{"id":4893,"date":"2024-10-17T16:47:58","date_gmt":"2024-10-17T08:47:58","guid":{"rendered":"https:\/\/nullthought.net\/?p=4893"},"modified":"2024-10-17T16:48:00","modified_gmt":"2024-10-17T08:48:00","slug":"%e6%ae%8b%e5%b7%ae%e7%bd%91%e7%bb%9c%ef%bc%88resnet%ef%bc%89%e7%bb%8f%e5%85%b8%e8%ae%ba%e6%96%87","status":"publish","type":"post","link":"https:\/\/nullthought.net\/?p=4893","title":{"rendered":"\u6b8b\u5dee\u7f51\u7edc\uff08ResNet\uff09\u7ecf\u5178\u8bba\u6587"},"content":{"rendered":"\n<p>\u7ecf\u5178\u8bba\u6587<strong><a href=\"https:\/\/arxiv.org\/abs\/1512.03385\" target=\"_blank\" rel=\"noreferrer noopener\">Deep Residual Learning for Image Recognition<\/a><\/strong>\u662f\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u5177\u6709\u91cc\u7a0b\u7891\u610f\u4e49\u7684\u5de5\u4f5c\u4e4b\u4e00\u3002\u5b83\u5f15\u5165\u4e86\u4e00\u79cd\u5168\u65b0\u7684\u795e\u7ecf\u7f51\u7edc\u7ed3\u6784\u2014\u2014\u6b8b\u5dee\u7f51\u7edc\uff08ResNet\uff09\uff0c\u89e3\u51b3\u4e86\u6df1\u5c42\u795e\u7ecf\u7f51\u7edc\u4e2d\u666e\u904d\u5b58\u5728\u7684\u9000\u5316\u95ee\u9898\uff0c\u6781\u5927\u5730\u63d0\u9ad8\u4e86\u6df1\u5ea6\u7f51\u7edc\u7684\u8bad\u7ec3\u6548\u679c\u548c\u6cdb\u5316\u6027\u80fd\u3002\u5176\u6838\u5fc3\u521b\u65b0\u5728\u4e8e\u6b8b\u5dee\u5757\u548c\u6377\u5f84\u8fde\u63a5\u7684\u5f15\u5165\uff0c\u4f7f\u5f97\u6df1\u5ea6\u7f51\u7edc\u53ef\u4ee5\u66f4\u8f7b\u677e\u5730\u4f18\u5316\uff0c\u5e76\u6709\u6548\u5e94\u5bf9\u68af\u5ea6\u6d88\u5931\u95ee\u9898\u3002\u901a\u8fc7\u8be6\u7ec6\u7684\u5b9e\u9a8c\u9a8c\u8bc1\uff0c\u4f5c\u8005\u5c55\u793a\u4e86\u6b8b\u5dee\u7f51\u7edc\u5728ImageNet\u7b49\u5927\u89c4\u6a21\u6570\u636e\u96c6\u4e0a\u7684\u4f18\u8d8a\u6027\u80fd\uff0c\u63a8\u52a8\u4e86\u6df1\u5ea6\u5b66\u4e60\u5728\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u7684\u8fdb\u4e00\u6b65\u53d1\u5c55\u3002\u8fd9\u4e00\u521b\u65b0\u6027\u5de5\u4f5c\u4e0d\u4ec5\u4e3a\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7684\u8bbe\u8ba1\u63d0\u4f9b\u4e86\u65b0\u7684\u601d\u8def\uff0c\u4e5f\u4e3a\u540e\u7eed\u7684\u5404\u79cd\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u548c\u6280\u672f\u5960\u5b9a\u4e86\u57fa\u7840\u3002<strong>\u6b64\u8bba\u6587\u7684\u7814\u7a76\u6210\u679c\u88ab\u540e\u7eed\u7814\u7a76\u5927\u91cf\u5f15\u7528<\/strong>\u3002<\/p>\n\n\n\n<p>\u8bba\u6587\u4f5c\u8005\u4e3aKaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun\uff08\u4f55\u607a\u660e\u3001\u5f20\u7fd4\u5b87\u3001\u4efb\u5c11\u537f\u548c\u5b59\u5251\uff09\uff0c\u6765\u81eaMicrosoft Research\uff08\u5fae\u8f6f\u7814\u7a76\u9662\uff09\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"327\" height=\"169\" src=\"https:\/\/nullthought.net\/wp-content\/uploads\/2024\/10\/image-12.png\" alt=\"\" class=\"wp-image-4894\" srcset=\"https:\/\/nullthought.net\/wp-content\/uploads\/2024\/10\/image-12.png 327w, https:\/\/nullthought.net\/wp-content\/uploads\/2024\/10\/image-12-300x155.png 300w\" sizes=\"auto, (max-width: 327px) 100vw, 327px\" \/><figcaption class=\"wp-element-caption\"><strong><a href=\"https:\/\/arxiv.org\/abs\/1512.03385\" target=\"_blank\" rel=\"noreferrer noopener\">Deep Residual Learning for Image Recognition<\/a><\/strong><\/figcaption><\/figure>\n\n\n\n<p>\u4ee5\u4e0b\u4e3a\u8bba\u6587\u5185\u5bb9\u4ecb\u7ecd\uff1a<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">1. \u80cc\u666f\u4e0e\u6311\u6218<\/h5>\n\n\n\n<p>\u8fd1\u5e74\u6765\uff0c\u6df1\u5ea6\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\uff08Convolutional Neural Networks, CNNs\uff09\u5728\u56fe\u50cf\u5206\u7c7b\u3001\u76ee\u6807\u68c0\u6d4b\u7b49\u9886\u57df\u53d6\u5f97\u4e86\u5de8\u5927\u6210\u529f\uff0c\u4f8b\u5982AlexNet\u3001VGG\u7b49\u6a21\u578b\u90fd\u5c55\u793a\u4e86\u6df1\u5ea6\u5b66\u4e60\u5728\u89c6\u89c9\u4efb\u52a1\u4e2d\u7684\u5f3a\u5927\u8868\u73b0\u3002\u7136\u800c\uff0c\u968f\u7740\u795e\u7ecf\u7f51\u7edc\u7684\u6df1\u5ea6\u4e0d\u65ad\u589e\u52a0\uff0c\u7814\u7a76\u4eba\u5458\u53d1\u73b0\uff0c\u7f51\u7edc\u7684\u6027\u80fd\u5e76\u6ca1\u6709\u968f\u6df1\u5ea6\u7684\u589e\u52a0\u800c\u4e00\u76f4\u63d0\u9ad8\uff0c\u53cd\u800c\u53ef\u80fd\u51fa\u73b0\u201c\u9000\u5316\u95ee\u9898\u201d\uff08degradation problem\uff09\u3002\u8fd9\u79cd\u9000\u5316\u8868\u73b0\u4e3a\uff1a\u5728\u7f51\u7edc\u6df1\u5ea6\u589e\u52a0\u65f6\uff0c\u8bad\u7ec3\u8bef\u5dee\u548c\u6d4b\u8bd5\u8bef\u5dee\u4e0d\u964d\u53cd\u5347\u3002\u8fd9\u4e00\u73b0\u8c61\u4e0e\u68af\u5ea6\u6d88\u5931\u6216\u68af\u5ea6\u7206\u70b8\u4e0d\u540c\uff0c\u751a\u81f3\u5728\u6279\u5f52\u4e00\u5316\uff08Batch Normalization, BN\uff09\u6280\u672f\u7684\u5e2e\u52a9\u4e0b\uff0c\u68af\u5ea6\u7684\u524d\u5411\u4f20\u64ad\u548c\u53cd\u5411\u4f20\u64ad\u90fd\u53ef\u4ee5\u4fdd\u6301\u5065\u5eb7\uff0c\u4f46\u6a21\u578b\u4f9d\u7136\u65e0\u6cd5\u6709\u6548\u6536\u655b\u3002<\/p>\n\n\n\n<p>\u6df1\u5ea6\u7f51\u7edc\u4e2d\u7684\u9000\u5316\u73b0\u8c61\u8868\u660e\uff0c\u73b0\u6709\u4f18\u5316\u7b97\u6cd5\u5728\u9762\u5bf9\u66f4\u6df1\u7684\u7f51\u7edc\u65f6\uff0c\u65e0\u6cd5\u627e\u5230\u4e00\u4e2a\u8f83\u597d\u7684\u89e3\uff0c\u8fd9\u4f7f\u5f97\u589e\u52a0\u7f51\u7edc\u6df1\u5ea6\u6210\u4e3a\u4e00\u4e2a\u91cd\u5927\u6311\u6218\u3002<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">2. \u6b8b\u5dee\u5b66\u4e60\u7684\u6838\u5fc3\u601d\u60f3<\/h5>\n\n\n\n<p>\u4e3a\u4e86\u5e94\u5bf9\u4e0a\u8ff0\u6311\u6218\uff0c\u4f5c\u8005\u63d0\u51fa\u4e86<strong>\u6b8b\u5dee\u5b66\u4e60\u6846\u67b6\uff08Residual Learning Framework\uff09<\/strong>\uff0c\u5176\u6838\u5fc3\u601d\u60f3\u662f\u901a\u8fc7\u5b66\u4e60\u201c\u6b8b\u5dee\u6620\u5c04\u201d\uff08Residual Mapping\uff09\u800c\u975e\u76f4\u63a5\u5b66\u4e60\u590d\u6742\u7684\u6620\u5c04\uff0c\u4ece\u800c\u964d\u4f4e\u6df1\u5c42\u7f51\u7edc\u7684\u5b66\u4e60\u96be\u5ea6\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u76ee\u6807\u6620\u5c04\u4e0e\u6b8b\u5dee\u6620\u5c04<\/strong>\uff1a\u5047\u8bbe\u6211\u4eec\u6709\u4e00\u4e2a\u8f93\u5165 x\uff0c\u76ee\u6807\u662f\u5b66\u5f97\u4e00\u4e2a\u6620\u5c04 H(x)\u3002\u5728\u4f20\u7edf\u7f51\u7edc\u4e2d\uff0c\u6211\u4eec\u5e0c\u671b\u5806\u53e0\u7684\u7f51\u7edc\u5c42\u76f4\u63a5\u5b66\u4e60\u5230\u8fd9\u4e2a\u6620\u5c04 H(x)\u3002\u4f46\u5728\u6b8b\u5dee\u7f51\u7edc\u4e2d\uff0c\u4f5c\u8005\u63d0\u51fa\u6539\u4e3a\u5b66\u4e60\u4e00\u4e2a\u6b8b\u5dee\u51fd\u6570 F(x)=H(x)\u2212x\uff0c\u8fd9\u6837\u539f\u672c\u7684\u6620\u5c04\u5c31\u53ef\u4ee5\u88ab\u8868\u793a\u4e3a H(x)=F(x)+x\u3002\u901a\u8fc7\u8fd9\u79cd\u91cd\u6784\uff0c\u7f51\u7edc\u53ea\u9700\u8981\u5b66\u4e60\u5982\u4f55\u8c03\u6574\u8f93\u5165\u4ee5\u5f97\u5230\u76ee\u6807\u8f93\u51fa\uff0c\u8fd9\u6bd4\u76f4\u63a5\u5b66\u4e60\u590d\u6742\u7684\u975e\u7ebf\u6027\u6620\u5c04\u66f4\u5bb9\u6613\u4f18\u5316\u3002<\/li>\n\n\n\n<li><strong>\u6377\u5f84\u8fde\u63a5\uff08Shortcut Connections\uff09<\/strong>\uff1a\u5728\u6b8b\u5dee\u5b66\u4e60\u4e2d\uff0c\u6377\u5f84\u8fde\u63a5\uff08\u4e5f\u79f0\u4e3a\u8df3\u8dc3\u8fde\u63a5\uff0cskip connections\uff09\u662f\u81f3\u5173\u91cd\u8981\u7684\u7ed3\u6784\uff0c\u5b83\u4f7f\u5f97\u8f93\u5165\u76f4\u63a5\u7ed5\u8fc7\u82e5\u5e72\u5c42\u5230\u8fbe\u8f93\u51fa\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u6b8b\u5dee\u5757\u901a\u8fc7\u6377\u5f84\u8fde\u63a5\u5c06\u8f93\u5165\u76f4\u63a5\u6dfb\u52a0\u5230\u540e\u7eed\u51e0\u5c42\u7684\u8f93\u51fa\u4e2d\uff0c\u5f62\u5f0f\u4e3a\uff1ay=F(x,{Wi})+x\uff0c\u5176\u4e2d\uff0cF(x,{Wi})\u8868\u793a\u51e0\u5c42\u5377\u79ef\u3001\u6279\u5f52\u4e00\u5316\u548c\u6fc0\u6d3b\u7684\u7ec4\u5408\uff0cx \u901a\u8fc7\u6377\u5f84\u8fde\u63a5\u76f4\u63a5\u52a0\u5165\u5230\u8f93\u51fa\u4e2d\uff0c\u8fd9\u6837\u7684\u7ed3\u6784\u4f7f\u5f97\u4fe1\u606f\u6d41\u80fd\u591f\u76f4\u63a5\u4ece\u8f93\u5165\u4f20\u9012\u5230\u8f93\u51fa\uff0c\u5373\u4f7f\u4e2d\u95f4\u7684\u5c42\u5b66\u5f97\u4e0d\u591f\u597d\uff0c\u8f93\u5165\u4fe1\u606f\u4e5f\u4e0d\u4f1a\u4e22\u5931\u3002<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">3. \u6b8b\u5dee\u5757\u4e0e\u7f51\u7edc\u67b6\u6784<\/h5>\n\n\n\n<p>\u8bba\u6587\u4e2d\u8be6\u7ec6\u63cf\u8ff0\u4e86<strong>\u6b8b\u5dee\u5757\uff08Residual Block\uff09<\/strong>\u7684\u8bbe\u8ba1\u53ca\u5176\u5728\u6df1\u5ea6\u7f51\u7edc\u4e2d\u7684\u5e94\u7528\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u6b8b\u5dee\u5757\u7684\u8bbe\u8ba1<\/strong>\uff1a\u6bcf\u4e2a\u6b8b\u5dee\u5757\u5305\u542b\u4e24\u5c42\u5377\u79ef\u64cd\u4f5c\uff0c\u6bcf\u4e00\u5c42\u540e\u9762\u90fd\u8ddf\u6709\u6279\u5f52\u4e00\u5316\uff08Batch Normalization\uff09\u548cReLU\u6fc0\u6d3b\u51fd\u6570\u3002\u8f93\u5165\u901a\u8fc7\u6377\u5f84\u8fde\u63a5\u76f4\u63a5\u52a0\u5230\u6700\u540e\u4e00\u5c42\u7684\u8f93\u51fa\u4e0a\u3002\u5728\u8fd9\u79cd\u8bbe\u8ba1\u4e2d\uff0c\u6377\u5f84\u8fde\u63a5\u662f\u8eab\u4efd\u6620\u5c04\uff08identity mapping\uff09\uff0c\u8fd9\u610f\u5473\u7740\u5b83\u4e0d\u4f1a\u5f15\u5165\u4efb\u4f55\u989d\u5916\u7684\u53c2\u6570\u6216\u8ba1\u7b97\u590d\u6742\u5ea6\u3002<\/li>\n\n\n\n<li><strong>\u6b8b\u5dee\u7f51\u7edc\u7684\u67b6\u6784<\/strong>\uff1a\u4f5c\u8005\u5728ImageNet\u6570\u636e\u96c6\u4e0a\u6d4b\u8bd5\u4e86\u4e0d\u540c\u6df1\u5ea6\u7684\u6b8b\u5dee\u7f51\u7edc\uff0c\u4ece18\u5c42\u5230152\u5c42\u3002\u4e0e\u4f20\u7edf\u7684VGG\u7f51\u7edc\u76f8\u6bd4\uff0c\u6b8b\u5dee\u7f51\u7edc\u5177\u6709\u66f4\u6df1\u7684\u5c42\u6b21\uff08\u6700\u591a152\u5c42\uff09\uff0c\u4f46\u53c2\u6570\u91cf\u548c\u8ba1\u7b97\u590d\u6742\u5ea6\u5374\u76f8\u5bf9\u8f83\u4f4e\u3002\u8fd9\u4e9b\u7f51\u7edc\u7684\u4e3b\u8981\u7ed3\u6784\u5305\u62ec\u4e00\u4e2a7\u00d77\u5377\u79ef\u5c42\u4f5c\u4e3a\u521d\u59cb\u5377\u79ef\u5c42\uff0c\u540e\u7eed\u4e3a\u4e00\u7cfb\u5217\u7684\u6b8b\u5dee\u5757\uff0c\u6700\u540e\u662f\u5168\u5c40\u5e73\u5747\u6c60\u5316\u548c\u5168\u8fde\u63a5\u5c42\u8f93\u51fa\u5206\u7c7b\u7ed3\u679c\u3002<\/li>\n\n\n\n<li><strong>\u74f6\u9888\u8bbe\u8ba1\uff08Bottleneck Design\uff09<\/strong>\uff1a\u5728\u66f4\u6df1\u7684\u7f51\u7edc\uff08\u598250\u5c42\u3001101\u5c42\u548c152\u5c42\uff09\u4e2d\uff0c\u4f5c\u8005\u5f15\u5165\u4e86\u74f6\u9888\u8bbe\u8ba1\u6765\u51cf\u5c11\u8ba1\u7b97\u5f00\u9500\u3002\u74f6\u9888\u5757\u5305\u542b\u4e09\u5c42\u5377\u79ef\uff1a1\u00d71\u5377\u79ef\u7528\u4e8e\u964d\u7ef4\uff0c3\u00d73\u5377\u79ef\u7528\u4e8e\u7279\u5f81\u63d0\u53d6\uff0c\u518d\u901a\u8fc71\u00d71\u5377\u79ef\u6062\u590d\u7ef4\u5ea6\u3002\u8fd9\u79cd\u8bbe\u8ba1\u6781\u5927\u5730\u51cf\u5c11\u4e86\u53c2\u6570\u91cf\uff0c\u540c\u65f6\u4fdd\u7559\u4e86\u6df1\u5c42\u7f51\u7edc\u7684\u975e\u7ebf\u6027\u80fd\u529b\u3002<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">4. \u5b9e\u9a8c\u4e0e\u7ed3\u679c\u5206\u6790<\/h5>\n\n\n\n<p>\u4f5c\u8005\u901a\u8fc7\u4e00\u7cfb\u5217\u5b9e\u9a8c\u9a8c\u8bc1\u4e86\u6b8b\u5dee\u7f51\u7edc\u5728\u6df1\u5ea6\u589e\u52a0\u65f6\u7684\u4f18\u8d8a\u6027\u80fd\uff0c\u4e3b\u8981\u5305\u62ecImageNet\u548cCIFAR-10\u4e24\u4e2a\u6570\u636e\u96c6\u7684\u5b9e\u9a8c\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>ImageNet\u6570\u636e\u96c6<\/strong>\uff1a\n<ul class=\"wp-block-list\">\n<li>\u4f5c\u8005\u5728ImageNet 2012\u5206\u7c7b\u4efb\u52a1\u4e0a\u4f7f\u7528\u4e86\u591a\u79cd\u6df1\u5ea6\u7684\u6b8b\u5dee\u7f51\u7edc\uff0c\u5e76\u4e0e\u666e\u901a\u7f51\u7edc\u8fdb\u884c\u5bf9\u6bd4\u3002\u7ed3\u679c\u8868\u660e\uff0c\u968f\u7740\u7f51\u7edc\u6df1\u5ea6\u4ece18\u5c42\u589e\u52a0\u5230152\u5c42\uff0c\u6b8b\u5dee\u7f51\u7edc\u7684\u9519\u8bef\u7387\u6301\u7eed\u964d\u4f4e\u3002\u5728\u6d4b\u8bd5\u96c6\u4e0a\u7684Top-5\u9519\u8bef\u7387\u4e3a3.57%\uff0c\u663e\u8457\u4f18\u4e8e\u4e4b\u524d\u7684\u7f51\u7edc\u6a21\u578b\uff0c\u5982VGG\u548cGoogLeNet\u3002<\/li>\n\n\n\n<li>\u5728\u6df1\u5ea6\u4e3a34\u5c42\u65f6\uff0c\u6b8b\u5dee\u7f51\u7edc\u4e0e\u666e\u901a\u7f51\u7edc\u76f8\u6bd4\uff0c\u5177\u6709\u663e\u8457\u66f4\u4f4e\u7684\u8bad\u7ec3\u8bef\u5dee\u548c\u6d4b\u8bd5\u8bef\u5dee\uff0c\u8bc1\u660e\u4e86\u6b8b\u5dee\u5b66\u4e60\u80fd\u591f\u6709\u6548\u5e94\u5bf9\u6df1\u5c42\u7f51\u7edc\u7684\u9000\u5316\u95ee\u9898\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>CIFAR-10\u6570\u636e\u96c6<\/strong>\uff1a\n<ul class=\"wp-block-list\">\n<li>\u5728CIFAR-10\u6570\u636e\u96c6\u4e0a\uff0c\u4f5c\u8005\u6784\u5efa\u4e86\u4ece20\u5c42\u5230110\u5c42\u7684\u6b8b\u5dee\u7f51\u7edc\uff0c\u5e76\u8fdb\u4e00\u6b65\u63a2\u7d22\u4e861202\u5c42\u7684\u8d85\u6df1\u7f51\u7edc\u3002\u7ed3\u679c\u8868\u660e\uff0c\u6b8b\u5dee\u7f51\u7edc\u80fd\u591f\u6709\u6548\u6536\u655b\uff0c\u5e76\u4e14\u968f\u7740\u6df1\u5ea6\u7684\u589e\u52a0\uff0c\u7f51\u7edc\u7684\u6027\u80fd\u4e0d\u65ad\u63d0\u5347\u3002<\/li>\n\n\n\n<li>\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u5c3d\u7ba11202\u5c42\u7684\u7f51\u7edc\u5728\u8bad\u7ec3\u8bef\u5dee\u4e0a\u51e0\u4e4e\u4e3a\u96f6\uff0c\u4f46\u5728\u6d4b\u8bd5\u96c6\u4e0a\u7684\u8bef\u5dee\u7565\u9ad8\u4e8e110\u5c42\u7684\u7f51\u7edc\uff0c\u8fd9\u8868\u660e\u7f51\u7edc\u8fc7\u6df1\u53ef\u80fd\u5bfc\u81f4\u4e00\u5b9a\u7a0b\u5ea6\u7684\u8fc7\u62df\u5408\u3002\u56e0\u6b64\uff0c\u4f5c\u8005\u5efa\u8bae\u5728\u67d0\u4e9b\u5c0f\u578b\u6570\u636e\u96c6\u4e0a\uff0c\u8fc7\u6df1\u7684\u7f51\u7edc\u53ef\u80fd\u9700\u8981\u7ed3\u5408\u66f4\u5f3a\u7684\u6b63\u5219\u5316\u624b\u6bb5\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>\u5176\u4ed6\u89c6\u89c9\u4efb\u52a1<\/strong>\uff1a\n<ul class=\"wp-block-list\">\n<li>\u4f5c\u8005\u8fd8\u5728COCO\u548cPASCAL VOC\u7b49\u6570\u636e\u96c6\u4e0a\u9a8c\u8bc1\u4e86\u6b8b\u5dee\u7f51\u7edc\u5728\u76ee\u6807\u68c0\u6d4b\u4efb\u52a1\u4e2d\u7684\u6709\u6548\u6027\u3002\u5728\u8fd9\u4e9b\u4efb\u52a1\u4e2d\uff0c\u6b8b\u5dee\u7f51\u7edc\u901a\u8fc7\u66f4\u597d\u7684\u7279\u5f81\u8868\u793a\uff0c\u663e\u8457\u63d0\u5347\u4e86\u76ee\u6807\u68c0\u6d4b\u548c\u5b9a\u4f4d\u7684\u7cbe\u5ea6\u3002\u4f8b\u5982\uff0c\u5728COCO\u6570\u636e\u96c6\u4e0a\uff0c\u4f7f\u7528ResNet-101\u7684\u76ee\u6807\u68c0\u6d4b\u6a21\u578b\u5728mAP\uff08Mean Average Precision\uff09\u6307\u6807\u4e0a\u76f8\u5bf9\u4e8eVGG-16\u63d0\u9ad8\u4e8628%\u7684\u76f8\u5bf9\u589e\u76ca\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">5. \u4f18\u52bf\u4e0e\u6027\u80fd\u63d0\u5347\u7684\u539f\u56e0\u5206\u6790<\/h5>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u4f18\u5316\u7684\u7b80\u5316<\/strong>\uff1a\u6b8b\u5dee\u5b66\u4e60\u7684\u4e00\u4e2a\u5173\u952e\u4f18\u52bf\u662f\uff0c\u5b83\u5c06\u539f\u672c\u590d\u6742\u7684\u975e\u7ebf\u6027\u6620\u5c04\u5206\u89e3\u4e3a\u5bf9\u8f93\u5165\u7684\u7ec6\u5fae\u8c03\u6574\uff0c\u8fd9\u4f7f\u5f97\u7f51\u7edc\u66f4\u5bb9\u6613\u8fdb\u884c\u4f18\u5316\u3002\u7279\u522b\u662f\u5f53\u7406\u60f3\u6620\u5c04\u63a5\u8fd1\u4e8e\u8eab\u4efd\u6620\u5c04\u65f6\uff0c\u6b8b\u5dee\u5757\u53ea\u9700\u5c06\u5176\u8f93\u51fa\u5c3d\u91cf\u9760\u8fd1\u96f6\uff0c\u4ece\u800c\u4f7f\u5f97\u8bad\u7ec3\u66f4\u4e3a\u7b80\u5355\u3002<\/li>\n\n\n\n<li><strong>\u68af\u5ea6\u7684\u6709\u6548\u4f20\u64ad<\/strong>\uff1a\u6377\u5f84\u8fde\u63a5\u4f7f\u5f97\u524d\u5411\u548c\u53cd\u5411\u4f20\u64ad\u65f6\uff0c\u68af\u5ea6\u53ef\u4ee5\u7ed5\u8fc7\u4e2d\u95f4\u5c42\u76f4\u63a5\u6d41\u5411\u524d\u540e\u5c42\uff0c\u8fd9\u6781\u5927\u5730\u7f13\u89e3\u4e86\u6df1\u5ea6\u7f51\u7edc\u4e2d\u5e38\u89c1\u7684\u68af\u5ea6\u6d88\u5931\u95ee\u9898\uff0c\u4f7f\u5f97\u6df1\u5c42\u7f51\u7edc\u4e5f\u80fd\u9ad8\u6548\u8bad\u7ec3\u3002<\/li>\n\n\n\n<li><strong>\u53c2\u6570\u6548\u7387\u4e0e\u8ba1\u7b97\u590d\u6742\u5ea6<\/strong>\uff1a\u6b8b\u5dee\u7f51\u7edc\u5728\u6df1\u5ea6\u589e\u52a0\u7684\u540c\u65f6\uff0c\u5e76\u672a\u663e\u8457\u589e\u52a0\u53c2\u6570\u91cf\u548c\u8ba1\u7b97\u590d\u6742\u5ea6\u3002\u901a\u8fc7\u5f15\u5165\u74f6\u9888\u8bbe\u8ba1\uff0c\u6b8b\u5dee\u7f51\u7edc\u5728\u4fdd\u6301\u9ad8\u6548\u975e\u7ebf\u6027\u7279\u5f81\u8868\u8fbe\u7684\u540c\u65f6\uff0c\u663e\u8457\u51cf\u5c11\u4e86\u8ba1\u7b97\u91cf\u3002\u4f8b\u5982\uff0cResNet-152\u7684\u8ba1\u7b97\u590d\u6742\u5ea6\u4f4e\u4e8eVGG-19\uff0c\u4f46\u5728\u5206\u7c7b\u4efb\u52a1\u4e0a\u7684\u8868\u73b0\u66f4\u597d\u3002<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">6. \u6b8b\u5dee\u5b66\u4e60\u7684\u5f71\u54cd\u4e0e\u5e94\u7528<\/h5>\n\n\n\n<p>\u6b8b\u5dee\u7f51\u7edc\u7684\u5f15\u5165\u4e3a\u6df1\u5ea6\u5b66\u4e60\u7279\u522b\u662f\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u5e26\u6765\u4e86\u6df1\u8fdc\u5f71\u54cd\u3002\u5b83\u7684\u63d0\u51fa\u4e0d\u4ec5\u89e3\u51b3\u4e86\u6df1\u5ea6\u7f51\u7edc\u7684\u4f18\u5316\u96be\u9898\uff0c\u8fd8\u4e3a\u540e\u7eed\u7684\u7f51\u7edc\u8bbe\u8ba1\u63d0\u4f9b\u4e86\u4e00\u4e2a\u65b0\u7684\u65b9\u5411\u548c\u7075\u611f\u3002\u5176\u540e\u7684\u8bb8\u591a\u5de5\u4f5c\uff0c\u5305\u62ec<strong>DenseNet<\/strong>\u3001<strong>ResNeXt<\/strong>\u7b49\uff0c\u90fd\u662f\u5728\u6b8b\u5dee\u5b66\u4e60\u7684\u57fa\u7840\u4e0a\u8fdb\u4e00\u6b65\u6539\u8fdb\u548c\u6269\u5c55\u7684\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u5e94\u7528\u9886\u57df<\/strong>\uff1a\u6b8b\u5dee\u7f51\u7edc\u88ab\u5e7f\u6cdb\u5e94\u7528\u4e8e\u5404\u79cd\u89c6\u89c9\u4efb\u52a1\u4e2d\uff0c\u5305\u62ec\u56fe\u50cf\u5206\u7c7b\u3001\u76ee\u6807\u68c0\u6d4b\u3001\u8bed\u4e49\u5206\u5272\u7b49\u3002\u5728\u5de5\u4e1a\u754c\uff0cResNet\u5df2\u6210\u4e3a\u4f17\u591a\u8ba1\u7b97\u673a\u89c6\u89c9\u5e94\u7528\u7684\u57fa\u77f3\uff0c\u5c24\u5176\u662f\u5728\u5927\u89c4\u6a21\u56fe\u50cf\u8bc6\u522b\u548c\u89c6\u9891\u5206\u6790\u4efb\u52a1\u4e2d\uff0c\u5b83\u7684\u6027\u80fd\u548c\u7a33\u5b9a\u6027\u4f7f\u5176\u6210\u4e3a\u9996\u9009\u6a21\u578b\u3002<\/li>\n\n\n\n<li><strong>\u6cdb\u5316\u80fd\u529b<\/strong>\uff1a\u6b8b\u5dee\u5b66\u4e60\u7684\u601d\u60f3\u5e76\u4e0d\u4ec5\u9650\u4e8e\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\uff0c\u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\u3001\u8bed\u97f3\u8bc6\u522b\u7b49\u5176\u4ed6\u9886\u57df\u4e5f\u6709\u5e94\u7528\u3002\u5176\u6838\u5fc3\u601d\u60f3\u662f\u5c06\u6df1\u5ea6\u6a21\u578b\u7684\u5b66\u4e60\u4efb\u52a1\u5206\u89e3\u4e3a\u76f8\u5bf9\u7b80\u5355\u7684\u6b8b\u5dee\u6620\u5c04\uff0c\u8fd9\u79cd\u65b9\u6cd5\u53ef\u4ee5\u6709\u6548\u5e94\u5bf9\u6df1\u5c42\u7ed3\u6784\u4e2d\u7684\u4f18\u5316\u56f0\u96be\uff0c\u4f7f\u5f97\u66f4\u590d\u6742\u7684\u4efb\u52a1\u80fd\u591f\u901a\u8fc7\u8f83\u6df1\u7684\u7f51\u7edc\u6765\u5904\u7406\u3002<\/li>\n<\/ul>\n\n\n\n<p><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u7ecf\u5178\u8bba\u6587Deep Residual Learning for Image Recognition\u662f\u8ba1\u7b97\u673a\u89c6\u89c9\u9886 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[8],"tags":[39,79,80],"class_list":["post-4893","post","type-post","status-publish","format-standard","hentry","category-tech","tag-ai","tag-microsoft","tag-cv"],"rttpg_featured_image_url":null,"rttpg_author":{"display_name":"NullThought","author_link":"https:\/\/nullthought.net\/?author=1"},"rttpg_comment":0,"rttpg_category":"<a href=\"https:\/\/nullthought.net\/?cat=8\" rel=\"category\">Tech<\/a>","rttpg_excerpt":"\u7ecf\u5178\u8bba\u6587Deep Residual Learning for Image Recognition\u662f\u8ba1\u7b97\u673a\u89c6\u89c9\u9886&hellip;","_links":{"self":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4893","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=4893"}],"version-history":[{"count":1,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4893\/revisions"}],"predecessor-version":[{"id":4895,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4893\/revisions\/4895"}],"wp:attachment":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4893"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4893"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4893"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}