u53c2\u6570&#xff0c;\u4f46\u53ea\u9700 64.6 GFLOPs\u3002<\/p>\n<hr \/>\n<h3 style=\"text-align:justify\">\u4e09\u3001\u4e3a\u4ec0\u4e48\u6bd4\u5df2\u6709\u5377\u79ef\u6269\u6563\u6a21\u578b DiCo \u66f4\u9ad8\u6548&#xff1f;<\/h3>\n<p style=\"text-align:justify\">\u4f5c\u8005\u8fd8\u4e13\u95e8\u6bd4\u8f83\u4e86\u5f53\u524d\u5377\u79ef\u6269\u6563\u6a21\u578b DiCo\u3002<\/p>\n<p class=\"img-center\"><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"367\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035751-6ac5c33fd85c1.png\" width=\"470\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Figure 4&#xff1a;DiCo Block \u4e0e FCDM Block \u7684\u7ed3\u6784\u5dee\u5f02<\/p>\n<p style=\"text-align:justify\">\u4e24\u8005\u90fd\u6709 Depthwise\/Separable Convolution&#xff0c;\u4f46 FCDM \u6709\u4e09\u4e2a\u5173\u952e\u5dee\u5f02\u3002<\/p>\n<p style=\"text-align:justify\">\u7b2c\u4e00&#xff0c;\u5148 Depthwise Conv&#xff0c;\u518d\u505a\u901a\u9053\u6269\u5f20\u3002\u8fd9\u6837 7\u00d77 Depthwise Conv \u59cb\u7ec8\u5728\u8f83\u4f4e\u901a\u9053\u7ef4\u5ea6\u4e0a\u6267\u884c&#xff0c;\u6269\u5f20\u540e\u7684\u9ad8\u7ef4\u7a7a\u95f4\u4e3b\u8981\u4ea4\u7ed9 1\u00d71 Conv \u5904\u7406&#xff0c;\u56e0\u6b64\u65e2\u4fdd\u7559\u8f83\u5927\u611f\u53d7\u91ce&#xff0c;\u53c8\u4e0d\u4f1a\u56e0\u4e3a\u901a\u9053\u81a8\u80c0\u663e\u8457\u589e\u52a0\u5927\u5377\u79ef\u6838\u6210\u672c\u3002<\/p>\n<p style=\"text-align:justify\">\u7b2c\u4e8c&#xff0c;FCDM \u4f7f\u7528 GRN \u4ee3\u66ff DiCo \u7684 Compact Channel Attention&#xff08;CCA&#xff09;\u3002\u4e8c\u8005\u90fd\u5e0c\u671b\u51cf\u5c11\u901a\u9053\u5197\u4f59\u3001\u589e\u5f3a\u4e0d\u540c\u901a\u9053\u54cd\u5e94&#xff0c;\u4f46 CCA \u9700\u8981\u989d\u5916 1\u00d71 Conv&#xff0c;\u800c GRN \u4e3b\u8981\u7531 L2 Normalization \u548c Response Normalization \u6784\u6210&#xff0c;\u66f4\u8f7b\u91cf\u3002<\/p>\n<p style=\"text-align:justify\">\u7b2c\u4e09&#xff0c;FCDM \u4e0d\u518d\u989d\u5916\u589e\u52a0 Feedforward Module\u3002\u56e0\u4e3a Inverted Bottleneck \u672c\u8eab\u5df2\u7ecf\u5b8c\u6210\u901a\u9053\u6269\u5f20\u548c\u975e\u7ebf\u6027\u53d8\u6362&#xff0c;\u518d\u5806 FFN \u53cd\u800c\u4f1a\u5e26\u6765\u5197\u4f59\u3002<\/p>\n<p style=\"text-align:justify\">\u56e0\u6b64\u8fd9\u7bc7\u8bba\u6587\u7684\u6838\u5fc3\u5e76\u4e0d\u662f\u201c\u5377\u79ef\u4e00\u5b9a\u6bd4 Transformer \u5f3a\u201d&#xff0c;\u800c\u662f\u8bf4\u660e&#xff1a;\u5982\u679c\u91cd\u65b0\u8bbe\u8ba1\u5377\u79ef Block&#xff0c;ConvNeXt \u7684\u5c40\u90e8\u8ba1\u7b97\u548c\u5c42\u7ea7\u7ed3\u6784\u53ef\u4ee5\u975e\u5e38\u9002\u5408\u6269\u6563\u6a21\u578b\u3002<\/p>\n<hr \/>\n<h3 style=\"text-align:justify\">\u56db\u3001\u5b9e\u9a8c\u8bbe\u7f6e&#xff1a;\u5c3d\u91cf\u53ea\u6bd4\u8f83\u201c\u67b6\u6784\u672c\u8eab\u201d<\/h3>\n<p style=\"text-align:justify\">\u8bba\u6587\u5728 ImageNet-1K \u4e0a\u8fdb\u884c Class-Conditional Latent Diffusion \u5b9e\u9a8c&#xff0c;\u5206\u8fa8\u7387\u5305\u62ec 256\u00d7256 \u548c 512\u00d7512\u3002<\/p>\n<p style=\"text-align:justify\">\u8bad\u7ec3\u57fa\u672c\u6cbf\u7528 DiT \u7684\u6807\u51c6\u8bbe\u7f6e&#xff1a;<\/p>\n<ul>\n<li>\n<p style=\"text-align:justify\">AdamW&#xff1b;<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">Learning Rate&#xff1a;1\u00d710^-4&#xff1b;<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">Batch Size&#xff1a;256&#xff1b;<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">\u65e0 Weight Decay&#xff1b;<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">\u4ec5\u4f7f\u7528 Horizontal Flip&#xff1b;<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">EMA Decay&#xff1a;0.9999&#xff1b;<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">Diffusion Timestep&#xff1a;1000\u3002<\/p>\n<\/li>\n<\/ul>\n<p style=\"text-align:justify\">\u8bc4\u6d4b\u65f6\u751f\u6210 50K \u5f20\u56fe\u50cf&#xff0c;\u4f7f\u7528 250 \u4e2a DDPM Sampling Steps\u3002\u4e3b\u8981\u6307\u6807\u662f FID&#xff0c;\u540c\u65f6\u62a5\u544a IS\u3001Precision \u548c Recall\u3002<\/p>\n<p style=\"text-align:justify\">\u8ba1\u7b97\u8d44\u6e90\u4e0a&#xff0c;FCDM-XL \u5728 256\u00d7256 \u4e0b\u53ef\u4ee5\u7528 4 \u5f20 RTX 4090 24GB\u3001Global Batch Size 256 \u8bad\u7ec3&#xff0c;\u901f\u5ea6\u7ea6 0.9 it\/s&#xff1b;\u8bba\u6587\u8fd8\u9a8c\u8bc1\u4e86 Batch Size 256 \u53ef\u4ee5\u653e\u5165\u5355\u5f20 A100 40GB\u3002<\/p>\n<p style=\"text-align:justify\">\u8fd9\u90e8\u5206\u5f88\u91cd\u8981&#xff0c;\u56e0\u4e3a\u6587\u7ae0\u91cd\u70b9\u4e0d\u662f\u5237\u65b0\u7edd\u5bf9\u6700\u5f3a FID&#xff0c;\u800c\u662f\u9a8c\u8bc1&#xff1a;\u5728\u76f8\u5bf9\u5e38\u89c4\u7684 DiT \u8bad\u7ec3\u6846\u67b6\u4e0b&#xff0c;\u67b6\u6784\u672c\u8eab\u80fd\u4e0d\u80fd\u663e\u8457\u964d\u4f4e\u751f\u6210\u6210\u672c\u3002<\/p>\n<hr \/>\n<h3 style=\"text-align:justify\">\u4e94\u3001\u4e3b\u5b9e\u9a8c&#xff1a;\u66f4\u5c11 FLOPs\u3001\u66f4\u9ad8\u541e\u5410\u3001\u66f4\u5feb\u6536\u655b<\/h3>\n<h4 style=\"text-align:justify\">5.1 \u4ece Small \u5230 XL&#xff0c;\u5377\u79ef\u540c\u6837\u80fd Scaling<\/h4>\n<p><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"624\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035751-6ac5c33fec45a.png\" width=\"1013\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Table 3&#xff1a;ImageNet 256\u00d7256 \u4e0a\u4e0d\u540c\u5c3a\u5ea6\u6a21\u578b\u7684\u751f\u6210\u8d28\u91cf\u4e0e\u6548\u7387\u6bd4\u8f83<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"240\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035752-6ac5c34025b48.png\" width=\"1119\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Figure 5&#xff1a;FCDM \u4e0e DiT \u5728\u4e0d\u540c\u6a21\u578b\u89c4\u6a21\u4e0b\u7684 FID \u6536\u655b\u66f2\u7ebf<\/p>\n<p style=\"text-align:justify\">400K steps \u4e0b&#xff1a;<\/p>\n<ul>\n<li>\n<p style=\"text-align:justify\">FCDM-S&#xff1a;FID 48.52<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">FCDM-B&#xff1a;FID 26.21<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">FCDM-L&#xff1a;FID 13.83<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">FCDM-XL&#xff1a;FID 10.72<\/p>\n<\/li>\n<\/ul>\n<p style=\"text-align:justify\">\u6a21\u578b\u8d8a\u5927&#xff0c;FID \u6301\u7eed\u4e0b\u964d&#xff0c;\u6ca1\u6709\u51fa\u73b0\u201c\u5377\u79ef\u7f51\u7edc\u6269\u4e0d\u52a8\u201d\u7684\u73b0\u8c61\u3002<\/p>\n<p style=\"text-align:justify\">\u66f4\u5173\u952e\u7684\u662f FCDM-XL \u5728 1M steps \u65f6\u8fbe\u5230 FID 7.91&#xff0c;\u8ba1\u7b97\u91cf\u53ea\u6709 64.6 GFLOPs&#xff0c;\u541e\u5410\u91cf 272.7 it\/s\u3002\u540c\u8868\u4e2d\u7684 DiT-XL\/2 \u5373\u4f7f\u8bad\u7ec3 7M steps&#xff0c;FID \u4ecd\u4e3a 9.62&#xff0c;\u800c\u4e14\u5355\u6b21\u8ba1\u7b97\u4e3a 118.6 GFLOPs\u3001\u541e\u5410\u53ea\u6709 80.5 it\/s\u3002<\/p>\n<p style=\"text-align:justify\">\u8fd9\u8bf4\u660e FCDM \u7684\u4f18\u52bf\u540c\u65f6\u6765\u81ea\u4e24\u5c42&#xff1a;<\/p>\n<p style=\"text-align:justify\">\u5355\u6b65\u8bad\u7ec3\u66f4\u4fbf\u5b9c &#043; \u8fbe\u5230\u76ee\u6807\u8d28\u91cf\u9700\u8981\u7684\u8bad\u7ec3\u6b65\u6570\u66f4\u5c11\u3002<\/p>\n<h4 style=\"text-align:justify\">5.2 256\u00d7256&#xff1a;\u6027\u80fd\u548c\u6548\u7387\u4e00\u8d77\u770b<\/h4>\n<p><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"510\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035752-6ac5c34049150.png\" width=\"954\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Table 4&#xff1a;ImageNet 256\u00d7256 Class-Conditional Generation Benchmark<\/p>\n<p style=\"text-align:justify\">\u52a0\u5165 Classifier-Free Guidance \u540e&#xff0c;\u8bad\u7ec3 400 epochs \u7684 FCDM-XL \u8fbe\u5230&#xff1a;<\/p>\n<ul>\n<li>\n<p style=\"text-align:justify\">FID&#xff1a;2.03<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">IS&#xff1a;285.7<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">FLOPs&#xff1a;64.6G<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">Throughput&#xff1a;272.7 it\/s<\/p>\n<\/li>\n<\/ul>\n<p style=\"text-align:justify\">\u5176 FID \u4e0e SiT\u3001DiCo \u7b49\u5f3a\u57fa\u7ebf\u5904\u4e8e\u540c\u4e00\u6c34\u5e73\u751a\u81f3\u7565\u4f18&#xff0c;\u4f46\u5355\u6b21 FLOPs \u548c\u541e\u5410\u660e\u663e\u66f4\u6709\u4f18\u52bf\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"327\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035752-6ac5c3406e1ad.png\" width=\"1071\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Figure 6&#xff1a;FID \u4e0e\u603b\u8bad\u7ec3\u6210\u672c\u3001FID \u4e0e\u541e\u5410\u91cf\u7684\u8054\u5408\u6bd4\u8f83<\/p>\n<p style=\"text-align:justify\">Figure 6 \u66f4\u80fd\u4f53\u73b0\u4f5c\u8005\u60f3\u5f3a\u8c03\u7684\u70b9&#xff1a;FCDM \u4f4d\u4e8e\u201c\u8f83\u4f4e\u8bad\u7ec3\u6210\u672c &#043; \u8f83\u9ad8\u541e\u5410 &#043; \u8f83\u4f4e FID\u201d\u7684\u4f18\u52bf\u533a\u57df\u3002\u4e0e DiCo \u76f8\u6bd4&#xff0c;\u8bba\u6587\u6307\u51fa FCDM \u53ef\u4ee5\u7528\u7ea6 2.5\u00d7 \u66f4\u5c11\u7684\u603b\u8bad\u7ec3 FLOPs\u53d6\u5f97\u6709\u7ade\u4e89\u529b\u7684\u751f\u6210\u6027\u80fd&#xff0c;\u5e76\u83b7\u5f97\u7ea6 1.5\u00d7 \u66f4\u9ad8\u7684\u63a8\u7406\u541e\u5410\u3002<\/p>\n<p style=\"text-align:justify\">\u4e0d\u8fc7\u4f5c\u8005\u4e5f\u660e\u786e\u8bf4\u660e&#xff1a;FCDM \u76ee\u524d\u5e76\u6ca1\u6709\u8d85\u8fc7 EDM-2\u3001Simpler Diffusion \u7b49\u6700\u65b0\u7edd\u5bf9 SOTA\u3002\u56e0\u6b64\u5b83\u7684\u5356\u70b9\u5e94\u7406\u89e3\u4e3a\u6027\u80fd\u2014\u6548\u7387\u6743\u8861&#xff0c;\u800c\u4e0d\u662f\u5355\u7eaf\u8ffd\u6c42\u6700\u4f4e FID\u3002<\/p>\n<h4 style=\"text-align:justify\">5.3 512\u00d7512&#xff1a;\u5206\u8fa8\u7387\u8d8a\u9ad8&#xff0c;\u5377\u79ef\u4f18\u52bf\u8d8a\u660e\u663e<\/h4>\n<p><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"324\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035752-6ac5c34099395.png\" width=\"1110\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Table 5&#xff1a;ImageNet 512\u00d7512 \u4e0a\u7684\u751f\u6210\u8d28\u91cf\u4e0e\u6548\u7387\u5bf9\u6bd4<\/p>\n<p style=\"text-align:justify\">\u5728 512\u00d7512 \u4e0b&#xff0c;FCDM-XL 400K steps \u5df2\u8fbe\u5230 FID 10.23&#xff1b;\u8bad\u7ec3\u5230 1M steps \u540e&#xff0c;FID \u8fdb\u4e00\u6b65\u8fbe\u5230 7.46&#xff0c;FLOPs \u4e3a 257.7G&#xff0c;\u541e\u5410\u91cf 129.6 it\/s\u3002<\/p>\n<p style=\"text-align:justify\">\u76f8\u6bd4\u4e4b\u4e0b&#xff0c;DiT-XL\/2 \u7684\u8ba1\u7b97\u91cf\u4e3a 524.7G&#xff0c;\u541e\u5410\u4ec5 18.6 it\/s&#xff1b;\u5373\u4f7f\u8bad\u7ec3\u5230 3M steps&#xff0c;\u8868\u4e2d FID \u4ecd\u4e3a 12.03\u3002<\/p>\n<p style=\"text-align:justify\">\u8bba\u6587\u8fd8\u6307\u51fa&#xff0c;\u5f53\u5206\u8fa8\u7387\u4ece 256 \u7ffb\u500d\u5230 512 \u65f6&#xff0c;DiT \u541e\u5410\u4e0b\u964d\u7ea6 4\u00d7&#xff0c;\u800c FCDM \u53ea\u4e0b\u964d\u7ea6 2\u00d7\u3002\u8fd9\u6b63\u4f53\u73b0\u4e86 Attention \u4e0e\u5c40\u90e8\u5377\u79ef\u5728\u9ad8\u5206\u8fa8\u7387\u4e0b\u8ba1\u7b97\u590d\u6742\u5ea6\u7684\u5dee\u5f02\u3002<\/p>\n<hr \/>\n<h3 style=\"text-align:justify\">\u516d\u3001\u6d88\u878d\u5b9e\u9a8c&#xff1a;ConvNeXt \u7684\u54ea\u4e9b\u8bbe\u8ba1\u771f\u7684\u91cd\u8981&#xff1f;<\/h3>\n<p><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"450\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035752-6ac5c340c3106.png\" width=\"1281\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Figure 7&#xff1a;GRN \u524d\u540e Feature Activation \u53ef\u89c6\u5316<\/p>\n<p class=\"img-center\"><img loading=\"lazy\" decoding=\"async\" alt=\"\" height=\"271\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035753-6ac5c3413be8c.png\" width=\"469\" \/><\/p>\n<p style=\"text-align:justify\">\u8bba\u6587 Table 6&#xff1a;\u5377\u79ef\u6838\u5927\u5c0f\u3001GRN\u3001Feedforward\u3001Inverted Bottleneck \u4e0e Block \u8bbe\u8ba1\u6d88\u878d<\/p>\n<p style=\"text-align:justify\">\u9ed8\u8ba4 FCDM-L \u4f7f\u7528 7\u00d77 DWConv&#xff0c;\u5728 200K steps \u4e0b FID \u4e3a 19.97\u3002\u6539\u6210 5\u00d75 \u540e\u53d8\u4e3a 20.48&#xff0c;3\u00d73 \u540e\u8fdb\u4e00\u6b65\u53d8\u4e3a 21.28&#xff0c;\u8bf4\u660e\u5927\u5377\u79ef\u6838\u5e26\u6765\u7684\u66f4\u5927\u6709\u6548\u611f\u53d7\u91ce\u786e\u5b9e\u91cd\u8981\u3002<\/p>\n<p style=\"text-align:justify\">\u5c06 GRN \u66ff\u6362\u6210 DiCo \u7684 CCA \u540e&#xff0c;FID \u6076\u5316\u5230 23.85\u3002Figure 7 \u4e5f\u663e\u793a&#xff0c;GRN \u80fd\u660e\u663e\u964d\u4f4e\u4e0d\u540c Feature Channel \u4e4b\u95f4\u7684\u5197\u4f59\u3002<\/p>\n<p style=\"text-align:justify\">\u51e0\u4e2a\u66f4\u6fc0\u8fdb\u7684\u6d88\u878d\u4e0b\u964d\u5f97\u66f4\u660e\u663e&#xff1a;<\/p>\n<ul>\n<li>\n<p style=\"text-align:justify\">\u52a0\u5165\u989d\u5916 Feedforward&#xff1a;FID 28.52<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">\u53bb\u6389 Inverted Bottleneck&#xff1a;FID 28.76<\/p>\n<\/li>\n<li>\n<p style=\"text-align:justify\">\u5c06 FCDM Block \u6362\u6210 ResNet Block&#xff1a;FID 31.14<\/p>\n<\/li>\n<\/ul>\n<p style=\"text-align:justify\">\u8fd9\u4e9b\u7ed3\u679c\u8bf4\u660e&#xff0c;FCDM \u7684\u6027\u80fd\u4e0d\u662f\u201c\u53ea\u8981\u7528\u5377\u79ef\u5c31\u884c\u201d&#xff0c;\u800c\u662f\u4f9d\u8d56 ConvNeXt \u7684\u4e00\u6574\u5957\u73b0\u4ee3\u5316\u8bbe\u8ba1&#xff1a;\u5927\u6838 Depthwise Conv &#043; Inverted Bottleneck &#043; GRN &#043; \u7b80\u6d01 Block \u7ed3\u6784\u3002<\/p>\n<hr \/>\n<h3 style=\"text-align:justify\">\u4e03\u3001\u603b\u7ed3\u4e0e\u601d\u8003<\/h3>\n<p style=\"text-align:justify\">FCDM \u6700\u503c\u5f97\u8bb0\u4f4f\u7684\u89c2\u70b9\u662f&#xff1a;<\/p>\n<p style=\"text-align:justify\">\u6269\u6563\u6a21\u578b\u7684 Scaling \u5e76\u4e0d\u4e13\u5c5e\u4e8e Transformer\u3002<\/p>\n<p style=\"text-align:justify\">\u4f5c\u8005\u628a ConvNeXt \u91cd\u65b0\u8bbe\u8ba1\u4e3a\u6761\u4ef6\u6269\u6563 Backbone&#xff0c;\u901a\u8fc7 AdaLN \u6ce8\u5165\u6761\u4ef6\u3001U-shaped \u5c42\u7ea7\u7ed3\u6784\u4fdd\u6301\u591a\u5c3a\u5ea6\u4fe1\u606f&#xff0c;\u518d\u5229\u7528\u5927\u6838 Depthwise Conv\u3001Inverted Bottleneck \u548c GRN \u6784\u5efa\u9ad8\u6548 Block\u3002\u7ed3\u679c\u8868\u660e&#xff0c;\u5728\u4e0e DiT \u5bf9\u9f50\u53c2\u6570\u91cf\u7684\u60c5\u51b5\u4e0b&#xff0c;FCDM \u5927\u7ea6\u53ea\u9700\u8981\u4e00\u534a FLOPs&#xff0c;\u540c\u65f6\u62e5\u6709\u66f4\u9ad8\u541e\u5410\u548c\u66f4\u5feb FID \u6536\u655b\u3002<\/p>\n<p style=\"text-align:justify\">\u8fd9\u7bc7\u8bba\u6587\u771f\u6b63\u6311\u6218\u7684\u662f\u4e00\u4e2a\u8fd1\u5e74\u6765\u8d8a\u6765\u8d8a\u9ed8\u8ba4\u7684\u524d\u63d0&#xff1a;\u751f\u6210\u6a21\u578b\u8981\u7ee7\u7eed Scaling&#xff0c;\u5c31\u5fc5\u987b\u4f9d\u8d56\u8d8a\u6765\u8d8a\u5927\u7684 Transformer\u3002<\/p>\n<p style=\"text-align:justify\">FCDM \u7ed9\u51fa\u7684\u7b54\u6848\u66f4\u514b\u5236&#xff1a;Transformer \u5f53\u7136\u5f88\u5f3a&#xff0c;\u4f46\u73b0\u4ee3 ConvNet \u7684\u5c40\u90e8\u5f52\u7eb3\u504f\u7f6e\u548c\u786c\u4ef6\u6548\u7387\u5e76\u6ca1\u6709\u5931\u6548\u3002\u5c24\u5176\u5728\u9ad8\u5206\u8fa8\u7387\u548c\u6709\u9650 GPU \u8d44\u6e90\u4e0b&#xff0c;\u5377\u79ef\u67b6\u6784\u4ecd\u53ef\u80fd\u63d0\u4f9b\u975e\u5e38\u6709\u7ade\u4e89\u529b\u7684\u8def\u5f84\u3002<\/p>\n<p style=\"text-align:justify\">\u4ece\u5de5\u7a0b\u89d2\u5ea6\u770b&#xff0c;\u8fd9\u4e00\u70b9\u5c24\u5176\u503c\u5f97\u5173\u6ce8\u3002FCDM-XL \u80fd\u5728 4 \u5f20 RTX 4090 \u4e0a\u8bad\u7ec3&#xff0c;\u5728 512\u00d7512 \u4e0b\u4ecd\u4fdd\u6301\u660e\u663e\u541e\u5410\u4f18\u52bf&#xff1b;\u4e0e\u6b64\u540c\u65f6&#xff0c;\u5b83\u5e76\u6ca1\u6709\u4f9d\u8d56\u590d\u6742\u7684\u65b0\u8bad\u7ec3\u76ee\u6807&#xff0c;\u800c\u4e3b\u8981\u901a\u8fc7 Backbone \u8bbe\u8ba1\u83b7\u5f97\u6536\u76ca\u3002<\/p>\n<p style=\"text-align:justify\">\u6240\u4ee5\u8fd9\u7bc7\u5de5\u4f5c\u7684\u4ef7\u503c\u5e76\u4e0d\u53ea\u662f\u201cConvNeXt \u53c8\u8d62\u4e86\u4e00\u6b21\u201d&#xff0c;\u800c\u662f\u5728\u63d0\u9192\u6211\u4eec&#xff1a;<\/p>\n<p style=\"text-align:justify\">\u5f53\u4e00\u4e2a\u9886\u57df\u7684\u4e3b\u6d41\u67b6\u6784\u9010\u6e10\u6536\u655b\u5230 Transformer \u65f6&#xff0c;\u91cd\u65b0\u68c0\u67e5\u88ab\u5ffd\u7565\u7684\u5f52\u7eb3\u504f\u7f6e\u548c\u786c\u4ef6\u7279\u6027&#xff0c;\u6709\u65f6\u6bd4\u7ee7\u7eed\u5806\u8ba1\u7b97\u91cf\u66f4\u503c\u5f97\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u8bba\u6587\u9898\u76ee&#xff1a; Reviving ConvNeXt for Efficient Convolutional Diffusion Models<br \/>\n\u4e2d\u6587\u7ffb\u8bd1&#xff1a; \u91cd\u632f ConvNeXt&#xff1a;\u9762\u5411\u9ad8\u6548\u5377\u79ef\u6269\u6563\u6a21\u578b\u7684\u67b6\u6784\u8bbe\u8ba1<br \/>\n\u4f1a\u8bae&#xff1a; CVPR 2026\u6458\u8981&#xff1a; \u8fd1\u5e74\u6765\u6269\u6563\u6a21\u578b\u8d8a\u6765\u8d8a\u504f\u5411 Transformer \u4e3b\u5e72&#xff0c;\u56e0\u4e3a\u5168\u6ce8\u610f\u529b\u67b6\u6784\u8868\u73b0\u51fa\u5f88\u5f3a\u7684\u53ef\u6269\u5c55\u6027\u3002\u4f46\u5377\u79ef\u7f51\u7edc\u5177\u5907\u7684\u5c40\u90e8\u5f52\u7eb3\u504f\u7f6e\u3001\u53c2\u6570\u6548\u7387\u548c\u786c\u4ef6\u53cb\u597d\u6027&amp;#xff0<\/p>\n","protected":false},"author":2,"featured_media":113750,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[7460,2519,2520],"topic":[],"class_list":["post-113763","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-7460","tag-2519","tag-2520"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\uff08\u8bba\u6587\u901f\u8bfb\uff09FCDM\uff1aConvNeXt \u4e5f\u80fd\u505a\u9ad8\u6548\u6269\u6563\u6a21\u578b\uff0c\u5377\u79ef\u7f51\u7edc\u91cd\u65b0\u6311\u6218 DiT - \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\/113763.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\uff08\u8bba\u6587\u901f\u8bfb\uff09FCDM\uff1aConvNeXt \u4e5f\u80fd\u505a\u9ad8\u6548\u6269\u6563\u6a21\u578b\uff0c\u5377\u79ef\u7f51\u7edc\u91cd\u65b0\u6311\u6218 DiT - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u8bba\u6587\u9898\u76ee&#xff1a; Reviving ConvNeXt for Efficient Convolutional Diffusion Models \u4e2d\u6587\u7ffb\u8bd1&#xff1a; \u91cd\u632f ConvNeXt&#xff1a;\u9762\u5411\u9ad8\u6548\u5377\u79ef\u6269\u6563\u6a21\u578b\u7684\u67b6\u6784\u8bbe\u8ba1 \u4f1a\u8bae&#xff1a; CVPR 2026\u6458\u8981&#xff1a; \u8fd1\u5e74\u6765\u6269\u6563\u6a21\u578b\u8d8a\u6765\u8d8a\u504f\u5411 Transformer \u4e3b\u5e72&#xff0c;\u56e0\u4e3a\u5168\u6ce8\u610f\u529b\u67b6\u6784\u8868\u73b0\u51fa\u5f88\u5f3a\u7684\u53ef\u6269\u5c55\u6027\u3002\u4f46\u5377\u79ef\u7f51\u7edc\u5177\u5907\u7684\u5c40\u90e8\u5f52\u7eb3\u504f\u7f6e\u3001\u53c2\u6570\u6548\u7387\u548c\u786c\u4ef6\u53cb\u597d\u6027&amp;#xff0\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/113763.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-07T03:57:54+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035750-6ac5c33ec25d8.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"4 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/113763.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/113763.html\",\"name\":\"\uff08\u8bba\u6587\u901f\u8bfb\uff09FCDM\uff1aConvNeXt \u4e5f\u80fd\u505a\u9ad8\u6548\u6269\u6563\u6a21\u578b\uff0c\u5377\u79ef\u7f51\u7edc\u91cd\u65b0\u6311\u6218 DiT - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-10-07T03:57:54+00:00\",\"dateModified\":\"2026-10-07T03:57:54+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/113763.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/113763.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/113763.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\uff08\u8bba\u6587\u901f\u8bfb\uff09FCDM\uff1aConvNeXt \u4e5f\u80fd\u505a\u9ad8\u6548\u6269\u6563\u6a21\u578b\uff0c\u5377\u79ef\u7f51\u7edc\u91cd\u65b0\u6311\u6218 DiT\"}]},{\"@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":"\uff08\u8bba\u6587\u901f\u8bfb\uff09FCDM\uff1aConvNeXt \u4e5f\u80fd\u505a\u9ad8\u6548\u6269\u6563\u6a21\u578b\uff0c\u5377\u79ef\u7f51\u7edc\u91cd\u65b0\u6311\u6218 DiT - \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\/113763.html","og_locale":"zh_CN","og_type":"article","og_title":"\uff08\u8bba\u6587\u901f\u8bfb\uff09FCDM\uff1aConvNeXt \u4e5f\u80fd\u505a\u9ad8\u6548\u6269\u6563\u6a21\u578b\uff0c\u5377\u79ef\u7f51\u7edc\u91cd\u65b0\u6311\u6218 DiT - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u8bba\u6587\u9898\u76ee&#xff1a; Reviving ConvNeXt for Efficient Convolutional Diffusion Models \u4e2d\u6587\u7ffb\u8bd1&#xff1a; \u91cd\u632f ConvNeXt&#xff1a;\u9762\u5411\u9ad8\u6548\u5377\u79ef\u6269\u6563\u6a21\u578b\u7684\u67b6\u6784\u8bbe\u8ba1 \u4f1a\u8bae&#xff1a; CVPR 2026\u6458\u8981&#xff1a; \u8fd1\u5e74\u6765\u6269\u6563\u6a21\u578b\u8d8a\u6765\u8d8a\u504f\u5411 Transformer \u4e3b\u5e72&#xff0c;\u56e0\u4e3a\u5168\u6ce8\u610f\u529b\u67b6\u6784\u8868\u73b0\u51fa\u5f88\u5f3a\u7684\u53ef\u6269\u5c55\u6027\u3002\u4f46\u5377\u79ef\u7f51\u7edc\u5177\u5907\u7684\u5c40\u90e8\u5f52\u7eb3\u504f\u7f6e\u3001\u53c2\u6570\u6548\u7387\u548c\u786c\u4ef6\u53cb\u597d\u6027&amp;#xff0","og_url":"https:\/\/www.wsisp.com\/helps\/113763.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-10-07T03:57:54+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261007035750-6ac5c33ec25d8.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"4 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/113763.html","url":"https:\/\/www.wsisp.com\/helps\/113763.html","name":"\uff08\u8bba\u6587\u901f\u8bfb\uff09FCDM\uff1aConvNeXt \u4e5f\u80fd\u505a\u9ad8\u6548\u6269\u6563\u6a21\u578b\uff0c\u5377\u79ef\u7f51\u7edc\u91cd\u65b0\u6311\u6218 DiT - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-10-07T03:57:54+00:00","dateModified":"2026-10-07T03:57:54+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/113763.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/113763.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/113763.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\uff08\u8bba\u6587\u901f\u8bfb\uff09FCDM\uff1aConvNeXt \u4e5f\u80fd\u505a\u9ad8\u6548\u6269\u6563\u6a21\u578b\uff0c\u5377\u79ef\u7f51\u7edc\u91cd\u65b0\u6311\u6218 DiT"}]},{"@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\/113763","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=113763"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/113763\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/113750"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=113763"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=113763"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=113763"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=113763"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}