{"id":6501,"date":"2022-08-28T19:41:53","date_gmt":"2022-08-28T11:41:53","guid":{"rendered":"http:\/\/139.9.1.231\/?p=6501"},"modified":"2022-08-28T19:41:55","modified_gmt":"2022-08-28T11:41:55","slug":"fcn","status":"publish","type":"post","link":"http:\/\/139.9.1.231\/index.php\/2022\/08\/28\/fcn\/","title":{"rendered":"FCN\u5168\u5377\u79ef\u7f51\u7edc–\u56fe\u50cf\u5206\u5272\u7684\u5f00\u5c71\u4e4b\u4f5c"},"content":{"rendered":"\n
\"\"<\/figure>\n\n\n\n

\u8bba\u6587\u5730\u5740\uff1a https:\/\/arxiv.org\/abs\/1411.4038<\/a><\/p>\n\n\n\n

\u968f\u7740CNN\u5728\u56fe\u50cf\u8bc6\u522b\u4e2d\u53d6\u5f97\u5de8\u5927\u6210\u529f\uff0c\u4e00\u4e9b\u7ecf\u5178\u7684\u56fe\u50cf\u5206\u7c7b\u7f51\u7edc\uff08AlexNet\u3001VGG\u3001GoogLeNet\u3001ResNet\uff09\u4e5f\u9010\u6e10\u88ab\u5e94\u7528\u4e8e\u66f4\u52a0\u7ec6\u5206\u7684\u89c6\u89c9\u4efb\u52a1\u4e2d\u3002\u5f88\u591a\u7814\u7a76\u8005\u4e5f\u5728\u63a2\u7d22\u5982\u4f55\u5c06\u5206\u7c7b\u7f51\u7edc\u8fdb\u884c\u6539\u9020\u540e\u7528\u4e8e\u8bed\u4e49\u5206\u5272\u7684\u5bc6\u96c6\u9884\u6d4b\u95ee\u9898\uff08dense predictions\uff09\u3002\u5728\u66f4\u9ad8\u6548\u7684\u8bed\u4e49\u5206\u5272\u7f51\u7edc\u63d0\u51fa\u4e4b\u524d\uff0c\u5b66\u672f\u754c\u7528\u4e8e\u5bc6\u96c6\u9884\u6d4b\u4efb\u52a1\u7684\u6a21\u578b\u4e3b\u8981\u6709\u4ee5\u4e0b\u51e0\u4e2a\u7279\u70b9\uff1a<\/p>\n\n\n\n

\uff081\uff09\u5c0f\u6a21\u578b\u3002\u65e9\u671f\u7684\u7f51\u7edc\u7ed3\u6784\u53d7\u9650\u4e8e\u6570\u636e\u91cf\u548c\u9ad8\u6027\u80fd\u7684\u8ba1\u7b97\u8d44\u6e90\uff0c\u5728\u8bbe\u8ba1\u4e0a\u4e00\u822c\u4e0d\u4f1a\u4f7f\u7528\u8fc7\u5927\u7684\u6a21\u578b\u3002<\/p>\n\n\n\n

\uff082\uff09\u5206\u5757\u8bad\u7ec3\u3002\u5206\u5757\u8bad\u7ec3\uff08patchwise training\uff09\u5728\u5f53\u65f6\u662f\u56fe\u50cf\u8bad\u7ec3\u7684\u666e\u904d\u505a\u6cd5\uff0c\u4f46\u8be5\u65b9\u6cd5\u5bf9\u4e8e\u5168\u5377\u79ef\u7f51\u7edc\u7684\u8bad\u7ec3\u4f1a\u663e\u5f97\u76f8\u5bf9\u4f4e\u6548\uff0c\u4f46\u5206\u5757\u8bad\u7ec3\u7684\u4f18\u70b9\u5728\u4e8e\u80fd\u591f\u89c4\u907f\u7c7b\u522b\u4e0d\u5747\u8861\u95ee\u9898\uff0c\u5e76\u4e14\u80fd\u591f\u7f13\u89e3\u5bc6\u96c6\u5206\u5757\u7684\u7a7a\u95f4\u76f8\u5173\u6027\u95ee\u9898\u3002<\/p>\n\n\n\n

\uff083\uff09\u8f93\u5165\u79fb\u4f4d\u4e0e\u8f93\u51fa\u4ea4\u9519\u3002\u8be5\u65b9\u6cd5\u53ef\u4ee5\u89c6\u4e3a\u4e00\u79cd\u8f93\u5165\u4e0e\u8f93\u51fa\u7684\u53d8\u6362\u65b9\u6cd5\uff0c\u5728OverFeat\u7b49\u7ed3\u6784\u4e2d\u88ab\u5e7f\u6cdb\u4f7f\u7528\u3002<\/p>\n\n\n\n

\uff084\uff09\u540e\u5904\u7406\u3002\u5bf9\u4e8e\u795e\u7ecf\u7f51\u7edc\u8f93\u51fa\u8d28\u91cf\u4e0d\u9ad8\u7684\u95ee\u9898\uff0c\u5bf9\u8f93\u51fa\u52a0\u540e\u5904\u7406\u4e5f\u662f\u5e38\u89c1\u505a\u6cd5\uff0c\u5e38\u7528\u7684\u540e\u5904\u7406\u65b9\u6cd5\u5305\u62ec\u8d85\u50cf\u7d20\u6295\u5f71\uff08superpixel projection\uff09\u3001\u968f\u673a\u573a\u6b63\u5219\u5316\uff08random field regularization\uff09\u548c\u56fe\u50cf\u6ee4\u6ce2\u5904\u7406\u7b49\u3002<\/p>\n\n\n\n

\u53ef\u4ee5\u770b\u5230\uff0c\u65e9\u671f\u7528\u4e8e\u76ee\u6807\u68c0\u6d4b\u3001\u5173\u952e\u70b9\u9884\u6d4b\u548c\u8bed\u4e49\u5206\u5272\u7b49\u5bc6\u96c6\u9884\u6d4b\u95ee\u9898\u6574\u4f53\u6765\u770b\u6709\u4e24\u4e2a\u7f3a\u9677\uff0c\u4e00\u662f\u65e0\u6cd5\u5b9e\u73b0\u7aef\u5230\u7aef\uff08end-to-end\uff09\u7684\u6d41\u7a0b\uff0c\u6a21\u578b\u6574\u4f53\u6548\u7387\u4e0d\u4f73\uff1b\u7b2c\u4e8c\u4e2a\u5219\u662f\u4e0d\u80fd\u505a\u5230\u771f\u6b63\u7684\u5bc6\u96c6\u9884\u6d4b\u7684\u7279\u5f81\uff1a\u50cf\u7d20\u5230\u50cf\u7d20\uff08pixels-to-pixels\uff09\u7684\u9884\u6d4b\u3002<\/p>\n\n\n\n

\u5168\u5377\u79ef\u7f51\u7edc\uff08Fully Convolutional Networks, FCN\uff09\u7684\u63d0\u51fa\uff0c\u6b63\u597d\u53ef\u4ee5\u89e3\u51b3\u65e9\u671f\u7f51\u7edc\u7ed3\u6784\u666e\u904d\u5b58\u5728\u7684\u4e0a\u8ff0\u4e24\u4e2a\u7f3a\u9677\u3002FCN\u57282015\u5e74\u7684\u4e00\u7bc7\u8bba\u6587Fully Convolutional Networks for Semantic Segmentation\u4e2d\u63d0\u51fa\uff0c\u5176\u4e3b\u8981\u601d\u8def\u5728\u4e8e\u7528\u5377\u79ef\u5c42\u4ee3\u66ff\u6b64\u524d\u5206\u7c7b\u7f51\u7edc\u4e2d\u7684\u5168\u8fde\u63a5\u5c42\uff0c\u5c06\u5168\u8fde\u63a5\u5c42\u7684\u8bed\u4e49\u6807\u7b7e\u8f93\u51fa\u6539\u4e3a\u5377\u79ef\u5c42\u7684\u8bed\u4e49\u70ed\u56fe\uff08heatmap\uff09\u8f93\u51fa\uff0c\u518d\u7ed3\u5408\u4e0a\u91c7\u6837\u6280\u672f\u5b9e\u73b0\u50cf\u7d20\u5230\u50cf\u7d20\u7684\u5bc6\u96c6\u9884\u6d4b\u3002\u5982\u4e0b\u56fe\u6240\u793a\uff0c\u4e0a\u56fe\u4e3a\u5e38\u89c1\u5206\u7c7b\u7f51\u7edc\u7684\u6d41\u7a0b\uff0c\u5728\u4e94\u5c42\u5377\u79ef\u7f51\u7edc\u4e4b\u540e\u6709\u4e09\u5c42\u5168\u8fde\u63a5\u7f51\u7edc\uff0c\u6700\u540e\u8f93\u51fa\u4e00\u4e2a\u5305\u542b\u7c7b\u522b\u8bed\u4e49\u4fe1\u606f\u7684\u8f93\u51fa\u6982\u7387\uff1b\u4e0b\u56fe\u4e3aFCN\u7f51\u7edc\u6d41\u7a0b\uff0c\u5728\u4e0a\u56fe\u5206\u7c7b\u7f51\u7edc\u7684\u57fa\u7840\u4e0a\uff0c\u5c06\u6700\u540e\u4e09\u5c42\u5168\u8fde\u63a5\u5c42\u6539\u4e3a\u5377\u79ef\u5c42\uff0c\u8f93\u51fa\u4e5f\u76f8\u5e94\u7684\u53d8\u4e3a\u5206\u7c7b\u9884\u6d4b\u7684\u70ed\u56fe\uff0c\u8fd9\u6837\u5c31\u4e3a\u4e86\u6700\u540e\u7684\u50cf\u7d20\u7ea7\u7684\u5bc6\u96c6\u9884\u6d4b\u63d0\u4f9b\u4e86\u57fa\u7840\u3002<\/p>\n\n\n\n

\"\"<\/figure><\/div>\n\n\n\n

\u6240\u4ee5\uff0cFCN\u5b9e\u73b0\u5bc6\u96c6\u9884\u6d4b\u7684\u5173\u952e\u5728\u4e8e\u4fee\u6539\u5168\u8fde\u63a5\u5c42\u4e3a\u5377\u79ef\u5c42\uff0c\u90a3\u4e48\u5177\u4f53\u662f\u5982\u4f55\u4fee\u6539\u7684\u5462\uff1f\u5148\u6765\u8be6\u7ec6\u5206\u6790\u4e00\u4e0b\u7684\u5377\u79ef\u5c42\u548c\u5168\u8fde\u63a5\u5c42\u7684\u7279\u5f81\u3002\u5377\u79ef\u5c42\u4e0e\u5168\u8fde\u63a5\u5c42\u6700\u5927\u7684\u533a\u522b\u5728\u4e8e\u5377\u79ef\u5c42\u6bcf\u6b21\u8ba1\u7b97\u65f6\u53ea\u4e0e\u8f93\u5165\u56fe\u50cf\u4e2d\u4e00\u4e2a\u5177\u4f53\u7684\u5c40\u90e8\u505a\u8fd0\u7b97\uff0c\u4f46\u4e8c\u8005\u90fd\u662f\u505a\u70b9\u79ef\u8ba1\u7b97\uff0c\u5176\u51fd\u6570\u5f62\u5f0f\u662f\u7c7b\u4f3c\u7684\u3002\u5047\u8bbe\u7ed9\u5b9a\u5728\u6307\u5b9a\u7f51\u7edc\u5c42\u4efb\u610f\u5750\u6807\u70b9(i,j)\u7684\u6570\u636e\u5411\u91cfXij\uff0c\u800c\u4e0b\u4e00\u5c42\u5bf9\u5e94\u5750\u6807\u70b9\u7684\u6570\u636e\u5411\u91cf\u4e3aYij\uff0c\u6709\uff1a<\/p>\n\n\n\n

\"\"<\/figure><\/div>\n\n\n\n

\u5176\u4e2d\u4e3a\u5377\u79ef\u6838\u5927\u5c0f\u6216\u8005\u6743\u91cd\u5411\u91cf\u957f\u5ea6\uff0cs\u4e3a\u6b65\u957f\uff08stride\uff09\uff0c\u800cf_ks\u5219\u8868\u793a\u5f53\u524d\u5c42\u5230\u4e0b\u4e00\u5c42\u7684\u6620\u5c04\u51fd\u6570\uff0cf_ks\u65e2\u53ef\u4ee5\u8868\u793a\u4e3a\u5377\u79ef\u5c42\u53c8\u53ef\u4ee5\u8868\u793a\u4e3a\u5168\u8fde\u63a5\u5c42\uff0c\u6240\u4ee5\u4e8c\u8005\u4e4b\u95f4\u7684\u8f6c\u6362\u662f\u6709\u7406\u8bba\u57fa\u7840\u7684\u3002<\/p>\n\n\n\n

FCN\u5206\u522b\u5728AlexNet\u3001VGG\u548cGoogLeNet\u4e0a\u8fdb\u884c\u4e86\u5168\u8fde\u63a5\u5c42\u8f6c\u5377\u79ef\u5c42\u7684\u4fee\u6539\uff0c\u901a\u8fc7\u5b9e\u9a8c\u53d1\u73b0\u4ee5VGG16\u4f5c\u4e3a\u4e3b\u5e72\u7f51\u7edc\u6548\u679c\u6700\u597d\uff0c\u5b8c\u6574\u7684FCN\u7ed3\u6784\u5982\u4e0b\u56fe\u6240\u793a\uff0c\u7b2c\u4e00\u884c\u6700\u5de6\u8fb9\u4e3a\u539f\u59cb\u8f93\u5165\u56fe\u50cf\uff0c\u56fe\u50cf\u5c3a\u5bf8\u4e3a32×32\uff0cconv\u4e3a\u5377\u79ef\u5c42\uff0cpool\u4e3a\u6c60\u5316\u5c42\uff0c\u53ef\u4ee5\u6ce8\u610f\u5230conv6-7\u662f\u6700\u540e\u7684\u5377\u79ef\u5c42\uff0c\u6b64\u65f6\u5f97\u5230\u7684\u5bc6\u96c6\u9884\u6d4b\u70ed\u56fe\u5c3a\u5bf8\u4e3a\u8f93\u5165\u56fe\u50cf\u76841\/32\uff0c\u4e3a\u4e86\u5b9e\u73b0\u50cf\u7d20\u5230\u50cf\u7d20\u7684\u9884\u6d4b\uff0c\u8fd8\u9700\u8981\u5bf9\u70ed\u56fe\u8fdb\u884c\u4e0a\u91c7\u6837\uff0cFCN\u91c7\u7528\u53cc\u7ebf\u6027\u63d2\u503c\uff08bilinear interpolation\uff09\u8fdb\u884c\u4e0a\u91c7\u6837\uff0c\u6240\u4ee5\u8fd9\u91cc\u9700\u8981\u5c06\u70ed\u56fe\u4e0a\u91c7\u683732\u500d\u6765\u6062\u590d\u5230\u539f\u59cb\u56fe\u50cf\u7684\u5c3a\u5bf8\uff0c\u56e0\u800c\u7b2c\u4e00\u884c\u7684\u7f51\u7edc\u7ed3\u6784\u4e5f\u53ebFCN-32s\u3002\u76f4\u63a5\u8fdb\u884c32\u500d\u4e0a\u91c7\u6837\u5f97\u5230\u7684\u8f93\u51fa\u65e0\u7591\u662f\u8f83\u4e3a\u7c97\u7cd9\u7684\uff0c\u4e3a\u4e86\u63d0\u9ad8\u50cf\u7d20\u9884\u6d4b\u8d28\u91cf\uff0cFCN\u53c8\u5206\u522b\u6709FCN-16s\u548cFCN-8s\u7684\u6539\u8fdb\u7248\u672c\u3002\u56fe\u4e2d\u7b2c\u4e8c\u884c\u5373\u4e3aFCN-16s\uff0c\u4e3b\u8981\u533a\u522b\u5728\u4e8e\u5148\u5c06conv7\uff081×1\uff09\u7684\u8f93\u51fa\u70ed\u56fe\u8fdb\u884c2\u500d\u4e0a\u91c7\u6837\uff0c\u7136\u540e\u5c06\u5176\u4e0epool4\uff082×2\uff09\u8fdb\u884c\u878d\u5408\uff0c\u6700\u540e\u5bf9\u878d\u5408\u540e\u7684\u7ed3\u679c\u8fdb\u884c16\u500d\u4e0a\u91c7\u6837\u5f97\u5230\u6700\u7ec8\u9884\u6d4b\u7ed3\u679c\uff0c\u540c\u7406FCN-8s\u5c06pool3\uff084×4\uff09\u30012\u500d\u4e0a\u91c7\u6837\u540e\u7684pool4\uff084×4\uff09\u4ee5\u53ca4\u500d\u4e0a\u91c7\u6837\u7684conv7\uff084×4\uff09\u8fdb\u884c\u878d\u5408\uff0c\u6700\u540e\u518d\u8fdb\u884c8\u500d\u7684\u4e0a\u91c7\u6837\u5f97\u5230\u8bed\u4e49\u5206\u5272\u56fe\u50cf\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u6240\u4ee5\uff0c\u4eceFCN-32s\u5230FCN-8s\uff0c\u5176\u5b9e\u4e00\u79cd\u7c97\u5206\u5272\u5230\u7cbe\u7ec6\u5206\u5272\u7684\u6f14\u53d8\u8fc7\u7a0b\uff0cFCN\u901a\u8fc7\u878d\u5408\u6d45\u5c42\u56fe\u50cf\u7279\u5f81\u548c\u6df1\u5c42\u5377\u79ef\u70ed\u56fe\u7684\u65b9\u5f0f\u6765\u5f97\u5230\u5f53\u65f6\u7684SOTA\uff08State of the art\uff09\u6c34\u5e73\u7684\u8bed\u4e49\u5206\u5272\u6a21\u578b\u3002\u4e0b\u56fe\u662fFCN-32s\u3001FCN-16s\u548cFCN-8s\u5728\u540c\u4e00\u5f20\u56fe\u50cf\u4e0a\u7684\u5206\u5272\u6548\u679c\uff0c\u4e0e\u5206\u5272\u7684\u6807\u51c6\u56fe\u50cf\uff08Ground truth\uff09\u76f8\u6bd4\uff0c\u53ef\u4ee5\u770b\u5230\u4e09\u4e2a\u6a21\u578b\u7684\u5206\u5272\u7cbe\u5ea6\u662f\u5728\u4e0d\u65ad\u4f18\u5316\u7684\u3002<\/p>\n\n\n\n

\"\"<\/figure><\/div>\n\n\n\n

\u4e0b\u65b9\u4ee3\u7801\u7ed9\u51faFCN-8s\u7684\u4e00\u4e2aPyTorch\u7b80\u7565\u5b9e\u73b0\u65b9\u5f0f\uff0c\u4fbf\u4e8e\u8bfb\u8005\u52a0\u6df1\u5bf9FCN\u7684\u7406\u89e3\u3002\u4ee3\u7801\u4e2d\u5bf9\u4e8e\u5377\u79ef\u4e0b\u91c7\u6837\u4f7f\u7528\u4e86VGG16\u7684\u9884\u8bad\u7ec3\u6743\u91cd\uff0c\u5206\u522b\u6784\u5efa\u4e86\u56db\u4e2a\u7279\u5f81\u63d0\u53d6\u6a21\u5757\u3001\u4e00\u4e2a\u5377\u79ef\u5757\u548c\u4e09\u4e2a\u72ec\u7acb\u7684\u5377\u79ef\u5c42\u3002\u5728\u524d\u5411\u4f20\u64ad\u6d41\u7a0b\u4e2d\uff0c\u5c06conv7\u3001pool3\u548cpool4\u8fdb\u884c\u878d\u5408\uff0c\u6700\u540e\u518d\u505a8\u500d\u7684\u53cc\u7ebf\u6027\u63d2\u503c\u4e0a\u91c7\u6837\u3002<\/p>\n\n\n\n

# \u5bfc\u5165PyTorch\u76f8\u5173\u6a21\u5757\r\nimport torch\r\nimport torch.nn as nn\r\nimport torch.nn.functional as F\r\nfrom torchvision import models\r\n\r\n### \u5b9a\u4e49FCN-8s\u6a21\u578b\u7c7b\r\nclass FCN8(nn.Module):\r\n    def __init__(self, num_classes):\r\n        super().__init__()\r\n        # \u63d0\u53d6VGG16\u9884\u8bad\u7ec3\u6743\u91cd\u4f5c\u4e3a\u7279\u5f81\r\n        feats =list(models.vgg16(pretrained=True).features.children())\r\n        # \u53d6\u524d9\u5c42\u4e3a\u7b2c\u4e00\u7279\u5f81\u6a21\u5757\r\n        self.feat1 = nn.Sequential(*feats[0:9])\r\n        # \u53d6\u7b2c10-15\u5c42\u4e3a\u7b2c\u4e8c\u7279\u5f81\u6a21\u5757\r\n        self.feat2 = nn.Sequential(*feats[10:16])\r\n        # \u53d6\u7b2c16-22\u5c42\u4e3a\u7b2c\u4e09\u7279\u5f81\u6a21\u5757\r\n        self.feat3 = nn.Sequential(*feats[17:23])\r\n        # \u53d6\u540e6\u5c42\u4e3a\u7b2c\u56db\u7279\u5f81\u6a21\u5757\r\n        self.feat4 = nn.Sequential(*feats[24:30])\r\n        # \u5377\u79ef\u5c42\u6743\u91cd\u4e0d\u53c2\u4e0e\u8bad\u7ec3\u66f4\u65b0\r\n        for m in self.modules():\r\n            if isinstance(m, nn.Conv2d):\r\n                m.requires_grad = False\r\n        # \u5b9a\u4e49\u5377\u79ef\u5757\r\n        self.conv_blocks = nn.Sequential(\r\n            nn.Conv2d(512, 4096, 7),\r\n            nn.ReLU(inplace=True),\r\n            nn.Dropout(),\r\n            nn.Conv2d(4096, 4096, 1),\r\n            nn.ReLU(inplace=True),\r\n            nn.Dropout(),\r\n        )\r\n        # \u6539\u6700\u540e\u4e09\u5c42\u7684\u5168\u8fde\u63a5\u5c42\u4e3a\u5377\u79ef\u5c42\r\n        self.conv1 = nn.Conv2d(256, num_classes, 1)\r\n        self.conv2 = nn.Conv2d(512, num_classes, 1)\r\n        self.conv3 = nn.Conv2d(4096, num_classes, 1)\r\n\r\n    ### \u5b9a\u4e49\u524d\u5411\u8ba1\u7b97\u6d41\u7a0b\r\n    def forward(self, x):\r\n        feat1 = self.feat1(x)\r\n        feat2 = self.feat2(feat1)\r\n        feat3 = self.feat3(feat2)\r\n        feat4 = self.feat4(feat3)\r\n        conv_blocks = self.conv_blocks(feat4)\r\n\r\n        conv1 = self.conv1(feat2)\r\n        conv2 = self.conv2(feat3)\r\n        conv3 = self.conv3(conv_blocks)      \r\n        outputs = F.upsample_bilinear(conv_blocks, conv2.size()[2:])\r\n        # \u7b2c\u4e00\u6b21\u878d\u5408\r\n        outputs += conv2\r\n        outputs = F.upsample_bilinear(outputs, conv1.size()[2:])\r\n        # \u7b2c\u4e8c\u6b21\u878d\u5408\r\n        outputs += conv1\r\n        return F.upsample_bilinear(outputs, x.size()[2:]) <\/code><\/pre>\n\n\n\n
FCN\u662f\u6df1\u5ea6\u5b66\u4e60\u8bed\u4e49\u5206\u5272\u7f51\u7edc\u7684\u5f00\u5c71\u4e4b\u4f5c\uff0c\u5728\u7ed3\u6784\u8bbe\u8ba1\u4e0a\u7387\u5148\u5c06\u5168\u5377\u79ef\u7f51\u7edc\u7528\u4e8e\u6df1\u5ea6\u5b66\u4e60\u8bed\u4e49\u5206\u5272\u4efb\u52a1\uff0c\u5728\u7ecf\u5178\u5206\u7c7b\u7f51\u7edc\u7684\u57fa\u7840\u4e0a\u5b9e\u73b0\u4e86\u50cf\u7d20\u5230\u50cf\u7d20\u548c\u7aef\u5230\u7aef\u7684\u5206\u5272\u3002FCN\u6574\u4f53\u4e0a\u5df2\u5177\u5907\u7f16\u89e3\u7801\u67b6\u6784\u7684U\u5f62\u7f51\u7edc\u96cf\u5f62\uff0c\u4e3a\u540e\u7eed\u7684\u7f51\u7edc\u8bbe\u8ba1\u5f00\u521b\u4e86\u575a\u5b9e\u7684\u57fa\u7840\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
\u8bba\u6587\u5730\u5740\uff1a https:\/\/arxiv.org\/abs\/1411.4038 \u968f\u7740CNN\u5728\u56fe\u50cf\u8bc6\u522b\u4e2d\u53d6\u5f97\u5de8\u5927\u6210\u529f … \u7ee7\u7eed\u9605\u8bfbFCN\u5168\u5377\u79ef\u7f51\u7edc–\u56fe\u50cf\u5206\u5272\u7684\u5f00\u5c71\u4e4b\u4f5c<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[24,9],"tags":[],"_links":{"self":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/6501"}],"collection":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/comments?post=6501"}],"version-history":[{"count":3,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/6501\/revisions"}],"predecessor-version":[{"id":6510,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/6501\/revisions\/6510"}],"wp:attachment":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/media?parent=6501"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/categories?post=6501"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/tags?post=6501"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}