{"id":9071,"date":"2022-10-19T19:41:00","date_gmt":"2022-10-19T11:41:00","guid":{"rendered":"http:\/\/139.9.1.231\/?p=9071"},"modified":"2022-10-13T14:39:00","modified_gmt":"2022-10-13T06:39:00","slug":"resizemodel","status":"publish","type":"post","link":"http:\/\/139.9.1.231\/index.php\/2022\/10\/19\/resizemodel\/","title":{"rendered":"\u8ba9\u7f51\u7edc\u6765\u5b66\u4e60resize\uff1a\u63d2\u5373\u7528\u7684\u65b0\u578b\u56fe\u50cf\u8c03\u6574\u5668\u6a21\u578b"},"content":{"rendered":"\n

Learning to Resize Images for Computer Vision Tasks<\/strong><\/p>\n\n\n\n

\u8bba\u6587\u5730\u5740\uff1ahttps:\/\/arxiv.org\/abs\/2103.09950<\/a><\/strong><\/p>\n\n\n\n

\u4ee3\u7801\uff1ahttps:\/\/github.com\/KushajveerSingh\/resize_network_cv<\/a><\/p>\n\n\n\n

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

\u5c3d\u7ba1\u8fd1\u5e74\u6765\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u5f88\u5927\u5730\u4fc3\u8fdb\u4e86\u8ba1\u7b97\u673a\u89c6\u89c9\u7684\u53d1\u5c55\uff0c\u4f46\u4e00\u4e2a\u91cd\u8981\u65b9\u9762\u5f88\u5c11\u88ab\u5173\u6ce8\uff1a\u56fe\u50cf\u5927\u5c0f\u5bf9\u88ab\u8bad\u7ec3\u7684\u4efb\u52a1\u7684\u51c6\u786e\u6027\u7684\u5f71\u54cd<\/strong> \u3002\u901a\u5e38\uff0c\u8f93\u5165\u56fe\u50cf\u7684\u5927\u5c0f\u88ab\u8c03\u6574\u5230\u4e00\u4e2a\u76f8\u5bf9\u8f83\u5c0f\u7684\u7a7a\u95f4\u5206\u8fa8\u7387(\u4f8b\u5982\uff0c224\u00d7224)\uff0c\u7136\u540e\u518d\u8fdb\u884c\u8bad\u7ec3\u548c\u63a8\u7406\u3002\u8fd9\u79cd\u8c03\u6574\u5927\u5c0f\u7684\u673a\u5236\u901a\u5e38\u662f\u56fa\u5b9a\u7684\u56fe\u50cf\u8c03\u6574\u5668\uff08image resizer\uff09\uff08\u5982\uff1a\u53cc\u884c\u7ebf\u63d2\u503c\uff09\u4f46\u662f\u8fd9\u4e9b\u8c03\u6574\u5668\u662f\u5426\u9650\u5236\u4e86\u8bad\u7ec3\u7f51\u7edc\u7684\u4efb\u52a1\u6027\u80fd\u5462\uff1f<\/strong> \u4f5c\u8005\u901a\u8fc7\u5b9e\u9a8c\u8bc1\u660e\u4e86\u5178\u578b\u7684\u7ebf\u6027\u8c03\u6574\u5668\u53ef\u4ee5\u88ab\u53ef\u5b66\u4e60\u7684\u8c03\u6574\u5668\u53d6\u4ee3\uff0c\u4ece\u800c\u5927\u5927\u63d0\u9ad8\u6027\u80fd<\/strong> \u3002\u867d\u7136\u7ecf\u5178\u7684\u8c03\u6574\u5668\u901a\u5e38\u4f1a\u5177\u5907\u66f4\u597d\u7684\u5c0f\u56fe\u50cf\u611f\u77e5\u8d28\u91cf\uff08\u5373\u5bf9\u4eba\u7c7b\u8bc6\u522b\u56fe\u7247\u66f4\u52a0\u53cb\u597d\uff09\uff0c\u672c\u6587\u63d0\u51fa\u7684\u53ef\u5b66\u4e60\u8c03\u6574\u5668\u4e0d\u4e00\u5b9a\u4f1a\u5177\u5907\u66f4\u597d\u7684\u89c6\u89c9\u8d28\u91cf\uff0c\u4f46\u80fd\u591f\u63d0\u9ad8CV\u4efb\u52a1\u7684\u6027\u80fd\u3002<\/p>\n\n\n\n

\u5728\u4e0d\u540c\u7684\u4efb\u52a1\u4e2d\uff0c\u53ef\u5b66\u4e60\u7684\u56fe\u50cf\u8c03\u6574\u5668\u4e0ebaseline\u89c6\u89c9\u6a21\u578b\u8fdb\u884c\u8054\u5408\u8bad\u7ec3\u3002\u8fd9\u79cd\u53ef\u5b66\u4e60\u7684\u57fa\u4e8ecnn\u7684\u8c03\u6574\u5668\u521b\u5efa\u4e86\u673a\u5668\u53cb\u597d\u7684\u89c6\u89c9\u64cd\u4f5c\uff0c\u56e0\u6b64\u5728\u4e0d\u540c\u7684\u89c6\u89c9\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u4e86\u66f4\u597d\u7684\u6027\u80fd<\/strong> \u3002\u4f5c\u8005\u4f7f\u7528ImageNet\u6570\u636e\u96c6\u6765\u8fdb\u884c\u5206\u7c7b\u4efb\u52a1\uff0c\u5b9e\u9a8c\u4e2d\u4f7f\u7528\u56db\u79cd\u4e0d\u540c\u7684baseline\u6a21\u578b\u6765\u5b66\u4e60\u4e0d\u540c\u7684\u8c03\u6574\u5668\uff0c\u76f8\u6bd4\u4e8ebaseline\u6a21\u578b\uff0c\u4f7f\u7528\u672c\u6587\u63d0\u51fa\u7684\u53ef\u5b66\u4e60\u8c03\u6574\u5668\u80fd\u591f\u83b7\u5f97\u66f4\u9ad8\u7684\u6027\u80fd\u63d0\u5347\u3002<\/p>\n\n\n\n

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

\u4e3b\u8981\u5305\u62ec\u4e86\u4e24\u4e2a\u91cd\u8981\u7684\u7279\u6027\uff1a\uff081\uff09 \u53cc\u7ebf\u6027\u7279\u5f81\u8c03\u6574\u5927\u5c0f\uff08bilinear feature resizing\uff09\uff0c\u4ee5\u53ca\uff082\uff09\u8df3\u8fc7\u8fde\u63a5\uff08skip connection\uff09\uff0c\u8be5\u8fde\u63a5\u53ef\u5bb9\u7eb3\u53cc\u7ebf\u6027\u8c03\u6574\u5927\u5c0f\u7684\u56fe\u50cf\u548cCNN\u529f\u80fd\u7684\u7ec4\u5408\u3002<\/p>\n\n\n\n

\u7b2c\u4e00\u4e2a\u7279\u6027\u8003\u8651\u5230\u4ee5\u539f\u59cb\u5206\u8fa8\u7387\u8ba1\u7b97\u7684\u7279\u5f81\u4e0e\u6a21\u578b\u7684\u4e00\u81f4\u6027\u3002\u8df3\u8fc7\u8fde\u63a5\u53ef\u4ee5\u7b80\u5316\u5b66\u4e60\u8fc7\u7a0b\uff0c\u56e0\u4e3a\u91cd\u5b9a\u5927\u5c0f\u5668\u6a21\u578b\u53ef\u4ee5\u76f4\u63a5\u5c06\u53cc\u7ebf\u6027\u91cd\u5b9a\u5927\u5c0f\u7684\u56fe\u50cf\u4f20\u9012\u5230\u57fa\u7ebf\u4efb\u52a1\u4e2d\u3002<\/p>\n\n\n\n

\u4e0e\u4e00\u822c\u7684\u7f16\u7801\u5668-\u89e3\u7801\u5668\u67b6\u6784\u4e0d\u540c\uff0c\u8fd9\u7bc7\u8bba\u6587\u4e2d\u6240\u63d0\u51fa\u7684\u4f53\u7cfb\u7ed3\u6784\u5141\u8bb8\u5c06\u56fe\u50cf\u5927\u5c0f\u8c03\u6574\u4e3a\u4efb\u4f55\u76ee\u6807\u5927\u5c0f\u548c\u7eb5\u6a2a\u6bd4\uff08\u6ce8\u610f\uff1a\u8fd9\u4e2a\u5927\u5c0f\u5fc5\u987b\u662f\u6211\u4eec\u81ea\u5df1\u8bbe\u5b9a\u7684\uff0c\u800c\u4e0d\u662f\u7f51\u7edc \u81ea\u5df1\u5b66\u4e60\u7684\uff09\u3002\u5e76\u4e14\u53ef\u5b66\u4e60\u7684resizer\u6027\u80fd\u51e0\u4e4e\u4e0d\u4f9d\u8d56\u4e8e\u53cc\u7ebf\u6027\u91cd\u5b9a\u5668\u7684\u9009\u62e9\uff0c\u8fd9\u610f\u5473\u7740\u5b83\u53ef\u4ee5\u76f4\u63a5\u66ff\u6362\u5176\u4ed6\u73b0\u6210\u7684\u65b9\u6cd5\u3002<\/p>\n\n\n\n

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

\u4ee5\u4e0a\u4e4b\u540e\uff0c\u5c31\u6ca1\u6709\u522b\u7684\u4e86\uff0c\u8fd8\u4ee5\u4e3a\u662f\u4ec0\u4e48\u6837\u5b50\u7684\u60ca\u5929\u8bbe\u8ba1\uff0c\u6700\u540e\u4e0d\u5c31\u662f\uff1a\u7ed9\u7f51\u7edc\u53d8\u5f97\u590d\u6742\u4e86\u5417\uff0c\u628a\u8fd9\u79cd\u590d\u6742\u8bf4\u6210\u662f\u53ef\u5b66\u4e60\u7684resizer\uff0c\u8fd9\u6837\u7684\u8bdd\uff0c\u666e\u901a\u7f51\u7edc\u7684\u6d45\u5c42\u90fd\u53ef\u4ee5\u8bf4\u6210\u662f\u53ef\u5b66\u4e60\u7684resizer\u4e0d\u662f\u5417\uff1f<\/p>\n\n\n\n

\u53e6\u5916\uff0c\u901a\u8fc7\u4e00\u4e9b\u5b9e\u9a8c \u6765\u8bf4\uff0c\u786e\u5b9e\u80fd\u591f\u63d0\u5347\u6548\u679c\u3002\u4e2a\u4eba\u8ba4\u4e3a \u4f5c\u8005\u63d0\u51fa\u7684resizer\u6a21\u578b\u5b9e\u9645\u4e0a\u662f\u4e00\u4e2a\u53ef\u8bad\u7ec3\u7684\u6570\u636e\u589e\u5f3a\u65b9\u6cd5\uff0c\u751a\u81f3\u4e5f\u53ef\u4ee5\u8ba4\u4e3a\u5c31\u662f\u5c06\u6a21\u578b\u53d8\u5f97\u66f4\u52a0\u590d\u6742\u3002\u6574\u4f53\u7684\u7f51\u7edc\u5c31\u50cf\u662f\u4e00\u822c\u6a21\u578b\u4e2dresblock\u3002<\/p>\n\n\n\n

\u4f5c\u8005\u7684\u5bf9\u6bd4\u8bd5\u9a8c\u662f\u8fd9\u6837\u505a\u7684\uff1a\u9996\u5148\u901a\u8fc7\u5e38\u7528\u7684reisze\u65b9\u6cd5\u8bad\u7ec3\u7f51\u7edc\u6a21\u578b\uff0c\u4f5c\u4e3abaseline\uff0c\u7136\u540e\u5728\u8bad\u7ec3\u597d\u7684\u7f51\u7edc\u6a21\u578b\u524d\u9762\u6dfb\u52a0\u53ef\u5b66\u4e60\u7684resizer\uff0c\u7136\u540e\u8fdb\u884c\u8bad\u7ec3\uff0c\u4f5c\u4e3a\u81ea\u5df1\u7684\u65b9\u6cd5\u3002\u611f\u53d7\u4e00\u4e0b\u4f5c\u8005\u7684\u5b9e\u9a8c\u7ed3\u679c\u5427\u3002\u5982\u88683\u548c\u88684<\/p>\n\n\n\n

\"\"\/
\u88683 \u8bad\u7ec3\u914d\u7f6e<\/figcaption><\/figure>\n\n\n\n
\"\"\/
\u88684 \u5bf9\u6bd4\u8bd5\u9a8c<\/figcaption><\/figure>\n\n\n\n

\u56db. \u603b\u7ed3<\/strong><\/p>\n\n\n\n

\u6587\u7ae0\u5199\u5f97\u597d\uff0c\u52a0\u4e0a\u70b9\u8fd0\u6c14\uff0c\u90fd\u53ef\u4ee5\u53d1\u9ad8\u8d28\u91cf\u8bba\u6587\uff0c\u4ee5\u4e0a<\/strong><\/p>\n\n\n\n

\u4ee3\u7801\u5b9e\u73b0\uff1a<\/p>\n\n\n\n

import torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nfrom functools import partial\n\n\"\"\"\n    Learning to Resize Images for Computer Vision Tasks\n    https://arxiv.org\/pdf\/2105.04714.pdf\n\"\"\"\n\ndef conv1x1(in_chs, out_chs = 16):\n    return nn.Conv2d(in_chs, out_chs, kernel_size=1, stride=1, padding=0)\n\n\ndef conv3x3(in_chs, out_chs = 16):\n    return nn.Conv2d(in_chs, out_chs, kernel_size=3, stride=1, padding=1)\n\n\ndef conv7x7(in_chs, out_chs = 16):\n    return nn.Conv2d(in_chs, out_chs, kernel_size=7, stride=1, padding=3)\n\n\nclass ResBlock(nn.Module):\n    def __init__(self, in_chs,out_chs = 16):\n        super(ResBlock, self).__init__()\n        self.layers = nn.Sequential(\n            conv3x3(in_chs, out_chs),\n            nn.BatchNorm2d(out_chs),\n            nn.LeakyReLU(0.2),\n            conv3x3(out_chs, out_chs),\n            nn.BatchNorm2d(out_chs)\n        )\n    def forward(self, x):\n        identity = x\n        out = self.layers(x)\n        out += identity\n        return out\n\n\nclass Resizer(nn.Module):\n    def __init__(self, in_chs, out_size, n_filters = 16, n_res_blocks = 1, mode = 'bilinear'):\n        super(Resizer, self).__init__()\n        self.interpolate_layer = partial(F.interpolate, size=out_size, mode=mode,\n            align_corners=(True if mode in ('linear', 'bilinear', 'bicubic', 'trilinear') else None))\n        self.conv_layers = nn.Sequential(\n            conv7x7(in_chs, n_filters),\n            nn.LeakyReLU(0.2),\n            conv1x1(n_filters, n_filters),\n            nn.LeakyReLU(0.2),\n            nn.BatchNorm2d(n_filters)\n        )\n        self.residual_layers = nn.Sequential()\n        for i in range(n_res_blocks):\n            self.residual_layers.add_module(f'res{i}', ResBlock(n_filters, n_filters))\n        self.residual_layers.add_module('conv3x3', conv3x3(n_filters, n_filters))\n        self.residual_layers.add_module('bn', nn.BatchNorm2d(n_filters))\n        self.final_conv = conv7x7(n_filters, in_chs)\n\n    def forward(self, x):\n        identity = self.interpolate_layer(x)\n        conv_out = self.conv_layers(x)\n        conv_out = self.interpolate_layer(conv_out)\n        conv_out_identity = conv_out\n        res_out = self.residual_layers(conv_out)\n        res_out += conv_out_identity\n        out = self.final_conv(res_out)\n        out += identity\n        return<\/code><\/pre>\n","protected":false},"excerpt":{"rendered":"
Learning to Resize Images for Computer Vision Tasks \u8bba\u6587\u5730 … \u7ee7\u7eed\u9605\u8bfb\u8ba9\u7f51\u7edc\u6765\u5b66\u4e60resize\uff1a\u63d2\u5373\u7528\u7684\u65b0\u578b\u56fe\u50cf\u8c03\u6574\u5668\u6a21\u578b<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[4,24,9],"tags":[],"_links":{"self":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/9071"}],"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=9071"}],"version-history":[{"count":17,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/9071\/revisions"}],"predecessor-version":[{"id":9179,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/9071\/revisions\/9179"}],"wp:attachment":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/media?parent=9071"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/categories?post=9071"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/tags?post=9071"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}