{"id":2421,"date":"2024-07-01T11:23:26","date_gmt":"2024-07-01T03:23:26","guid":{"rendered":"https:\/\/17aitech.com\/?p=2421"},"modified":"2024-10-08T15:17:18","modified_gmt":"2024-10-08T07:17:18","slug":"%e3%80%90%e8%af%be%e7%a8%8b%e6%80%bb%e7%bb%93%e3%80%91day13%ef%bc%88%e4%b8%8b%ef%bc%89%ef%bc%9a%e4%ba%ba%e8%84%b8%e8%af%86%e5%88%ab%e5%92%8cmtcnn%e6%a8%a1%e5%9e%8b","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=2421","title":{"rendered":"\u3010\u8bfe\u7a0b\u603b\u7ed3\u3011Day13\uff08\u4e0b\uff09\uff1a\u4eba\u8138\u8bc6\u522b\u548cMTCNN\u6a21\u578b"},"content":{"rendered":"<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_78 ez-toc-wrap-left-text counter-hierarchy ez-toc-counter ez-toc-light-blue ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title\" style=\"cursor:inherit\">\u6587\u7ae0\u76ee\u5f55<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"Toggle Table of Content\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Toggle<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #999;color:#999\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewBox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #999;color:#999\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewBox=\"0 0 24 24\" version=\"1.2\" baseProfile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1 ' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E5%89%8D%E8%A8%80\" >\u524d\u8a00<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E4%BA%BA%E8%84%B8%E8%AF%86%E5%88%AB\" >\u4eba\u8138\u8bc6\u522b<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E4%BA%BA%E8%84%B8%E6%A3%80%E6%B5%8B\" >\u4eba\u8138\u68c0\u6d4b<\/a><ul class='ez-toc-list-level-4' ><li class='ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E7%AE%80%E8%BF%B0\" >\u7b80\u8ff0<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E8%AF%86%E5%88%AB%E8%BF%87%E7%A8%8B\" >\u8bc6\u522b\u8fc7\u7a0b<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E8%BE%93%E5%85%A5%E8%BE%93%E5%87%BA\" >\u8f93\u5165\u8f93\u51fa<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-7\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E5%BA%94%E7%94%A8%E5%9C%BA%E6%99%AF\" >\u5e94\u7528\u573a\u666f<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-8\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E4%BA%BA%E8%84%B8%E6%A3%80%E6%B5%8B%E7%89%B9%E7%82%B9\" >\u4eba\u8138\u68c0\u6d4b\u7279\u70b9<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-9\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E4%BA%BA%E8%84%B8%E8%BA%AB%E4%BB%BD%E8%AF%86%E5%88%AB\" >\u4eba\u8138\u8eab\u4efd\u8bc6\u522b<\/a><ul class='ez-toc-list-level-4' ><li class='ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-10\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E7%AE%80%E8%BF%B0-2\" >\u7b80\u8ff0<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-11\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E8%AF%86%E5%88%AB%E8%BF%87%E7%A8%8B-2\" >\u8bc6\u522b\u8fc7\u7a0b<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-12\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E5%BA%94%E7%94%A8%E9%A2%86%E5%9F%9F\" >\u5e94\u7528\u9886\u57df<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-13\" href=\"https:\/\/17aitech.com\/?p=2421\/#MTCNN%E6%A8%A1%E5%9E%8B\" >MTCNN\u6a21\u578b<\/a><ul class='ez-toc-list-level-4' ><li class='ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-14\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E7%AE%80%E4%BB%8B\" >\u7b80\u4ecb<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-15\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E6%A8%A1%E5%9E%8B%E7%BB%93%E6%9E%84\" >\u6a21\u578b\u7ed3\u6784<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-16\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E6%95%B4%E4%BD%93%E6%B5%81%E7%A8%8B\" >\u6574\u4f53\u6d41\u7a0b<\/a><ul class='ez-toc-list-level-5' ><li class='ez-toc-heading-level-5'><a class=\"ez-toc-link ez-toc-heading-17\" href=\"https:\/\/17aitech.com\/?p=2421\/#P-net%EF%BC%9A%E4%BA%BA%E8%84%B8%E6%A3%80%E6%B5%8B\" >P-net\uff1a\u4eba\u8138\u68c0\u6d4b<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-5'><a class=\"ez-toc-link ez-toc-heading-18\" href=\"https:\/\/17aitech.com\/?p=2421\/#R-net%EF%BC%9A%E4%BA%BA%E8%84%B8%E5%AF%B9%E9%BD%90\" >R-net\uff1a\u4eba\u8138\u5bf9\u9f50<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-5'><a class=\"ez-toc-link ez-toc-heading-19\" href=\"https:\/\/17aitech.com\/?p=2421\/#O-net%EF%BC%9A%E4%BA%BA%E8%84%B8%E8%AF%86%E5%88%AB\" >O-net\uff1a\u4eba\u8138\u8bc6\u522b<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-20\" href=\"https:\/\/17aitech.com\/?p=2421\/#MTCNN%E7%94%A8%E5%88%B0%E7%9A%84%E4%B8%BB%E8%A6%81%E6%A8%A1%E5%9D%97\" >MTCNN\u7528\u5230\u7684\u4e3b\u8981\u6a21\u5757<\/a><ul class='ez-toc-list-level-5' ><li class='ez-toc-heading-level-5'><a class=\"ez-toc-link ez-toc-heading-21\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E5%9B%BE%E5%83%8F%E9%87%91%E5%AD%97%E5%A1%94\" >\u56fe\u50cf\u91d1\u5b57\u5854<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-5'><a class=\"ez-toc-link ez-toc-heading-22\" href=\"https:\/\/17aitech.com\/?p=2421\/#IOU\" >IOU<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-5'><a class=\"ez-toc-link ez-toc-heading-23\" href=\"https:\/\/17aitech.com\/?p=2421\/#NMS%EF%BC%88Non-Maximum_Suppression%EF%BC%8C%E9%9D%9E%E6%9E%81%E5%A4%A7%E5%80%BC%E6%8A%91%E5%88%B6%EF%BC%89\" >NMS\uff08Non-Maximum Suppression\uff0c\u975e\u6781\u5927\u503c\u6291\u5236\uff09<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-24\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E4%BB%A3%E7%A0%81%E5%AE%9E%E7%8E%B0\" >\u4ee3\u7801\u5b9e\u73b0<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-25\" href=\"https:\/\/17aitech.com\/?p=2421\/#MTCNN%E8%AE%AD%E7%BB%83%E9%80%BB%E8%BE%91\" >MTCNN\u8bad\u7ec3\u903b\u8f91<\/a><ul class='ez-toc-list-level-4' ><li class='ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-26\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E5%87%86%E5%A4%87%E8%AE%AD%E7%BB%83%E6%95%B0%E6%8D%AE%E9%9B%86\" >\u51c6\u5907\u8bad\u7ec3\u6570\u636e\u96c6<\/a><ul class='ez-toc-list-level-5' ><li class='ez-toc-heading-level-5'><a class=\"ez-toc-link ez-toc-heading-27\" href=\"https:\/\/17aitech.com\/?p=2421\/#CelebA%E6%95%B0%E6%8D%AE%E9%9B%86%E7%AE%80%E4%BB%8B\" >CelebA\u6570\u636e\u96c6\u7b80\u4ecb<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-5'><a class=\"ez-toc-link ez-toc-heading-28\" href=\"https:\/\/17aitech.com\/?p=2421\/#CelebA%E6%95%B0%E6%8D%AE%E9%9B%86%E4%B8%8B%E8%BD%BD\" >CelebA\u6570\u636e\u96c6\u4e0b\u8f7d<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-29\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E4%B8%8B%E8%BD%BD%E5%92%8C%E5%87%86%E5%A4%87%E8%AE%AD%E7%BB%83%E9%9B%86\" >\u4e0b\u8f7d\u548c\u51c6\u5907\u8bad\u7ec3\u96c6<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-30\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E8%AE%AD%E7%BB%83%E9%9B%86%E9%A2%84%E5%A4%84%E7%90%86\" >\u8bad\u7ec3\u96c6\u9884\u5904\u7406<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-4'><a class=\"ez-toc-link ez-toc-heading-31\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E4%B8%89%E4%B8%AA%E6%A8%A1%E5%9E%8B%E5%88%86%E5%88%AB%E8%AE%AD%E7%BB%83\" >\u4e09\u4e2a\u6a21\u578b\u5206\u522b\u8bad\u7ec3<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-32\" href=\"https:\/\/17aitech.com\/?p=2421\/#MTCNN%E6%8E%A8%E7%90%86%E9%80%BB%E8%BE%91\" >MTCNN\u63a8\u7406\u903b\u8f91<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-33\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E5%86%85%E5%AE%B9%E5%B0%8F%E7%BB%93\" >\u5185\u5bb9\u5c0f\u7ed3<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-34\" href=\"https:\/\/17aitech.com\/?p=2421\/#%E5%8F%82%E8%80%83%E8%B5%84%E6%96%99\" >\u53c2\u8003\u8d44\u6599<\/a><\/li><\/ul><\/nav><\/div>\n<h2><span class=\"ez-toc-section\" id=\"%E5%89%8D%E8%A8%80\"><\/span>\u524d\u8a00<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>\u5728\u4e0a\u4e00\u7ae0\u8bfe\u7a0b<a href=\"https:\/\/17aitech.com\/?p=2398\">\u3010\u8bfe\u7a0b\u603b\u7ed3\u3011Day13\uff08\u4e0a\uff09\uff1a\u4f7f\u7528YOLO\u8fdb\u884c\u76ee\u6807\u68c0\u6d4b<\/a>\uff0c\u6211\u4eec\u4e86\u89e3\u5230\u76ee\u6807\u68c0\u6d4b\u6709\u4e24\u79cd\u7b56\u7565\uff0c\u4e00\u79cd\u662f\u4ee5YOLO\u4e3a\u4ee3\u8868\u7684\u7b56\u7565\uff1a\u7279\u5f81\u63d0\u53d6\u2192\u5207\u7247\u2192\u5206\u7c7b\u56de\u5f52\uff1b\u53e6\u5916\u4e00\u79cd\u662f\u4ee5MTCNN\u4e3a\u4ee3\u8868\u7684\u7b56\u7565\uff1a\u5148\u56fe\u50cf\u5207\u7247\u2192\u7279\u5f81\u63d0\u53d6\u2192\u5206\u7c7b\u548c\u56de\u5f52\u3002\u56e0\u6b64\uff0c\u672c\u7ae0\u5185\u5bb9\u5c06\u6df1\u5165\u4e86\u89e3MTCNN\u6a21\u578b\uff0c\u5305\u62ec\uff1aMTCNN\u7684\u6a21\u578b\u7ec4\u6210\u3001\u6a21\u578b\u8bad\u7ec3\u8fc7\u7a0b\u3001\u6a21\u578b\u9884\u6d4b\u8fc7\u7a0b\u7b49\u3002<\/p>\n<h2><span class=\"ez-toc-section\" id=\"%E4%BA%BA%E8%84%B8%E8%AF%86%E5%88%AB\"><\/span>\u4eba\u8138\u8bc6\u522b<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>\u5728\u5c55\u5f00\u4e86\u89e3MTCNN\u4e4b\u524d\uff0c\u6211\u4eec\u5bf9\u4eba\u8138\u68c0\u6d4b\u5148\u505a\u4e00\u4e2a\u521d\u6b65\u7684\u68b3\u7406\u548c\u4e86\u89e3\u3002\u4eba\u8138\u8bc6\u522b\u7ec6\u5206\u6709\u4e24\u79cd\uff1a\u4eba\u8138\u68c0\u6d4b\u548c\u4eba\u8138\u8eab\u4efd\u8bc6\u522b\u3002<\/p>\n<h3><span class=\"ez-toc-section\" id=\"%E4%BA%BA%E8%84%B8%E6%A3%80%E6%B5%8B\"><\/span>\u4eba\u8138\u68c0\u6d4b<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<h4><span class=\"ez-toc-section\" id=\"%E7%AE%80%E8%BF%B0\"><\/span>\u7b80\u8ff0<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<p>\u4eba\u8138\u68c0\u6d4b\u662f\u4e00\u4e2a\u91cd\u8981\u7684\u5e94\u7528\u9886\u57df\uff0c\u5b83\u901a\u5e38\u7528\u4e8e\u8bc6\u522b\u56fe\u50cf\u6216\u89c6\u9891\u4e2d\u7684\u4eba\u8138\uff0c\u5e76\u5b9a\u4f4d\u5176\u4f4d\u7f6e\u3002<\/p>\n<h4><span class=\"ez-toc-section\" id=\"%E8%AF%86%E5%88%AB%E8%BF%87%E7%A8%8B\"><\/span>\u8bc6\u522b\u8fc7\u7a0b<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<ol>\n<li><strong>\u8f93\u5165\u56fe\u50cf<\/strong>\uff1a\u9996\u5148\uff0c\u5c06\u5305\u542b\u4eba\u8138\u7684\u56fe\u50cf\u8f93\u5165\u5230\u4eba\u8138\u68c0\u6d4b\u6a21\u578b\u4e2d\u3002<\/li>\n<li><strong>\u7279\u5f81\u63d0\u53d6<\/strong>\uff1a\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u5c06\u5b66\u4e60\u63d0\u53d6\u56fe\u50cf\u4e2d\u7684\u7279\u5f81\uff0c\u4ee5\u4fbf\u8bc6\u522b\u4eba\u8138\u3002<\/li>\n<li><strong>\u4eba\u8138\u5b9a\u4f4d<\/strong>\uff1a\u6a21\u578b\u901a\u8fc7\u5728\u56fe\u50cf\u4e2d\u5b9a\u4f4d\u4eba\u8138\u7684\u4f4d\u7f6e\uff0c\u901a\u5e38\u4f7f\u7528\u77e9\u5f62\u8fb9\u754c\u6846\u6765\u6846\u5b9a\u4eba\u8138\u533a\u57df\u3002<\/li>\n<li><strong>\u8f93\u51fa\u7ed3\u679c<\/strong>\uff1a\u6700\u7ec8\u8f93\u51fa\u5305\u542b\u4eba\u8138\u4f4d\u7f6e\u4fe1\u606f\u7684\u7ed3\u679c\uff0c\u53ef\u4ee5\u662f\u8fb9\u754c\u6846\u7684\u5750\u6807\u6216\u5176\u4ed6\u5f62\u5f0f\u7684\u6807\u6ce8\u3002<\/li>\n<\/ol>\n<h4><span class=\"ez-toc-section\" id=\"%E8%BE%93%E5%85%A5%E8%BE%93%E5%87%BA\"><\/span>\u8f93\u5165\u8f93\u51fa<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<ul>\n<li>\u8f93\u5165\uff1a\u4e00\u5f20\u56fe\u50cf<\/li>\n<li>\u8f93\u51fa\uff1a\u6240\u6709\u4eba\u8138\u7684\u5750\u6807\u6846<\/li>\n<\/ul>\n<h4><span class=\"ez-toc-section\" id=\"%E5%BA%94%E7%94%A8%E5%9C%BA%E6%99%AF\"><\/span>\u5e94\u7528\u573a\u666f<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<ul>\n<li>\u8868\u60c5\u8bc6\u522b\uff1a\u8bc6\u522b\u4eba\u8138\u7684\u8868\u60c5\uff0c\u5982\u5feb\u4e50\u3001\u60b2\u4f24\u7b49\u3002<\/li>\n<li>\u5e74\u9f84\u8bc6\u522b\uff1a\u6839\u636e\u4eba\u8138\u7279\u5f81\u63a8\u65ad\u51fa\u4eba\u7684\u5e74\u9f84\u6bb5\u3002<\/li>\n<li>\u4eba\u8138\u8868\u60c5\u751f\u6210\uff1a\u901a\u8fc7\u68c0\u6d4b\u5230\u7684\u4eba\u8138\u751f\u6210\u4e0d\u540c\u7684\u8868\u60c5\u3002<\/li>\n<li>&#8230;<br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u8868\u60c5\u8bc6\u522b.jpeg\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u8868\u60c5\u8bc6\u522b.jpeg\" alt=\"\" \/><\/a><\/li>\n<\/ul>\n<h4><span class=\"ez-toc-section\" id=\"%E4%BA%BA%E8%84%B8%E6%A3%80%E6%B5%8B%E7%89%B9%E7%82%B9\"><\/span>\u4eba\u8138\u68c0\u6d4b\u7279\u70b9<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<p>\u4eba\u8138\u68c0\u6d4b\u662f\u76ee\u6807\u68c0\u6d4b\u4e2d\u6700\u7b80\u5355\u7684\u4efb\u52a1<\/p>\n<ul>\n<li>\u7c7b\u522b\u5c11<\/li>\n<li>\u4eba\u8138\u5f62\u72b6\u6bd4\u8f83\u56fa\u5b9a<\/li>\n<li>\u4eba\u8138\u7279\u5f81\u6bd4\u8f83\u56fa\u5b9a<\/li>\n<li>\u5468\u56f4\u73af\u5883\u4e00\u822c\u6bd4\u8f83\u597d<\/li>\n<\/ul>\n<h3><span class=\"ez-toc-section\" id=\"%E4%BA%BA%E8%84%B8%E8%BA%AB%E4%BB%BD%E8%AF%86%E5%88%AB\"><\/span>\u4eba\u8138\u8eab\u4efd\u8bc6\u522b<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<h4><span class=\"ez-toc-section\" id=\"%E7%AE%80%E8%BF%B0-2\"><\/span>\u7b80\u8ff0<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<p>\u4eba\u8138\u8eab\u4efd\u8bc6\u522b\u662f\u6307\u901a\u8fc7\u8bc6\u522b\u4eba\u8138\u4e0a\u7684\u72ec\u7279\u7279\u5f81\u6765\u786e\u5b9a\u4e00\u4e2a\u4eba\u7684\u8eab\u4efd\u3002<\/p>\n<h4><span class=\"ez-toc-section\" id=\"%E8%AF%86%E5%88%AB%E8%BF%87%E7%A8%8B-2\"><\/span>\u8bc6\u522b\u8fc7\u7a0b<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<p><strong>\u4eba\u8138\u5f55\u5165\u6d41\u7a0b<\/strong>\uff1a<\/p>\n<ol>\n<li>\u6570\u636e\u91c7\u96c6\uff1a\u91c7\u96c6\u5305\u542b\u4eba\u8138\u7684\u56fe\u50cf\u6570\u636e\u96c6\u3002<\/li>\n<li>\u4eba\u8138\u68c0\u6d4b\uff1a\u4f7f\u7528\u4eba\u8138\u68c0\u6d4b\u7b97\u6cd5\u5b9a\u4f4d\u56fe\u50cf\u4e2d\u7684\u4eba\u8138\u533a\u57df\u3002<\/li>\n<li>\u4eba\u8138\u7279\u5f81\u63d0\u53d6\uff1a\u901a\u8fc7\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u63d0\u53d6\u4eba\u8138\u56fe\u50cf\u7684\u7279\u5f81\u5411\u91cf\u3002<\/li>\n<li>\u7279\u5f81\u5411\u91cf\u5b58\u50a8\uff1a\u5c06\u63d0\u53d6\u5230\u7684\u7279\u5f81\u5411\u91cf\u5b58\u50a8\u5728\u5411\u91cf\u6570\u636e\u5e93\u4e2d\u3002<\/li>\n<\/ol>\n<p><strong>\u4eba\u8138\u9a8c\u8bc1\u6d41\u7a0b<\/strong>\uff1a<\/p>\n<ol>\n<li>\u4eba\u8138\u68c0\u6d4b\uff1a\u4f7f\u7528\u4eba\u8138\u68c0\u6d4b\u7b97\u6cd5\u5b9a\u4f4d\u56fe\u50cf\u4e2d\u7684\u4eba\u8138\u533a\u57df\u3002<\/li>\n<li>\u4eba\u8138\u7279\u5f81\u63d0\u53d6\uff1a\u901a\u8fc7\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u63d0\u53d6\u4eba\u8138\u56fe\u50cf\u7684\u7279\u5f81\u5411\u91cf\u3002<\/li>\n<li>\u4eba\u8138\u7279\u5f81\u5339\u914d\uff1a\u5c06\u8f93\u5165\u4eba\u8138\u7684\u7279\u5f81\u5411\u91cf\u4e0e\u5411\u91cf\u6570\u636e\u5e93\u4e2d\u7684\u7279\u5f81\u5411\u91cf\u8fdb\u884c\u5339\u914d\u3002<\/li>\n<li>\u8eab\u4efd\u8bc6\u522b\uff1a\u6839\u636e\u5339\u914d\u7ed3\u679c\u786e\u5b9a\u8f93\u5165\u4eba\u8138\u7684\u8eab\u4efd\u4fe1\u606f\u3002<\/li>\n<\/ol>\n<h4><span class=\"ez-toc-section\" id=\"%E5%BA%94%E7%94%A8%E9%A2%86%E5%9F%9F\"><\/span>\u5e94\u7528\u9886\u57df<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<ul>\n<li>\u5b89\u9632\u76d1\u63a7\uff1a\u7528\u4e8e\u95e8\u7981\u7cfb\u7edf\u3001\u76d1\u63a7\u7cfb\u7edf\u7b49\uff0c\u5b9e\u73b0\u4eba\u8138\u8bc6\u522b\u8fdb\u51fa\u63a7\u5236\u3002<\/li>\n<li>\u79fb\u52a8\u652f\u4ed8\uff1a\u901a\u8fc7\u4eba\u8138\u8bc6\u522b\u6765\u8fdb\u884c\u8eab\u4efd\u9a8c\u8bc1\uff0c\u5b9e\u73b0\u5b89\u5168\u7684\u79fb\u52a8\u652f\u4ed8\u529f\u80fd\u3002<\/li>\n<li>\u793e\u4ea4\u5a92\u4f53\uff1a\u7528\u4e8e\u81ea\u52a8\u6807\u8bb0\u7167\u7247\u4e2d\u7684\u4eba\u7269\uff0c\u65b9\u4fbf\u7528\u6237\u7ba1\u7406\u7167\u7247\u3002<\/li>\n<li>\u4eba\u673a\u4ea4\u4e92\uff1a\u5b9e\u73b0\u4eba\u8138\u8bc6\u522b\u767b\u5f55\u3001\u4eba\u8138\u89e3\u9501\u7b49\u529f\u80fd\u3002<br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u4eba\u8138\u8eab\u4efd\u8bc6\u522b.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u4eba\u8138\u8eab\u4efd\u8bc6\u522b.png\" alt=\"\" \/><\/a><\/li>\n<\/ul>\n<blockquote>\n<p>\u4e00\u822c\u6765\u8bf4\uff0c\u4e00\u5207\u76ee\u6807\u68c0\u6d4b\u7b97\u6cd5\u90fd\u53ef\u4ee5\u505a\u4eba\u8138\u68c0\u6d4b\uff0c\u4f46\u662f\u7531\u4e8e\u901a\u7528\u76ee\u6807\u68c0\u6d4b\u7b97\u6cd5\u505a\u4eba\u8138\u68c0\u6d4b\u592a\u91cd\u4e86\uff0c\u6240\u4ee5\u4f1a\u4f7f\u7528\u4e13\u95e8\u7684\u4eba\u8138\u8bc6\u522b\u7b97\u6cd5\uff0c\u800cMTCNN\u5c31\u662f\u8fd9\u6837\u4e00\u4e2a\u8f7b\u91cf\u7ea7\u548c\u4e13\u4e1a\u7ea7\u7684\u4eba\u8138\u68c0\u6d4b\u7f51\u7edc\u3002<\/p>\n<\/blockquote>\n<h3><span class=\"ez-toc-section\" id=\"MTCNN%E6%A8%A1%E5%9E%8B\"><\/span>MTCNN\u6a21\u578b<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<h4><span class=\"ez-toc-section\" id=\"%E7%AE%80%E4%BB%8B\"><\/span>\u7b80\u4ecb<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<p>MTCNN\uff08Multi-Task Cascaded Convolutional Neural Networks\uff09\u662f\u4e00\u79cd\u7528\u4e8e\u4eba\u8138\u68c0\u6d4b\u548c\u9762\u90e8\u5bf9\u9f50\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u3002<\/p>\n<p>\u8bba\u6587\u5730\u5740\uff1a<a href=\"https:\/\/arxiv.org\/abs\/1604.02878v1\">https:\/\/arxiv.org\/abs\/1604.02878v1<\/a><\/p>\n<h4><span class=\"ez-toc-section\" id=\"%E6%A8%A1%E5%9E%8B%E7%BB%93%E6%9E%84\"><\/span>\u6a21\u578b\u7ed3\u6784<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<ul>\n<li>MTCNN\u91c7\u7528\u4e86\u7ea7\u8054\u7ed3\u6784\uff0c\u5305\u62ec\u4e09\u4e2a\u9636\u6bb5\u7684\u6df1\u5ea6\u5377\u79ef\u7f51\u7edc\uff0c\u5206\u522b\u7528\u4e8e\u4eba\u8138\u68c0\u6d4b\u548c\u9762\u90e8\u5bf9\u9f50\u3002<\/li>\n<li>\u6bcf\u4e2a\u9636\u6bb5\u90fd\u6709\u4e0d\u540c\u7684\u4efb\u52a1\uff0c\u5305\u62ec\u4eba\u8138\u8fb9\u754c\u6846\u56de\u5f52\u3001\u4eba\u8138\u5173\u952e\u70b9\u5b9a\u4f4d\u7b49\u3002<br \/>\n<blockquote>\n<p>\u8fd9\u4e2a\u7ea7\u8054\u8fc7\u7a0b\uff0c\u76f8\u5f53\u4e8e<code>\u6d77\u9009\u2192\u6dd8\u6c70\u8d5b\u2192\u51b3\u8d5b<\/code>\u7684\u8fc7\u7a0b\u3002<\/p>\n<\/blockquote>\n<\/li>\n<\/ul>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/MTCNN\u6a21\u578b\u7ed3\u6784.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/MTCNN\u6a21\u578b\u7ed3\u6784.png\" alt=\"\" \/><\/a><\/p>\n<h4><span class=\"ez-toc-section\" id=\"%E6%95%B4%E4%BD%93%E6%B5%81%E7%A8%8B\"><\/span>\u6574\u4f53\u6d41\u7a0b<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<p>\u4e0a\u56fe\u662f\u8bba\u6587\u4e2d\u5bf9\u4e8eMTCNN\u6574\u4f53\u8fc7\u7a0b\u7684\u56fe\u793a\uff0c\u6211\u4eec\u6362\u4e00\u79cd\u8f83\u4e3a\u5bb9\u6613\u6613\u61c2\u7684\u56fe\u793a\u6765\u7406\u89e3\u6574\u4f53\u8fc7\u7a0b\uff1a<br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/MTCNN\u6d41\u7a0b1.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/MTCNN\u6d41\u7a0b1.png\" alt=\"\" \/><\/a><br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/MTCNN\u6d41\u7a0b2.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/MTCNN\u6d41\u7a0b2.png\" alt=\"\" \/><\/a><\/p>\n<ol>\n<li>\u5148\u5c06\u56fe\u7247\u751f\u6210\u4e0d\u540c\u5c3a\u5bf8\u7684\u56fe\u50cf\u91d1\u5b57\u5854\uff0c\u4ee5\u4fbf\u8bc6\u522b\u4e0d\u540c\u5927\u5c0f\u7684\u4eba\u8138\u3002<\/li>\n<li>\u5c06\u56fe\u7247\u8f93\u5165\u5230P-net\u4e2d\uff0c\u8bc6\u522b\u51fa\u53ef\u80fd\u5305\u542b\u4eba\u8138\u7684\u5019\u9009\u7a97\u53e3\u3002<\/li>\n<li>\u5c06P-net\u4e2d\u8bc6\u522b\u7684\u53ef\u80fd\u4eba\u8138\u7684\u5019\u9009\u7a97\u53e3\u8f93\u5165\u5230R-net\u4e2d\uff0c\u8bc6\u522b\u51fa\u66f4\u7cbe\u786e\u7684\u4eba\u8138\u4f4d\u7f6e\u3002<\/li>\n<li>\u5c06R-net\u4e2d\u8bc6\u522b\u7684\u4eba\u8138\u4f4d\u7f6e\u8f93\u5165\u5230O-net\u4e2d\uff0c\u8fdb\u884c\u66f4\u52a0\u7cbe\u7ec6\u5316\u8bc6\u522b\uff0c\u4ece\u800c\u627e\u5230\u4eba\u8138\u533a\u57df\u3002<\/li>\n<\/ol>\n<blockquote>\n<p>\u5907\u6ce8\uff1a\u4e0a\u56fe\u5f15\u7528\u81ea<a href=\"https:\/\/www.bilibili.com\/video\/BV1XJ41167pZ?vd_source=f825e70d1502bfc21582d29e60d89bb5\">\u79d1\u666e\uff1a\u4ec0\u4e48\u662fmtcnn\u4eba\u8138\u68c0\u6d4b\u7b97\u6cd5<\/a><\/p>\n<\/blockquote>\n<h5><span class=\"ez-toc-section\" id=\"P-net%EF%BC%9A%E4%BA%BA%E8%84%B8%E6%A3%80%E6%B5%8B\"><\/span>P-net\uff1a\u4eba\u8138\u68c0\u6d4b<span class=\"ez-toc-section-end\"><\/span><\/h5>\n<ul>\n<li><strong>\u540d\u79f0<\/strong>\uff1a\u63d0\u8bae\u7f51\u7edc\uff08proposal network\uff09<\/li>\n<li><strong>\u4f5c\u7528<\/strong>\uff1aP\u7f51\u7edc\u901a\u8fc7\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\uff08CNN\uff09\u5bf9\u8f93\u5165\u56fe\u50cf\u8fdb\u884c\u5904\u7406\uff0c\u8bc6\u522b\u51fa\u53ef\u80fd\u5305\u542b\u4eba\u8138\u7684\u5019\u9009\u7a97\u53e3\uff0c\u5e76\u5bf9\u8fd9\u4e9b\u5019\u9009\u7a97\u53e3\u8fdb\u884c\u8fb9\u754c\u6846\u7684\u56de\u5f52\uff0c\u4ee5\u66f4\u51c6\u786e\u5730\u5b9a\u4f4d\u4eba\u8138\u4f4d\u7f6e\u3002<\/li>\n<li><strong>\u7279\u70b9<\/strong>\uff1a\n<ul>\n<li>\u7eaf\u5377\u79ef\u7f51\u7edc\uff0c\u65e0\u5168\u94fe\u63a5\uff08<strong>\u7cbe\u9ad3\u6240\u5728<\/strong>\uff09<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h5><span class=\"ez-toc-section\" id=\"R-net%EF%BC%9A%E4%BA%BA%E8%84%B8%E5%AF%B9%E9%BD%90\"><\/span>R-net\uff1a\u4eba\u8138\u5bf9\u9f50<span class=\"ez-toc-section-end\"><\/span><\/h5>\n<ul>\n<li><strong>\u540d\u79f0<\/strong>\uff1a\u7cbe\u4fee\u7f51\u7edc\uff08refine network\uff09<\/li>\n<li><strong>\u4f5c\u7528<\/strong>\uff1aR\u7f51\u7edc\u901a\u8fc7\u5206\u7c7b\u5668\u548c\u56de\u5f52\u5668\u5bf9P\u7f51\u7edc\u751f\u6210\u7684\u5019\u9009\u7a97\u53e3\u8fdb\u884c\u5904\u7406\uff0c\u8fdb\u4e00\u6b65\u7b5b\u9009\u51fa\u5305\u542b\u4eba\u8138\u7684\u533a\u57df\uff0c\u5e76\u5bf9\u4eba\u8138\u4f4d\u7f6e\u8fdb\u884c\u4fee\u6b63\uff0c\u4ee5\u63d0\u9ad8\u4eba\u8138\u68c0\u6d4b\u7684\u51c6\u786e\u6027\u3002<\/li>\n<\/ul>\n<h5><span class=\"ez-toc-section\" id=\"O-net%EF%BC%9A%E4%BA%BA%E8%84%B8%E8%AF%86%E5%88%AB\"><\/span>O-net\uff1a\u4eba\u8138\u8bc6\u522b<span class=\"ez-toc-section-end\"><\/span><\/h5>\n<ul>\n<li><strong>\u540d\u79f0<\/strong>\uff1a\u8f93\u51fa\u7f51\u7edc\uff08output network\uff09<\/li>\n<li><strong>\u4f5c\u7528<\/strong>\uff1aO\u7f51\u7edc\u901a\u8fc7\u66f4\u6df1\u5c42\u6b21\u7684\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u5904\u7406\u4eba\u8138\u533a\u57df\uff0c\u4f18\u5316\u4eba\u8138\u4f4d\u7f6e\u548c\u59ff\u6001\uff0c\u5e76\u8f93\u51fa\u9762\u90e8\u5173\u952e\u70b9\u4fe1\u606f\uff0c\u4e3a\u540e\u7eed\u7684\u9762\u90e8\u5bf9\u9f50\u63d0\u4f9b\u91cd\u8981\u53c2\u8003\u3002<\/li>\n<\/ul>\n<h4><span class=\"ez-toc-section\" id=\"MTCNN%E7%94%A8%E5%88%B0%E7%9A%84%E4%B8%BB%E8%A6%81%E6%A8%A1%E5%9D%97\"><\/span>MTCNN\u7528\u5230\u7684\u4e3b\u8981\u6a21\u5757<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<h5><span class=\"ez-toc-section\" id=\"%E5%9B%BE%E5%83%8F%E9%87%91%E5%AD%97%E5%A1%94\"><\/span>\u56fe\u50cf\u91d1\u5b57\u5854<span class=\"ez-toc-section-end\"><\/span><\/h5>\n<p>MTCNN\u7684P\u7f51\u7edc\u4f7f\u7528\u7684\u68c0\u6d4b\u65b9\u5f0f\u662f\uff1a\u8bbe\u7f6e\u5efa\u8bae\u6846\uff0c\u7528\u5efa\u8bae\u6846\u5728\u56fe\u7247\u4e0a\u6ed1\u52a8\u68c0\u6d4b\u4eba\u8138<br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u56fe\u50cf\u91d1\u5b57\u5854.jpeg\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u56fe\u50cf\u91d1\u5b57\u5854.jpeg\" alt=\"\" \/><\/a><\/p>\n<blockquote>\n<p>\u7531\u4e8eP\u7f51\u7edc\u7684\u5efa\u8bae\u6846\u7684\u5927\u5c0f\u662f\u56fa\u5b9a\u7684\uff0c\u53ea\u80fd\u68c0\u6d4b12*12\u8303\u56f4\u5185\u7684\u4eba\u8138\uff0c\u6240\u4ee5\u5176\u4e0d\u65ad\u7f29\u5c0f\u56fe\u7247\u4ee5\u9002\u5e94\u4e8e\u5efa\u8bae\u6846\u7684\u5927\u5c0f\uff0c\u5f53\u4e0b\u4e00\u6b21\u56fe\u50cf\u7684\u6700\u5c0f\u8fb9\u957f\u5c0f\u4e8e12\u65f6\uff0c\u505c\u6b62\u7f29\u653e\u3002<\/p>\n<\/blockquote>\n<h5><span class=\"ez-toc-section\" id=\"IOU\"><\/span>IOU<span class=\"ez-toc-section-end\"><\/span><\/h5>\n<p>\u5b9a\u4e49\uff1aIOU\uff08Intersection over Union\uff09\u662f\u6307\u4ea4\u5e76\u6bd4\uff0c\u662f\u76ee\u6807\u68c0\u6d4b\u9886\u57df\u5e38\u7528\u7684\u4e00\u79cd\u8bc4\u4f30\u6307\u6807\uff0c\u7528\u4e8e\u8861\u91cf\u4e24\u4e2a\u8fb9\u754c\u6846\uff08Bounding Box\uff09\u4e4b\u95f4\u7684\u91cd\u53e0\u7a0b\u5ea6\u3002<br \/>\n\u4e24\u79cd\u65b9\u5f0f\uff1a<\/p>\n<ul>\n<li>\u4ea4\u96c6\u6bd4\u5e76\u96c6<\/li>\n<li>\u4ea4\u96c6\u6bd4\u6700\u5c0f\u96c6<br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u4ea4\u5e76\u6bd4\u7684\u4e24\u79cd\u65b9\u5f0f.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u4ea4\u5e76\u6bd4\u7684\u4e24\u79cd\u65b9\u5f0f.png\" alt=\"\" \/><\/a><\/li>\n<\/ul>\n<blockquote>\n<p>O\u7f51\u7edciou\u503c\u5927\u4e8e\u9608\u503c\u7684\u6846\u88ab\u8ba4\u4e3a\u662f\u91cd\u590d\u7684\u6846\u4f1a\u4e22\u5f03\uff0c\u7559\u4e0biou\u503c\u5c0f\u7684\u6846\uff0c\u4f46\u662f\u5982\u679c\u51fa\u73b0\u4e86\u4e0b\u56fe\u4e2d\u5927\u6846\u5957\u5c0f\u6846\u7684\u60c5\u51b5\uff0c\u5219iou\u503c\u504f\u5c0f\u4e5f\u4f1a\u88ab\u4fdd\u7559\uff0c\u662f\u6211\u4eec\u4e0d\u60f3\u770b\u5230\u7684\uff0c\u56e0\u6b64\u6211\u4eec\u5728O\u7f51\u7edc\u91c7\u7528\u4e86\u7b2c\u4e8c\u79cd\u65b9\u5f0f\u7684iou\u4ee5\u63d0\u9ad8\u8bef\u68c0\u7387\u3002<\/p>\n<\/blockquote>\n<h5><span class=\"ez-toc-section\" id=\"NMS%EF%BC%88Non-Maximum_Suppression%EF%BC%8C%E9%9D%9E%E6%9E%81%E5%A4%A7%E5%80%BC%E6%8A%91%E5%88%B6%EF%BC%89\"><\/span>NMS\uff08Non-Maximum Suppression\uff0c\u975e\u6781\u5927\u503c\u6291\u5236\uff09<span class=\"ez-toc-section-end\"><\/span><\/h5>\n<p><strong>\u5b9a\u4e49<\/strong>\uff1a<br \/>\nNMS\u662f\u4e00\u79cd\u76ee\u6807\u68c0\u6d4b\u4e2d\u5e38\u7528\u7684\u6280\u672f\uff0c\u65e8\u5728\u6d88\u9664\u91cd\u53e0\u8f83\u591a\u7684\u5019\u9009\u6846\uff0c\u4fdd\u7559\u6700\u5177\u4ee3\u8868\u6027\u7684\u8fb9\u754c\u6846\uff0c\u4ee5\u63d0\u9ad8\u68c0\u6d4b\u7684\u51c6\u786e\u6027\u548c\u6548\u7387\u3002<\/p>\n<p><strong>\u5de5\u4f5c\u539f\u7406<\/strong>\uff1a<br \/>\nNMS\u7684\u5de5\u4f5c\u539f\u7406\u662f\u901a\u8fc7\u8bbe\u7f6e\u4e00\u4e2a\u9608\u503c\uff0c\u6bd4\u5982IOU\uff08\u4ea4\u5e76\u6bd4\uff09\u9608\u503c\uff0c\u5bf9\u6240\u6709\u5019\u9009\u6846\u6309\u7167\u7f6e\u4fe1\u5ea6\u8fdb\u884c\u6392\u5e8f\uff0c\u7136\u540e\u4ece\u7f6e\u4fe1\u5ea6\u6700\u9ad8\u7684\u5019\u9009\u6846\u5f00\u59cb\uff0c\u5c06\u4e0e\u5176\u91cd\u53e0\u5ea6\u9ad8\u4e8e\u9608\u503c\u7684\u5019\u9009\u6846\u5254\u9664\uff0c\u4fdd\u7559\u7f6e\u4fe1\u5ea6\u6700\u9ad8\u7684\u5019\u9009\u6846\u3002<\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/NMS\u793a\u4f8b.jpeg\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/NMS\u793a\u4f8b.jpeg\" alt=\"\" \/><\/a><\/p>\n<ul>\n<li>\u5982\u4e0a\u56fe\u6240\u793a\u6846\u51fa\u4e86\u4e94\u4e2a\u4eba\u8138\uff0c\u7f6e\u4fe1\u5ea6\u5206\u522b\u4e3a0.98\uff0c0.83\uff0c0.75\uff0c0.81\uff0c0.67\uff0c<strong>\u524d\u4e09\u4e2a<\/strong>\u7f6e\u4fe1\u5ea6\u5bf9\u5e94\u5de6\u4fa7\u7684Rose\uff0c<strong>\u540e\u4e24\u4e2a<\/strong>\u5bf9\u5e94\u53f3\u4fa7\u7684Jack\u3002<\/li>\n<li>NMS\u5c06\u8fd9\u4e94\u4e2a\u6846<strong>\u6839\u636e\u7f6e\u4fe1\u5ea6\u6392\u5e8f<\/strong>\uff0c\u53d6\u51fa\u6700\u5927\u7684\u7f6e\u4fe1\u5ea6(0.98)\u7684\u6846\u5206\u522b\u548c\u5269\u4e0b\u7684\u6846<strong>\u505aiou<\/strong>\uff0c<strong>\u4fdd\u7559iou\u5c0f\u4e8e\u9608\u503c\u7684\u6846<\/strong>(\u4ee3\u7801\u4e2d\u9608\u503c\u8bbe\u7f6e\u7684\u662f0.3)\uff0c\u8fd9\u6837\u5c31\u5269\u4e0b0.81\u548c0.67\u8fd9\u4e24\u4e2a\u6846\u4e86\u3002<\/li>\n<li>\u91cd\u590d\u4e0a\u9762\u7684\u8fc7\u7a0b\uff0c\u53d6\u51fa\u7f6e\u4fe1\u5ea6(0.81)\u5927\u7684\u6846\u548c\u5269\u4e0b\u7684\u6846\u505aiou\uff0c\u4fdd\u7559iou\u5c0f\u4e8e\u9608\u503c\u7684\u6846\u3002\u8fd9\u6837\u6700\u540e\u53ea\u5269\u4e0b0.98\u548c0.81\u8fd9\u4e24\u4e2a\u4eba\u8138\u6846\u4e86\u3002<\/li>\n<\/ul>\n<h4><span class=\"ez-toc-section\" id=\"%E4%BB%A3%E7%A0%81%E5%AE%9E%E7%8E%B0\"><\/span>\u4ee3\u7801\u5b9e\u73b0<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<p>P-Net<br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/PNet\u7f51\u7edc\u7ed3\u6784.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/PNet\u7f51\u7edc\u7ed3\u6784.png\" alt=\"\" \/><\/a><\/p>\n<pre><code class=\"language-python\">import torch\nfrom torch import nn\n\n&quot;&quot;&quot;\n    P-Net\n&quot;&quot;&quot;\n\nclass PNet(nn.Module):\n    def __init__(self):\n        super().__init__()\n        self.features_extractor = nn.Sequential(\n            # \u7b2c\u4e00\u5c42\u5377\u79ef\n            nn.Conv2d(in_channels=3, out_channels=10, kernel_size=3, stride=1, padding=0),\n            nn.BatchNorm2d(num_features=10),\n            nn.ReLU(),\n\n            # \u7b2c\u4e00\u5c42\u6c60\u5316\n            nn.MaxPool2d(kernel_size=3,stride=2, padding=1),\n\n            # \u7b2c\u4e8c\u5c42\u5377\u79ef\n            nn.Conv2d(in_channels=10, out_channels=16, kernel_size=3, stride=1, padding=0),\n            nn.BatchNorm2d(num_features=16),\n            nn.ReLU(),\n\n            # \u7b2c\u4e09\u5c42\u5377\u79ef\n            nn.Conv2d(in_channels=16, out_channels=32, kernel_size=3, stride=1, padding=0),\n            nn.BatchNorm2d(num_features=32),\n            nn.ReLU()\n        )\n\n        # \u6982\u7387\u8f93\u51fa\n        self.cls_out = nn.Conv2d(in_channels=32, out_channels=2, kernel_size=1, stride=1, padding=0)\n        # \u56de\u5f52\u91cf\u8f93\u51fa\n        self.reg_out = nn.Conv2d(in_channels=32, out_channels=4, kernel_size=1, stride=1, padding=0)\n\n    def forward(self, x):\n        print(x.shape)\n        x = self.features_extractor(x)\n        cls_out = self.cls_out(x)\n        reg_out = self.reg_out(x)\n\n        return cls_out, reg_out<\/code><\/pre>\n<p>R-Net<br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/RNet\u7f51\u7edc\u7ed3\u6784.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/RNet\u7f51\u7edc\u7ed3\u6784.png\" alt=\"\" \/><\/a><\/p>\n<pre><code class=\"language-python\">import torch\nfrom torch import nn\n\nclass RNet(nn.Module):\n    def __init__(self):\n        super().__init__()\n\n        self.feature_extractor = nn.Sequential(\n            # \u7b2c\u4e00\u5c42\u5377\u79ef 24 x 24\n            nn.Conv2d(in_channels=3, out_channels=28, kernel_size=3, stride=1, padding=0),\n            nn.BatchNorm2d(num_features=28),\n            nn.ReLU(),\n\n            # \u7b2c\u4e00\u5c42\u6c60\u5316 11 x 11\n            nn.MaxPool2d(kernel_size=3, stride=2, padding=1, ceil_mode=False),\n\n            # \u7b2c\u4e8c\u5c42\u5377\u79ef 9 x 9\n            nn.Conv2d(in_channels=28, out_channels=48, kernel_size=3, stride=1, padding=0),\n            nn.BatchNorm2d(num_features=48),\n            nn.ReLU(),\n\n            # \u7b2c\u4e8c\u5c42\u6c60\u5316 (\u6ca1\u6709\u8865\u96f6) 4 x 4\n            nn.MaxPool2d(kernel_size=3, stride=2, padding=0, ceil_mode=False),\n\n            # \u7b2c\u4e09\u5c42\u5377\u79ef 3 x 3\n            nn.Conv2d(in_channels=48, out_channels=64, kernel_size=2, stride=1, padding=0),\n            nn.BatchNorm2d(num_features=64),\n            nn.ReLU(),\n\n            # \u5c55\u5e73\n            nn.Flatten(),\n\n            # \u5168\u8fde\u63a5\u5c42 [batch_size, 128]\n            nn.Linear(in_features=3 * 3 * 64, out_features=128)\n        )\n\n        # \u6982\u7387\u8f93\u51fa\n        self.cls_out = nn.Linear(in_features=128, out_features=1)\n\n        # \u56de\u5f52\u91cf\u8f93\u51fa\n        self.reg_out = nn.Linear(in_features=128, out_features=4)\n\n    def forward(self, x):\n        x = self.feature_extractor(x)\n        cls = self.cls_out(x)\n        reg = self.reg_out(x)\n        return cls, reg   <\/code><\/pre>\n<p>O-Net<br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/ONet\u7f51\u7edc\u7ed3\u6784.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/ONet\u7f51\u7edc\u7ed3\u6784.png\" alt=\"\" \/><\/a><\/p>\n<pre><code class=\"language-python\">import torch\nfrom torch import nn\n\nclass ONet(nn.Module):\n    def __init__(self):\n        super().__init__()\n        self.feature_extractor = nn.Sequential(\n            # \u7b2c1\u5c42\u5377\u79ef 48 x 48\n            nn.Conv2d(in_channels=3, out_channels=32, kernel_size=3, stride=1, padding=0),\n            nn.BatchNorm2d(num_features=32),\n            nn.ReLU(),\n\n            # \u7b2c1\u5c42\u6c60\u5316 11 x 11\n            nn.MaxPool2d(kernel_size=3, stride=2, padding=1, ceil_mode=False),\n\n            # \u7b2c2\u5c42\u5377\u79ef 9 x 9\n            nn.Conv2d(in_channels=32, out_channels=64, kernel_size=3, stride=1, padding=0),\n            nn.BatchNorm2d(num_features=64),\n            nn.ReLU(),\n\n            # \u7b2c2\u5c42\u6c60\u5316 \n            nn.MaxPool2d(kernel_size=3, stride=2, padding=0, ceil_mode=False),\n\n            # \u7b2c3\u5c42\u5377\u79ef \n            nn.Conv2d(in_channels=64, out_channels=64, kernel_size=3, stride=1, padding=0),\n            nn.BatchNorm2d(num_features=64),\n            nn.ReLU(),\n\n            # \u7b2c3\u5c42\u6c60\u5316 \n            nn.MaxPool2d(kernel_size=2, stride=2, padding=0, ceil_mode=False),\n\n            # \u7b2c4\u5c42\u5377\u79ef\n            nn.Conv2d(in_channels=64, out_channels=128, kernel_size=2, stride=1, padding=0),\n            nn.BatchNorm2d(num_features=128),\n            nn.ReLU(),\n\n            # \u5c55\u5e73 [batch_size, n_features]\n            nn.Flatten(),\n\n            # \u5168\u8fde\u63a5 [batch_size, 128]\n            nn.Linear(in_features=3 * 3 * 128, out_features=256)\n\n        )\n\n        # \u6982\u7387\u8f93\u51fa\n        self.cls_out = nn.Linear(in_features=256, out_features=1)\n\n        # \u56de\u5f52\u91cf\u8f93\u51fa\n        self.reg_out = nn.Linear(in_features=256, out_features=4)\n\n        # \u5173\u952e\u70b9\u8f93\u51fa\n        self.landmark_out = nn.Linear(in_features=256, out_features=10)\n\n    def forward(self, x):\n        x = self.feature_extractor(x)\n        cls = self.cls_out(x)\n        reg = self.reg_out(x)\n        landmark = self.landmark_out(x)\n        return cls, reg, landmark<\/code><\/pre>\n<h3><span class=\"ez-toc-section\" id=\"MTCNN%E8%AE%AD%E7%BB%83%E9%80%BB%E8%BE%91\"><\/span>MTCNN\u8bad\u7ec3\u903b\u8f91<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<h4><span class=\"ez-toc-section\" id=\"%E5%87%86%E5%A4%87%E8%AE%AD%E7%BB%83%E6%95%B0%E6%8D%AE%E9%9B%86\"><\/span>\u51c6\u5907\u8bad\u7ec3\u6570\u636e\u96c6<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<p>\u9996\u5148\uff0c\u6211\u4eec\u9700\u8981\u51c6\u5907\u4eba\u8138\u6807\u6ce8\u597d\u7684\u6570\u636e\u96c6,\u4eba\u8138\u8bc6\u522b\u6807\u6ce8\u597d\u7684\u6570\u636e\u96c6\u6bd4\u8f83\u6709\u540d\u662f\uff1a<a href=\"http:\/\/shuoyang1213.me\/WIDERFACE\/\">WIDEERFACE<\/a>\u548c<a href=\"http:\/\/mmlab.ie.cuhk.edu.hk\/projects\/CelebA.html\">CelebA<\/a>\u3002<br \/>\n\u672c\u6b21\u6211\u4eec\u4f7f\u7528CelebA\u6570\u636e\u96c6\u3002<\/p>\n<h5><span class=\"ez-toc-section\" id=\"CelebA%E6%95%B0%E6%8D%AE%E9%9B%86%E7%AE%80%E4%BB%8B\"><\/span>CelebA\u6570\u636e\u96c6\u7b80\u4ecb<span class=\"ez-toc-section-end\"><\/span><\/h5>\n<p>CelebA\u6570\u636e\u96c6\u662f\u7531\u9999\u6e2f\u4e2d\u6587\u5927\u5b66\u591a\u5a92\u4f53\u5b9e\u9a8c\u5ba4\u53d1\u5e03\u7684\u5927\u89c4\u6a21\u4eba\u8138\u5c5e\u6027\u6570\u636e\u96c6\uff0c\u5305\u542b\u8d85\u8fc7 20 \u4e07\u5f20\u540d\u4eba\u56fe\u50cf\uff0c\u6bcf\u5f20\u56fe\u50cf\u6709 40 \u4e2a\u5c5e\u6027\u6ce8\u91ca\u3002CelebA\u6570\u636e\u96c6\u5168\u62fc\u662fLarge-scale CelebFaces Attributes (CelebA) Dataset\u3002 \u8be5\u6570\u636e\u96c6\u4e2d\u7684\u56fe\u50cf\u6db5\u76d6\u4e86\u4e30\u5bcc\u7684\u4eba\u4f53\u59ff\u52bf\u53d8\u5316\u548c\u590d\u6742\u591a\u6837\u7684\u80cc\u666f\u4fe1\u606f\u3002\u6db5\u76d6\u4e86\u5206\u7c7b\u3001\u76ee\u6807\u68c0\u6d4b\u548c\u5173\u952e\u70b9\u68c0\u6d4b\u7b49\u6570\u636e\u3002<\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/CelebA\u7f51\u7ad9.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/CelebA\u7f51\u7ad9.png\" alt=\"\" \/><\/a><\/p>\n<h5><span class=\"ez-toc-section\" id=\"CelebA%E6%95%B0%E6%8D%AE%E9%9B%86%E4%B8%8B%E8%BD%BD\"><\/span>CelebA\u6570\u636e\u96c6\u4e0b\u8f7d<span class=\"ez-toc-section-end\"><\/span><\/h5>\n<p>\u4e0b\u8f7d\u5730\u5740\uff1a<\/p>\n<ul>\n<li>\u53ef\u4ee5\u5728<a href=\"https:\/\/mmlab.ie.cuhk.edu.hk\/projects\/CelebA.html\">CelebA\u5b98\u7f51<\/a>\u627e\u5230\u8c37\u6b4c\u7f51\u76d8\u4e0b\u8f7d\u94fe\u63a5\u6216\u767e\u5ea6\u7f51\u76d8\u4e0b\u8f7d\u94fe\u63a5\u3002<\/li>\n<\/ul>\n<h4><span class=\"ez-toc-section\" id=\"%E4%B8%8B%E8%BD%BD%E5%92%8C%E5%87%86%E5%A4%87%E8%AE%AD%E7%BB%83%E9%9B%86\"><\/span>\u4e0b\u8f7d\u548c\u51c6\u5907\u8bad\u7ec3\u96c6<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<p>\u7b2c\u4e00\u6b65\uff1a\u4e0b\u8f7d\u6587\u4ef6\u540e\u89e3\u538b\uff0c\u89e3\u538b\u540e\u76ee\u5f55\u7ed3\u6784\u5982\u4e0b<\/p>\n<pre><code class=\"language-shell\">CelebA\n  |-Anno\n    |-identity_CelebA.txt # \u56fe\u7247\u6807\u6ce8\u7684\u8eab\u4efd\u4fe1\u606f\n    |-list_attr_celeba.txt # \u56fe\u7247\u6807\u6ce8\u7684\u5c5e\u6027\u4fe1\u606f\n    |-list_bbox_celeba.txt # \u56fe\u7247\u6807\u6ce8\u7684\u4eba\u8138\u6846\n    |-list_landmarks_align_celeba.txt # \u56fe\u7247\u6807\u6ce8\u7684\u4eba\u8138\u5173\u952e\u70b9(\u5bf9\u9f50\u56fe\u7247)\n    |-list_landmarks_celeba.txt # \u56fe\u7247\u6807\u6ce8\u7684\u4eba\u8138\u5173\u952e\u70b9\n  |-Eval\n    |-list_eval_partition.txt # \u56fe\u7247\u5212\u5206\u8bad\u7ec3\u96c6\u3001\u9a8c\u8bc1\u96c6\u3001\u6d4b\u8bd5\u96c6\n  |-Img\n    |-img_align_celeba # \u56fe\u7247\u7ecf\u8fc7\u5bf9\u9f50\u540e\u7684\n    |-img_celeba # \u539f\u59cb\u56fe\u7247(\u672a\u505a\u5bf9\u9f50\u5904\u7406)\n    |-img_celeba_png # PNG\u683c\u5f0f\u7684\u56fe\u7247<\/code><\/pre>\n<p>\u7b2c\u4e8c\u6b65\uff1a\u5c06\u6240\u9700\u7684\u56fe\u7247\u4ee5\u53ca\u6807\u51c6\u4fe1\u606f\u62f7\u8d1d\u5230\u4ee3\u7801\u5de5\u7a0b\u9879\u76ee\u4e0b\uff0c\u5e76\u8c03\u6574\u76ee\u5f55\u7ed3\u6784\u5982\u4e0b\uff1a<\/p>\n<pre><code class=\"language-shell\">\u4ee3\u7801\u6839\u76ee\u5f55\n  |-datasets\n    |-celeba\n      |-identity_CelebA.txt\n      |-list_attr_celeba.txt\n      |-list_bbox_celeba.txt\n      |-list_landmarks_align_celeba.txt\n      |-list_landmarks_celeba.txt\n      |-Img\n        |-img_celeba\n  |-*.py \u4ee3\u7801\u6587\u4ef6  <\/code><\/pre>\n<blockquote>\n<p>\u8bf4\u660e\uff1a\u5728\u4f7f\u7528\u6570\u636e\u96c6\u65f6\u53ef\u4ee5\u8c03\u6574\u6210\u5982\u4e0b\u7684\u76ee\u5f55\u7ed3\u6784\uff0c\u8fd9\u6837\u6211\u4eec\u53ef\u4ee5\u5728Jupyter Notebook\u4e2d\u4f7f\u7528CelebA\u7684API\u67e5\u770b\u56fe\u7247\u7684\u4fe1\u606f\u3002<\/p>\n<\/blockquote>\n<p>\u7b2c\u4e09\u6b65\uff1a\u4f7f\u7528CelebA\u7684API\u67e5\u770b\u56fe\u7247\u7684\u4fe1\u606f\u3002<\/p>\n<pre><code class=\"language-python\">import numpy as np\nfrom torchvision.datasets import CelebA\nimport matplotlib.pyplot as plt\nimport cv2\nfrom PIL import Image\n\nroot_dir = &#039;datasets&#039;   # jupyter notebook\u7684\u6587\u4ef6\u653e\u5728\u4e0edatasets\u540c\u4e00\u7ea7\u76ee\u5f55\u4e0b\nceleba = CelebA(root=root_dir, split=&#039;train&#039;,\n                target_type=[&#039;attr&#039;, &#039;identity&#039;, &#039;bbox&#039;, &#039;landmarks&#039;],\n                download=False)\nattr_names = celeba.attr_names\nattr_names.pop()\nattr_names = np.array(attr_names)    \nfig = plt.figure(figsize=(14, 7))\n\nn = 4 # \u663e\u793a\u7684\u56fe\u7247\u6570\u91cf\nfor idx in range(n):\n    img, (attr, identity, bbox, landmarks) = celeba[idx]  # \u8bfb\u53d6\u56fe\u7247\u548c\u76f8\u5173\u6807\u7b7e\u503c\n    ax = fig.add_subplot(1, 4, idx + 1)\n    # \u4e0d\u663e\u793a\u523b\u5ea6\u6807\u7b7e\u548c\u8fb9\u6846\n    ax.set_xticks([])\n    ax.set_yticks([])\n    ax.set_frame_on(b=False)\n\n    # \u5c06CelebA\u6570\u636e\u96c6\u8bfb\u53d6\u5230\u7684PIL\u56fe\u7247\u683c\u5f0f\u8f6c\u6362\u6210OprnCV\u6240\u9700\u683c\u5f0f\n    img_cv2 = cv2.cvtColor(src=np.asanyarray(img), code=cv2.COLOR_RGB2BGR)\n\n    # \u7ed8\u5236\u7279\u5f81\u70b9\n    landmarks = landmarks.numpy()\n    for idx, point in enumerate(landmarks):\n        if idx % 2 == 0:\n            cv2.circle(img=img_cv2, center=(point, landmarks[idx + 1]),\n                        radius=1, color=(255, 0, 0), thickness=2)\n    attr_list = attr.numpy()\n    attrs = attr_names[attr_list==1]\n    label = &#039;&#039;     \n    # \u5c5e\u6027\u6807\u7b7e\n    for att in attrs:\n        label = label + att + &#039;\\n&#039;\n    ax.set_xlabel(label)\n    # \u8eab\u4efdID\n    cele_id = identity.numpy()\n    ax.set_title(f&#039;ID: {cele_id}&#039;)\n    # \u5c06OpenCV\u56fe\u7247\u518d\u6b21\u8f6c\u6210\u6210Pillow\u56fe\u7247\u683c\u5f0f\n    img_pil = Image.fromarray( cv2.cvtColor(src=img_cv2,\n                                            code=cv2.COLOR_BGR2RGB,))\n    ax.imshow(img_pil)\nplt.show()\n<\/code><\/pre>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/CelebA\u4fe1\u606f\u5c55\u793a1.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/CelebA\u4fe1\u606f\u5c55\u793a1.png\" alt=\"\" \/><\/a><\/p>\n<pre><code class=\"language-python\"># \u663e\u793a\u6807\u6ce8\u6846\nfig = plt.figure(figsize=(14, 7))\nfor idx in range(n):\n    img, (attr, identity, bbox, landmarks) = celeba[idx]\n    ax = fig.add_subplot(1, 4, idx + 1)\n    ax.set_xticks([])\n    ax.set_yticks([])\n    ax.set_frame_on(b=False)\n\n    # \u56fe\u7247\u7684\u8bfb\u53d6\u6539\u4e3a\u4f7f\u7528\u539f\u56fe\u50cf\n    file_path = os.path.join(celeba.root, celeba.base_folder, &#039;img_celeba&#039;, celeba.filename[idx])\n\n    img_cv2 = cv2.imread(file_path)\n    bbox = bbox.numpy()\n    cv2.rectangle(img=img_cv2, pt1=(bbox[0], bbox[1]),\n                  pt2=(bbox[0] + bbox[2], bbox[1] + bbox[3]),\n                  color=(255, 0, 0), thickness=2)\n    cele_id = identity.numpy()\n    ax.set_title(f&#039;ID: {cele_id}&#039;)\n    img_pil = Image.fromarray(cv2.cvtColor(src=img_cv2,\n                                           code=cv2.COLOR_BGR2RGB,))\n    ax.imshow(img_pil)\nplt.show()<\/code><\/pre>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/CelebA\u4fe1\u606f2.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/CelebA\u4fe1\u606f2.png\" alt=\"\" \/><\/a><\/p>\n<h4><span class=\"ez-toc-section\" id=\"%E8%AE%AD%E7%BB%83%E9%9B%86%E9%A2%84%E5%A4%84%E7%90%86\"><\/span>\u8bad\u7ec3\u96c6\u9884\u5904\u7406<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<p>\u56e0\u4e3a\u56fe\u7247\u6807\u6ce8\u7684\u4fe1\u606f\u5404\u79cd\u5404\u6837\uff0c\u4e0d\u80fd\u76f4\u63a5\u704c\u5165\u6a21\u578b\u4e2d\u8bad\u7ec3\uff0c\u6240\u4ee5\u9700\u8981\u8fdb\u884c\u9884\u5904\u7406\u3002<\/p>\n<blockquote>\n<p>\u4f20\u7ed9\u673a\u5668\u7684\u6570\u636e\u6700\u597d\u662f\u4ee50\u4e3a\u4e2d\u5fc3\u7684\uff0c\u4e14\u5f52\u4e00\u5316\u5230[-1, 1]\u4e4b\u95f4\u7684\u6570\u636e\u3002<\/p>\n<\/blockquote>\n<p>\u9884\u5904\u7406\u8fc7\u7a0b\u5927\u81f4\u5982\u4e0b\uff1a<\/p>\n<ul>\n<li>\u5728datasets\u76ee\u5f55\u4e0b\u521b\u5efatrain\/12\u3001train\/24\u3001train\/48\u6587\u4ef6\u5939\uff0c\u5206\u522b\u5b58\u653e12\u300124\u300148\u5927\u5c0f\u7684\u8bad\u7ec3\u6570\u636e\u3002<\/li>\n<li>\u8bfb\u53d6\u6807\u6ce8\u4fe1\u606f\u548c\u5173\u952e\u70b9\u4fe1\u606f<\/li>\n<li>\u8bfb\u53d6\u56fe\u7247<\/li>\n<li>\u6839\u636e\u56fe\u7247\u7684\u4fe1\u606f\uff0c\u968f\u673a\u751f\u62105\u4e2a\u5019\u9009\u88c1\u526a\u6846\n<ul>\n<li>\u5bf9\u5019\u9009\u88c1\u526a\u6846\u4e0e\u539f\u59cb\u6807\u6ce8\u6846\u8fdb\u884cIOU\u8ba1\u7b97<\/li>\n<li>\u5982\u679ciou&gt;0.7\uff0c\u4e3a\u6b63\u6837\u672c\uff1b\u5982\u679c0.4 &lt; iou &lt; 0.6\uff0c\u4e3a\u504f\u6837\u672c\uff1b\u5982\u679ciou&lt;0.4\uff0c\u4e3a\u8d1f\u6837\u672c<\/li>\n<\/ul>\n<\/li>\n<li>\u5c06\u751f\u6210\u7684\u6837\u672c\u6309\u7c7b\u4fdd\u5b58\u5728train\/12\/\u3001train\/24\/\u3001train\/48\/\u6587\u4ef6\u5939\u4e2d\u3002<\/li>\n<\/ul>\n<blockquote>\n<p>\u7531\u4e8eMTCNN\u539f\u59cb\u9879\u76ee\u4ee3\u7801\u53ef\u8bfb\u6027\u4e0d\u5f3a\uff0c\u6211\u5bf9\u9884\u5904\u7406\u8fc7\u7a0b\u8fdb\u884c\u4e86\u91cd\u6784\uff0c\u9884\u5904\u7406\u7684\u4ee3\u7801\u5982\u4e0b\u3002\u4ee5\u4e0b\u4ee3\u7801\u4e5f\u53ef\u4ee5\u53ef\u4ee5\u67e5\u770bGithub\u4ed3\u5e93\uff1a<a href=\"https:\/\/github.com\/domonic18\/detect_face_mtcnn\">Github\uff1adetect_face_mtcnn<\/a><\/p>\n<\/blockquote>\n<pre><code class=\"language-python\">import os\nimport random\nimport numpy as np\nimport torch\nfrom PIL import Image\nfrom utils.tool import iou as IOU\n\ncurrent_path = os.path.dirname(os.path.abspath(__file__))\nBASE_PATH = os.path.join(current_path, &quot;datasets&quot;)\nTARGET_PATH = os.path.join(BASE_PATH, &quot;celeba&quot;)\nIMG_PATH = os.path.join(BASE_PATH, &quot;celeba\/img_celeba&quot;)\nDST_PATH = os.path.join(BASE_PATH, &quot;train&quot;)\nLABEL_PATH = os.path.join(TARGET_PATH, &quot;list_bbox_celeba.txt&quot;)\nLANMARKS_PATH = os.path.join(TARGET_PATH, &quot;list_landmarks_celeba.txt&quot;)\n\n# \u6d4b\u8bd5\u6837\u672c\u4e2a\u6570\u9650\u5236,\u8bbe\u7f6e\u4e3a -1 \u8868\u793a\u5168\u90e8\nTEST_SAMPLE_LIMIT = 100\n\n# \u4e3a\u968f\u673a\u6570\u79cd\u5b50\u505a\u51c6\u5907\uff0c\u4f7f\u6b63\u6837\u672c\uff0c\u90e8\u5206\u6837\u672c\uff0c\u8d1f\u6837\u672c\u7684\u6bd4\u4f8b\u4e3a1\uff1a1\uff1a3\nfloat_num = [0.1, 0.1, 0.3, 0.5, 0.95, 0.95, 0.99, 0.99, 0.99, 0.99]\n\ndef create_directories(base_path, face_size):\n    paths = {}\n    base_path = os.path.join(base_path, f&quot;{face_size}&quot;)\n    if not os.path.exists(base_path):\n        os.makedirs(base_path)\n\n    paths[&#039;positive&#039;] = os.path.join(base_path, &quot;positive&quot;)\n    paths[&#039;negative&#039;] = os.path.join(base_path, &quot;negative&quot;)\n    paths[&#039;part&#039;] = os.path.join(base_path, &quot;part&quot;)\n\n    for path in paths.values():\n        if not os.path.exists(path):\n            os.makedirs(path)\n\n    return paths, base_path\n\ndef open_label_files(base_path):\n    files = {}\n    files[&#039;positive&#039;] = open(os.path.join(base_path, &quot;positive.txt&quot;), &quot;w&quot;)\n    files[&#039;negative&#039;] = open(os.path.join(base_path, &quot;negative.txt&quot;), &quot;w&quot;)\n    files[&#039;part&#039;] = open(os.path.join(base_path, &quot;part.txt&quot;), &quot;w&quot;)\n    return files\n\ndef parse_annotation_line(line):\n    strs = line.strip().split()\n    return strs\n\ndef adjust_bbox(x1, y1, w, h):\n    # \u6807\u6ce8\u4e0d\u6807\u51c6\uff0c\u7ed9\u6846\u9002\u5f53\u7684\u504f\u79fb\u91cf\n    x1 = int(x1 + w * 0.12)\n    y1 = int(y1 + h * 0.1)\n\n    x2 = int(x1 + w * 0.9)\n    y2 = int(y1 + h * 0.85)\n\n    w = int(x2 - x1)\n    h = int(y2 - y1)\n\n    return x1, y1, x2, y2, w, h\n\ndef generate_crop_boxes(cx, cy, max_side, img_w, img_h):\n    &quot;&quot;&quot;\n    \u6839\u636e\u7ed9\u5b9a\u7684\u4eba\u8138\u4e2d\u5fc3\u70b9\u5750\u6807\u548c\u5c3a\u5bf8,\u751f\u62105\u4e2a\u5019\u9009\u7684\u88c1\u526a\u6846\u3002\n\n    \u53c2\u6570:\n    cx (float): \u4eba\u8138\u4e2d\u5fc3\u70b9\u7684 x \u5750\u6807\n    cy (float): \u4eba\u8138\u4e2d\u5fc3\u70b9\u7684 y \u5750\u6807\n    max_side (int): \u4eba\u8138\u6846\u7684\u6700\u5927\u8fb9\u957f\n    img_w (int): \u56fe\u50cf\u5bbd\u5ea6\n    img_h (int): \u56fe\u50cf\u9ad8\u5ea6\n\n    \u8fd4\u56de:\n    crop_boxes (list): \u4e00\u4e2a\u5305\u542b5\u4e2a\u88c1\u526a\u6846\u5750\u6807\u7684\u5217\u8868,\u6bcf\u4e2a\u88c1\u526a\u6846\u7684\u683c\u5f0f\u4e3a [x1, y1, x2, y2]\n    &quot;&quot;&quot;\n\n    crop_boxes = []\n    for _ in range(5):\n        # \u968f\u673a\u504f\u79fb\u4e2d\u5fc3\u70b9\u5750\u6807\u4ee5\u53ca\u8fb9\u957f\n        seed = float_num[np.random.randint(0, len(float_num))]\n\n        # \u6700\u5927\u8fb9\u957f\u968f\u673a\u504f\u79fb\n        _max_side = max_side + np.random.randint(int(-max_side * seed), int(max_side * seed))\n\n        # \u4e2d\u5fc3\u70b9x\u5750\u6807\u968f\u673a\u504f\u79fb\n        _cx = cx + np.random.randint(int(-cx * seed), int(cx * seed))\n\n        # \u4e2d\u5fc3\u70b9y\u5750\u6807\u968f\u673a\u504f\u79fb\n        _cy = cy + np.random.randint(int(-cy * seed), int(cy * seed))\n\n        # \u5f97\u5230\u504f\u79fb\u540e\u7684\u5750\u6807\u503c\uff08\u65b9\u6846\uff09\n        _x1 = _cx - _max_side \/ 2\n        _y1 = _cy - _max_side \/ 2\n        _x2 = _x1 + _max_side\n        _y2 = _y1 + _max_side\n\n        # \u504f\u79fb\u8fc7\u5927\uff0c\u504f\u51fa\u56fe\u50cf\u4e86\uff0c\u6b64\u65f6\uff0c\u4e0d\u80fd\u7528\uff0c\u5e94\u8be5\u518d\u6b21\u5c1d\u8bd5\u504f\u79fb\n        if _x1 &lt; 0 or _y1 &lt; 0 or _x2 &gt; img_w or _y2 &gt; img_h:\n            continue\n\n        # \u6dfb\u52a0\u88c1\u526a\u6846\u5750\u6807\u5230\u5217\u8868\u4e2d\n        crop_boxes.append(np.array([_x1, _y1, _x2, _y2]))\n\n    return crop_boxes\n\ndef process_crop_box(img, face_size, max_side, crop_box, boxes, landmarks):\n    &quot;&quot;&quot;\n    \u5904\u7406\u5355\u4e2a\u88c1\u526a\u6846,\u751f\u6210\u6b63\u8d1f\u6837\u672c\u3002\n\n    \u53c2\u6570:\n    img (Image): \u539f\u59cb\u56fe\u50cf\n    crop_box (list): \u88c1\u526a\u6846\u5750\u6807 [x1, y1, x2, y2]\n    boxes (list): \u4eba\u8138\u6846\u5750\u6807\u5217\u8868\n    face_size (int): \u751f\u6210\u7684\u4eba\u8138\u56fe\u50cf\u5c3a\u5bf8\n\n    \u8fd4\u56de:\n    sample (dict): \u6837\u672c\u4fe1\u606f {&#039;image&#039;: image, &#039;label&#039;: label, &#039;bbox_offsets&#039;: offsets, &#039;landmark_offsets&#039;: landmark_offsets}\n    &quot;&quot;&quot;\n    x1, y1, x2, y2 = boxes[0][:4]\n    _x1, _y1, _x2, _y2 = crop_box[:4]\n    px1, py1, px2, py2, px3, py3, px4, py4, px5, py5 = landmarks\n    _max_side = max_side\n\n    offset_x1 = (x1 - _x1) \/ _max_side\n    offset_y1 = (y1 - _y1) \/ _max_side\n    offset_x2 = (x2 - _x2) \/ _max_side\n    offset_y2 = (y2 - _y2) \/ _max_side\n\n    offset_px1 = (px1 - _x1) \/ _max_side\n    offset_py1 = (py1 - _y1) \/ _max_side\n    offset_px2 = (px2 - _x1) \/ _max_side\n    offset_py2 = (py2 - _y1) \/ _max_side\n    offset_px3 = (px3 - _x1) \/ _max_side\n    offset_py3 = (py3 - _y1) \/ _max_side\n    offset_px4 = (px4 - _x1) \/ _max_side\n    offset_py4 = (py4 - _y1) \/ _max_side\n    offset_px5 = (px5 - _x1) \/ _max_side\n    offset_py5 = (py5 - _y1) \/ _max_side\n\n    face_crop = img.crop(crop_box)\n    face_resize = face_crop.resize((face_size, face_size), Image.Resampling.LANCZOS)\n\n    iou = IOU(torch.tensor([x1, y1, x2, y2]), torch.tensor([crop_box[:4]]))\n\n    if iou &gt; 0.7:  # \u6b63\u6837\u672c\n        label = 1\n    elif 0.4 &lt; iou &lt; 0.6:  # \u90e8\u5206\u6837\u672c\n        label = 2\n    elif iou &lt; 0.2:  # \u8d1f\u6837\u672c\n        label = 0\n    else:\n        return None  # \u4e0d\u7b26\u5408\u4efb\u4f55\u6761\u4ef6\u7684\u6837\u672c\u4e0d\u5904\u7406\n\n    return {\n        &#039;image&#039;: face_resize,\n        &#039;label&#039;: label,\n        &#039;bbox_offsets&#039;: (offset_x1, offset_y1, offset_x2, offset_y2),\n        &#039;landmark_offsets&#039;: (offset_px1, offset_py1, offset_px2, offset_py2, offset_px3, offset_py3, offset_px4, offset_py4, offset_px5, offset_py5)\n    }\ndef process_annotation(face_size, anno_line, landmarks):\n    &quot;&quot;&quot;\n    \u5904\u7406\u5355\u884c\u6ce8\u91ca\u4fe1\u606f,\u751f\u6210\u6b63\u8d1f\u6837\u672c\u3002\n\n    \u53c2\u6570:\n    anno_line (str): \u4e00\u884c\u6ce8\u91ca\u4fe1\u606f,\u683c\u5f0f\u4e3a &quot;image_filename x1 y1 w h&quot;\n    face_size (int): \u751f\u6210\u7684\u4eba\u8138\u56fe\u50cf\u5c3a\u5bf8\n    landmarks (str): \u5173\u952e\u70b9\u6807\u6ce8\u5b57\u7b26\u4e32\n\n    \u8fd4\u56de:\n    samples (list): \u751f\u6210\u7684\u6837\u672c\u5217\u8868\n    &quot;&quot;&quot;\n    # 5\u4e2a\u5173\u952e\u70b9\n    _landmarks = landmarks.split()\n\n    # \u4f7f\u7528\u5217\u8868\u89e3\u6790\u548c\u89e3\u5305\u4e00\u6b21\u6027\u83b7\u53d6\u6240\u6709\u5173\u952e\u70b9\u7684\u5750\u6807\n    landmarks = [float(x) for x in _landmarks[1:11]]\n\n    # \u89e3\u6790\u6ce8\u91ca\u884c,\u83b7\u53d6\u56fe\u50cf\u6587\u4ef6\u540d\u548c\u4eba\u8138\u4f4d\u7f6e\u4fe1\u606f\n    strs = parse_annotation_line(anno_line)\n    image_filename = strs[0].strip()\n    x1, y1, w, h = map(int, strs[1:])\n\n    # \u6807\u7b7e\u77eb\u6b63\n    x1, y1, x2, y2, w, h = adjust_bbox(x1, y1, w, h)\n    boxes = [[x1, y1, x2, y2]]\n\n    # \u8ba1\u7b97\u4eba\u8138\u4e2d\u5fc3\u70b9\u5750\u6807\n    cx = w \/ 2 + x1\n    cy = h \/ 2 + y1\n\n    # \u6700\u5927\u8fb9\u957f\n    max_side = max(w, h)\n\n    # \u6253\u5f00\u56fe\u50cf\u6587\u4ef6\n    image_filepath = os.path.join(IMG_PATH, image_filename)\n    with Image.open(image_filepath) as img:\n        # \u89e3\u6790\u51fa\u5bbd\u5ea6\u548c\u9ad8\u5ea6\n        img_w, img_h = img.size\n        # \u751f\u6210\u5019\u9009\u7684\u88c1\u526a\u6846\n        samples = []\n        for crop_box in generate_crop_boxes(cx, cy, max_side, img_w, img_h):\n            # \u5904\u7406\u6bcf\u4e2a\u5019\u9009\u88c1\u526a\u6846,\u751f\u6210\u6b63\u8d1f\u6837\u672c\n            sample = process_crop_box(img, face_size, max_side, crop_box, boxes, landmarks )\n            if sample:\n                samples.append(sample)\n\n    return samples\n\ndef save_samples(samples, files, base_path, counters):\n    &quot;&quot;&quot;\n    \u4fdd\u5b58\u6b63\u8d1f\u6837\u672c\u5230\u6587\u4ef6\u4e2d\u3002\n\n    \u53c2\u6570:\n    samples (list): \u6837\u672c\u5217\u8868, \u6bcf\u4e2a\u5143\u7d20\u4e3a\u4e00\u4e2a\u5b57\u5178, \u5305\u542b &#039;image&#039;, &#039;label&#039;, &#039;bbox_offsets&#039;, &#039;landmark_offsets&#039;\n    files (dict): \u5305\u542b\u6b63\u8d1f\u6837\u672c\u8f93\u51fa\u6587\u4ef6\u7684\u5b57\u5178\n    base_path (str): \u8f93\u51fa\u6587\u4ef6\u7684\u57fa\u7840\u8def\u5f84\n    counters (dict): \u6837\u672c\u8ba1\u6570\u5668\u5b57\u5178\n    &quot;&quot;&quot;\n    for sample in samples:\n        image = sample[&#039;image&#039;]\n        label = sample[&#039;label&#039;]\n        bbox_offsets = sample[&#039;bbox_offsets&#039;]\n        landmark_offsets = sample[&#039;landmark_offsets&#039;]\n\n        if label == 1:\n            category = &#039;positive&#039;\n            counters[&#039;positive&#039;] += 1\n        elif label == 2:\n            category = &#039;part&#039;\n            counters[&#039;part&#039;] += 1\n        else:\n            category = &#039;negative&#039;\n            counters[&#039;negative&#039;] += 1\n\n        filename = f&quot;{category}\/{counters[category]}.jpg&quot;\n        image.save(os.path.join(base_path, filename))\n\n        try:\n            bbox_str = &#039; &#039;.join(map(str, bbox_offsets))\n            landmark_str = &#039; &#039;.join(map(str, landmark_offsets))\n            files[category].write(f&quot;{filename} {label} {bbox_str} {landmark_str}\\n&quot;)\n        except IOError as e:\n            print(f&quot;Error writing to file: {e}&quot;)\n\ndef generate_samples(face_size, max_samples=-1):\n    &quot;&quot;&quot;\n    \u751f\u6210\u6307\u5b9a\u5927\u5c0f\u7684\u4eba\u8138\u6837\u672c,\u5e76\u4fdd\u5b58\u5230\u6587\u4ef6\u4e2d\u3002\n\n    \u53c2\u6570:\n    face_size (int): \u751f\u6210\u7684\u4eba\u8138\u56fe\u50cf\u5c3a\u5bf8\n    max_samples (int): \u6700\u5927\u751f\u6210\u6837\u672c\u6570\u91cf,\u8bbe\u7f6e\u4e3a -1 \u8868\u793a\u4e0d\u9650\u5236\n    &quot;&quot;&quot;\n    if not os.path.exists(DST_PATH):\n        os.makedirs(DST_PATH)\n\n    paths, base_path = create_directories(DST_PATH, face_size)\n    # \u65b0\u5efa\u6807\u6ce8\u6587\u4ef6\n    files = open_label_files(base_path)\n\n    # \u6837\u672c\u8ba1\u6570\n    counters = {&#039;positive&#039;: 0, &#039;negative&#039;: 0, &#039;part&#039;: 0}\n\n    # \u8bfb\u53d6\u6807\u6ce8\u4fe1\u606f\n    with open(LANMARKS_PATH) as f:\n        landmarks_list = f.readlines()\n    with open(LABEL_PATH) as f:\n        anno_list = f.readlines()\n\n    for i, (anno_line, landmarks) in enumerate(zip(anno_list, landmarks_list)):\n        print(f&quot;positive:{counters[&#039;positive&#039;]}, \\\n                negative:{counters[&#039;negative&#039;]}, \\\n                part:{counters[&#039;part&#039;]}&quot;)\n\n        # \u8df3\u8fc7\u524d\u4e24\u884c\n        if i &lt; 2:\n            continue\n\n        # \u5982\u679c\u5904\u7406\u4e86\u6307\u5b9a\u6570\u91cf\u7684\u6837\u672c,\u5219\u9000\u51fa\u5faa\u73af\n        if max_samples &gt; 0 and i &gt; max_samples:\n            break\n\n        # \u5904\u7406\u5355\u884c\u6807\u6ce8\u4fe1\u606f,\u751f\u6210\u6b63\u8d1f\u6837\u672c\n        samples = process_annotation(\n            face_size, anno_line, landmarks\n        )\n\n        # \u4fdd\u5b58\u6b63\u8d1f\u6837\u672c\u5230\u6587\u4ef6\n        save_samples(\n            samples,\n            files, base_path, counters\n        )\n\n    for file in files.values():\n        file.close()\n\ndef main():\n    # \u751f\u621012\u00d712\u7684\u6837\u672c\n    generate_samples(12, 1000)\n\n    generate_samples(24, 1000)\n\n    generate_samples(48, 1000)\n\nif __name__ == &quot;__main__&quot;:\n    main()\n<\/code><\/pre>\n<p>\u901a\u8fc7\u8fd0\u884c\u4e0a\u8ff0\u7684\u9884\u5904\u7406\u811a\u672c\uff0c\u4ee3\u7801\u4f1a\u5728datasets\u76ee\u5f55\u4e0b\u521b\u5efa\u5bf9\u5e94\u7684\u8bad\u7ec3\u6570\u636e\u96c6<br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/datasets\u8bad\u7ec3\u96c6.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/datasets\u8bad\u7ec3\u96c6.png\" alt=\"\" \/><\/a><\/p>\n<p>\u67e5\u770b48 \u00d7 48\u7684\u8bad\u7ec3\u96c6\uff0c\u53ef\u4ee5\u770b\u5230\u5bf9\u5e94\u7684\u6b63\u3001\u8d1f\u3001\u504f\u6837\u672c\u5982\u4e0b<br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u6b63\u8d1f\u504f\u6837\u672c.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u6b63\u8d1f\u504f\u6837\u672c.png\" alt=\"\" \/><\/a><\/p>\n<blockquote>\n<p>\u9650\u4e8e\u7bc7\u5e45\u539f\u56e0\uff0c\u4ee5\u4e0a\u6570\u636e\u96c6\u9884\u5904\u7406\u90e8\u5206\u7684\u89e3\u6790\u548c\u4ee3\u7801\u7406\u89e3\uff0c\u6211\u5c06\u653e\u5728\u4e0b\u7bc7\u6587\u7ae0\u8fdb\u884c\u3002<\/p>\n<\/blockquote>\n<h4><span class=\"ez-toc-section\" id=\"%E4%B8%89%E4%B8%AA%E6%A8%A1%E5%9E%8B%E5%88%86%E5%88%AB%E8%AE%AD%E7%BB%83\"><\/span>\u4e09\u4e2a\u6a21\u578b\u5206\u522b\u8bad\u7ec3<span class=\"ez-toc-section-end\"><\/span><\/h4>\n<p>\u7531\u4e8eMTCNN\u662f\u4e09\u4e2a\u7f51\u7edc\uff0c\u6240\u4ee5\u9700\u8981\u5206\u522b\u5bf9\u4e09\u4e2a\u7f51\u7edc\u8fdb\u884c\u8bad\u7ec3\u3002<\/p>\n<p>\u7b2c\u4e00\u6b65\uff1a\u6784\u5efa\u8bad\u7ec3\u7684\u516c\u5171\u57fa\u7840\u90e8\u5206\uff0c\u65b9\u4fbf\u4e09\u4e2a\u7f51\u7edc\u8bad\u7ec3\u65f6\u8c03\u7528\u3002<\/p>\n<pre><code class=\"language-python\">import torch\nimport os\nfrom torch.utils.data import DataLoader\nfrom train.FaceDataset import FaceDataset\nimport matplotlib.pyplot as plt\n\nclass Trainer:\n    def __init__(self, net, param_path, data_path):\n        # \u68c0\u6d4b\u662f\u5426\u6709GPU\n        self.device = &#039;cuda:0&#039; if torch.cuda.is_available() else &quot;cpu&quot;\n        # \u628a\u6a21\u578b\u642c\u5230device\n        self.net = net.to(self.device)\n\n        self.param_path = param_path\n\n        # \u6253\u5305\u6570\u636e\n        self.datasets = FaceDataset(data_path)\n\n        # \u5b9a\u4e49\u635f\u5931\u51fd\u6570\uff1a\u7c7b\u522b\u5224\u65ad\uff08\u5206\u7c7b\u4efb\u52a1\uff09\n        self.cls_loss_func = torch.nn.BCELoss()\n\n        # \u5b9a\u4e49\u635f\u5931\u51fd\u6570\uff1a\u6846\u7684\u504f\u7f6e\u56de\u5f52\n        self.offset_loss_func = torch.nn.MSELoss()\n\n        # \u5b9a\u4e49\u635f\u5931\u51fd\u6570\uff1a\u5173\u952e\u70b9\u7684\u504f\u7f6e\u56de\u5f52\n        self.point_loss_func = torch.nn.MSELoss()\n\n        # \u5b9a\u4e49\u4f18\u5316\u5668\n        self.optimizer = torch.optim.Adam(params=self.net.parameters(), lr=1e-3)\n\n    def compute_loss(self, out_cls, out_offset, out_point, cls, offset, point, landmark):\n        # \u9009\u53d6\u7f6e\u4fe1\u5ea6\u4e3a0\uff0c1\u7684\u6b63\u8d1f\u6837\u672c\u6c42\u7f6e\u4fe1\u5ea6\u635f\u5931\n        cls_mask = torch.lt(cls, 2)\n        cls_loss = self.cls_loss_func(torch.masked_select(out_cls, cls_mask), \n                                      torch.masked_select(cls, cls_mask))\n\n        # \u9009\u53d6\u6b63\u6837\u672c\u548c\u90e8\u5206\u6837\u672c\u6c42\u504f\u79fb\u7387\u7684\u635f\u5931\n        offset_mask = torch.gt(cls, 0)\n        offset_loss = self.offset_loss_func(torch.masked_select(out_offset, offset_mask),\n                                            torch.masked_select(offset, offset_mask))\n\n        if landmark:\n            point_loss = self.point_loss_func(torch.masked_select(out_point, offset_mask),\n                                              torch.masked_select(point, offset_mask))\n            return cls_loss, offset_loss, point_loss\n        else:\n            return cls_loss, offset_loss, None\n\n    def train(self, epochs, landmark=False):\n        &quot;&quot;&quot;\n            - \u65ad\u70b9\u7eed\u4f20 --&gt; \u77ed\u70b9\u7eed\u8bad\n            - transfer learning \u8fc1\u79fb\u5b66\u4e60\n            - pretrained model \u9884\u8bad\u7ec3\n\n        :param epochs: \u8bad\u7ec3\u7684\u8f6e\u6570\n        :param landmark: \u662f\u5426\u4e3alandmark\u4efb\u52a1\n        :return:\n        &quot;&quot;&quot;\n\n        # \u52a0\u8f7d\u4e0a\u6b21\u8bad\u7ec3\u7684\u53c2\u6570\n        if os.path.exists(self.param_path):\n            self.net.load_state_dict(torch.load(self.param_path))\n            print(&quot;\u52a0\u8f7d\u53c2\u6570\u6587\u4ef6,\u7ee7\u7eed\u8bad\u7ec3 ...&quot;)\n        else:\n            print(&quot;\u6ca1\u6709\u53c2\u6570\u6587\u4ef6,\u5168\u65b0\u8bad\u7ec3 ...&quot;)\n\n        # \u5c01\u88c5\u6570\u636e\u52a0\u8f7d\u5668\n        dataloader = DataLoader(self.datasets, batch_size=32, shuffle=True)\n\n        # \u5b9a\u4e49\u5217\u8868\u5b58\u50a8\u635f\u5931\u503c\n        cls_losses = []\n        offset_losses = []\n        point_losses = []\n        total_losses = []\n\n        for epoch in range(epochs):\n            # \u8bad\u7ec3\u4e00\u8f6e\n            for i, (img_data, _cls, _offset, _point) in enumerate(dataloader):\n                # \u6570\u636e\u642c\u5bb6 [32, 3, 12, 12]\n                img_data = img_data.to(self.device)\n                _cls = _cls.to(self.device)\n                _offset = _offset.to(self.device)\n                _point = _point.to(self.device)\n\n                if landmark:\n                    # O-Net\u8f93\u51fa\u4e09\u4e2a\n                    out_cls, out_offset, out_point = self.net(img_data)\n                    out_point = out_point.view(-1, 10)\n                else:\n                    # O-Net\u8f93\u51fa\u4e24\u4e2a\n                    out_cls, out_offset = self.net(img_data)\n                    out_point = None\n\n                # [B, C, H, W] \u8f6c\u6362\u4e3a [B, C]\n                out_cls = out_cls.view(-1, 1)\n                out_offset = out_offset.view(-1, 4)\n\n                if landmark:\n                    out_point = out_point.view(-1, 10)\n\n                # \u8ba1\u7b97\u635f\u5931\n                cls_loss, offset_loss, point_loss = self.compute_loss(out_cls, out_offset, out_point,\n                                                                    _cls, _offset, _point, landmark)\n\n                if landmark:\n                    loss = cls_loss + offset_loss + point_loss\n                else:\n                    loss = cls_loss + offset_loss\n\n                # \u6253\u5370\u635f\u5931\n                if landmark:\n                    print(f&quot;Epoch [{epoch+1}\/{epochs}], loss:{loss.item():.4f}, cls_loss:{cls_loss.item():.4f}, &quot;\n                        f&quot;offset_loss:{offset_loss.item():.4f}, point_loss:{point_loss.item():.4f}&quot;)\n                else:\n                    print(f&quot;Epoch [{epoch+1}\/{epochs}], loss:{loss.item():.4f}, cls_loss:{cls_loss.item():.4f}, &quot;\n                        f&quot;offset_loss:{offset_loss.item():.4f}&quot;)\n\n                # \u5b58\u50a8\u635f\u5931\u503c\n                cls_losses.append(cls_loss.item())\n                offset_losses.append(offset_loss.item())\n                if landmark:\n                    point_losses.append(point_loss.item())\n                total_losses.append(loss.item())\n\n                # \u6e05\u7a7a\u68af\u5ea6\n                self.optimizer.zero_grad()\n\n                # \u68af\u5ea6\u56de\u4f20\n                loss.backward()\n\n                # \u4f18\u5316\n                self.optimizer.step()\n\n            # \u4fdd\u5b58\u6a21\u578b\uff08\u53c2\u6570\uff09\n            torch.save(self.net.state_dict(), self.param_path)\n\n        # \u7ed8\u5236\u635f\u5931\u66f2\u7ebf\n        self.plot_losses(cls_losses, offset_losses, point_losses, total_losses, landmark)\n\n        print(&quot;\u8bad\u7ec3\u5b8c\u6210!&quot;)\n\n    def plot_losses(self, cls_losses, offset_losses, point_losses, total_losses, landmark):\n        &quot;&quot;&quot;\n        \u7ed8\u5236\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u7684\u635f\u5931\u66f2\u7ebf\n        :param cls_losses: \u5206\u7c7b\u635f\u5931\u5217\u8868\n        :param offset_losses: \u8fb9\u754c\u6846\u504f\u79fb\u635f\u5931\u5217\u8868\n        :param point_losses: \u5173\u952e\u70b9\u504f\u79fb\u635f\u5931\u5217\u8868\n        :param total_losses: \u603b\u635f\u5931\u5217\u8868\n        :param landmark: \u662f\u5426\u4e3alandmark\u4efb\u52a1\n        &quot;&quot;&quot;\n        plt.figure(figsize=(12, 6))\n        plt.subplot(1, 2, 1)\n        plt.plot(cls_losses, label=&#039;Classification Loss&#039;)\n        plt.plot(offset_losses, label=&#039;Offset Loss&#039;)\n        if landmark:\n            plt.plot(point_losses, label=&#039;Point Loss&#039;)\n        plt.legend()\n        plt.xlabel(&#039;Iteration&#039;)\n        plt.ylabel(&#039;Loss&#039;)\n        plt.title(&#039;Training Losses&#039;)\n\n        plt.subplot(1, 2, 2)\n        plt.plot(total_losses)\n        plt.xlabel(&#039;Iteration&#039;)\n        plt.ylabel(&#039;Total Loss&#039;)\n        plt.title(&#039;Total Training Loss&#039;)\n\n        plt.savefig(&#039;training_losses.png&#039;)\n        plt.close()\n<\/code><\/pre>\n<p>\u7b2c\u4e8c\u6b65\uff1a\u8bad\u7ec3\u5bf9\u5e94\u7684\u7f51\u7edc\u3002<\/p>\n<pre><code class=\"language-python\">from train import model_mtcnn as nets\nimport os\nimport train.train as train\n\nif __name__ == &#039;__main__&#039;:\n\n    current_path = os.path.dirname(os.path.abspath(__file__))\n    # \u6743\u91cd\u5b58\u653e\u5730\u5740\n    base_path = os.path.join(current_path, &quot;model&quot;)\n    model_path = os.path.join(base_path, &quot;p_net.pt&quot;)\n\n    # \u6570\u636e\u5b58\u653e\u5730\u5740\n    data_path = os.path.join(current_path, &quot;datasets\/train\/12&quot;)\n\n    # \u5982\u679c\u6ca1\u6709\u8fd9\u4e2a\u53c2\u6570\u5b58\u653e\u76ee\u5f55\uff0c\u5219\u521b\u5efa\u4e00\u4e2a\u76ee\u5f55\n    if not os.path.exists(base_path):\n        os.makedirs(base_path)\n\n    # \u6784\u5efa\u6a21\u578b\n    pnet = nets.PNet()\n\n    # \u5f00\u59cb\u8bad\u7ec3\n    t = train.Trainer(pnet, model_path, data_path)\n\n    # t.train2(0.01)\n    t.train(100)<\/code><\/pre>\n<p>\u8fd0\u884c\u7ed3\u679c\uff1a<br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u8bad\u7ec3\u8fc7\u7a0b.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u8bad\u7ec3\u8fc7\u7a0b.png\" alt=\"\" \/><\/a><\/p>\n<blockquote>\n<p>\u5907\u6ce8\uff1a\u4ee5\u4e0a\u4ee3\u7801\u5728Apple M3\u82af\u7247\u8fdb\u884c\u8bad\u7ec3\u548c\u63a8\u7406\u65f6\uff0c\u4f1a\u51fa\u73b0(Segmentation Fault)\u7684\u9519\u8bef\uff0c\u56e0\u6b64\u8bad\u7ec3\u548c\u9884\u6d4b\u6700\u597d\u662f\u5728x86\u67b6\u6784\u7684\u7535\u8111\u4e0a\u8fdb\u884c\u3002<\/p>\n<\/blockquote>\n<h3><span class=\"ez-toc-section\" id=\"MTCNN%E6%8E%A8%E7%90%86%E9%80%BB%E8%BE%91\"><\/span>MTCNN\u63a8\u7406\u903b\u8f91<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<ul>\n<li>\n<ol>\n<li>\u8f93\u5165\u4e00\u5f20\u56fe\u7247(\u4e0d\u9650\u5c3a\u5bf8)<\/li>\n<\/ol>\n<\/li>\n<li>\n<ol start=\"2\">\n<li>\u6784\u5efa\u56fe\u50cf\u91d1\u5b57\u5854<\/li>\n<\/ol>\n<\/li>\n<li>\n<ol start=\"3\">\n<li>\u904d\u5386\u91d1\u5b57\u5854\uff0c\u53d6\u51fa\u6bcf\u4e00\u4e2a\u7ea7\u522b\u7684\u56fe\u50cf\n<ul>\n<li>3.1 \u628a\u56fe\u50cf\u8f93\u5165P-Net\uff0c\u5f97\u5230P-Net\u7684\u8f93\u51fa<\/li>\n<li>3.2 \u628aP-Net\u7684\u8f93\u51fa\uff0cresize 24 \u00d7 24\uff0c \u8f93\u5165R-Net\uff0c\u5f97\u5230R-Net\u7684\u8f93\u51fa<\/li>\n<li>3.3 \u628aR-Net\u7684\u8f93\u51fa\uff0cresize 48 \u00d7 48\uff0c \u8f93\u5165O-Net\uff0c\u5f97\u5230O-Net\u7684\u8f93\u51fa<br \/>\n<blockquote>\n<p>\u7bc7\u5e45\u539f\u56e0\uff0c\u4ee3\u7801\u5c06\u5728\u4e0b\u4e00\u7ae0\u8fdb\u884c\u5206\u6790\u7406\u89e3\uff0c\u672c\u7ae0\u4e0d\u518d\u8d58\u8ff0\u3002<\/p>\n<\/blockquote>\n<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n<\/li>\n<\/ul>\n<p>\u63a8\u7406\u6548\u679c\uff1a<br \/>\n<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u63a8\u7406\u6548\u679c.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u63a8\u7406\u6548\u679c.png\" alt=\"\" \/><\/a><\/p>\n<h2><span class=\"ez-toc-section\" id=\"%E5%86%85%E5%AE%B9%E5%B0%8F%E7%BB%93\"><\/span>\u5185\u5bb9\u5c0f\u7ed3<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<ul>\n<li>\u4eba\u8138\u8bc6\u522b\n<ul>\n<li>\u4eba\u8138\u8bc6\u522b\u7ec6\u5206\u6709\u4e24\u79cd\uff1a\u4eba\u8138\u68c0\u6d4b\u548c\u4eba\u8138\u8eab\u4efd\u8bc6\u522b\u3002<\/li>\n<li>\u4eba\u8138\u68c0\u6d4b\u662f\u8bc6\u522b\u56fe\u50cf\u6216\u89c6\u9891\u4e2d\u7684\u4eba\u8138\uff0c\u5e76\u5b9a\u4f4d\u5176\u4f4d\u7f6e\u3002<\/li>\n<li>\u4eba\u8138\u8eab\u4efd\u8bc6\u522b\u662f\u6307\u901a\u8fc7\u8bc6\u522b\u4eba\u8138\u4e0a\u7684\u72ec\u7279\u7279\u5f81\u6765\u786e\u5b9a\u4e00\u4e2a\u4eba\u7684\u8eab\u4efd\u3002<\/li>\n<li>\u76ee\u6807\u68c0\u6d4b\u7b97\u6cd5\u90fd\u53ef\u4ee5\u505a\u4eba\u8138\u68c0\u6d4b\uff0c\u4f46\u662f\u592a\u91cd\u4e86\uff1b\u800cMTCNN\u5c31\u662f\u8fd9\u6837\u4e00\u4e2a\u8f7b\u91cf\u7ea7\u548c\u4e13\u4e1a\u7ea7\u7684\u4eba\u8138\u68c0\u6d4b\u7f51\u7edc\u3002<\/li>\n<\/ul>\n<\/li>\n<li>MTCNN\n<ul>\n<li>MTCNN\u91c7\u7528\u7ea7\u8054\u7ed3\u6784\uff0c\u5305\u542b\u4e09\u4e2a\u9636\u6bb5\u7684\u6df1\u5ea6\u5377\u79ef\u7f51\u7edc\uff0c\u76f8\u5f53\u4e8e<code>\u6d77\u9009\u2192\u6dd8\u6c70\u8d5b\u2192\u51b3\u8d5b<\/code>\u7684\u8fc7\u7a0b\u3002<\/li>\n<li>MTCNN\u7684PNet\u4e3a\u4e86\u80fd\u591f\u5c3d\u53ef\u80fd\u591a\u7684\u8bc6\u522b\u51fa\u4eba\u8138\uff0c\u91c7\u7528\u6ed1\u52a8\u7a97\u53e3\u65b9\u5f0f\uff0c\u751f\u6210\u56fe\u50cf\u91d1\u5b57\u5854\u3002<\/li>\n<li>RNet\u91c7\u7528NMS(\u975e\u6781\u5927\u503c\u6291\u5236)\u6280\u672f\uff0c\u901a\u8fc7\u8ba1\u7b97IOU\uff08\u4ea4\u5e76\u6bd4\uff09\uff0c\u5bf9\u6240\u6709\u5019\u9009\u6846\u6309\u7167\u7f6e\u4fe1\u5ea6\u8fdb\u884c\u6392\u5e8f\uff0c\u4fdd\u7559\u7f6e\u4fe1\u5ea6\u6700\u5927\u7684\u5019\u9009\u6846\u3002<\/li>\n<li>MTCNN\u7684\u8bad\u7ec3\u8fc7\u7a0b\u662f\u4e09\u9636\u6bb5\u5206\u5f00\u8bad\u7ec3\u7684\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h2><span class=\"ez-toc-section\" id=\"%E5%8F%82%E8%80%83%E8%B5%84%E6%96%99\"><\/span>\u53c2\u8003\u8d44\u6599<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p><a href=\"https:\/\/www.jianshu.com\/p\/2227cc959aee\">\u7b80\u4e66\uff1aMTCNN\u4e4b\u4eba\u8138\u68c0\u6d4b\u2014\u2014pytorch\u4ee3\u7801\u5b9e\u73b0<\/a><\/p>\n<p><a href=\"https:\/\/www.jianshu.com\/p\/158d21844723\">\u7b80\u4e66\uff1apytorch\u5b9e\u73b0\u4eba\u8138\u8bc6\u522b\u2014\u2014\uff08MTCNN+\u7279\u5f81\u63d0\u53d6\uff09<\/a><\/p>\n<p><a href=\"https:\/\/blog.csdn.net\/weixin_46470894\/article\/details\/106383539\">CSDN\uff1aMTCNN\u8d85\u8be6\u89e3<\/a><\/p>\n<p><a href=\"https:\/\/blog.csdn.net\/dongjinkun\/article\/details\/136535782\">CSDN\uff1aPyTorch\u57fa\u7840\uff0820\uff09<\/a><\/p>\n<p><a href=\"https:\/\/blog.csdn.net\/KRISNAT\/article\/details\/136086446\">CSDN\uff1a\u8be6\u89e3CelebA\u6570\u636e\u96c6\u4e0b\u8f7d\u3001\u8bfb\u53d6<\/a><\/p>\n<p align=\"center\">\u6b22\u8fce\u5173\u6ce8\u516c\u4f17\u53f7\u4ee5\u83b7\u5f97\u6700\u65b0\u7684\u6587\u7ae0\u548c\u65b0\u95fb<\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/\u626b\u7801_\u641c\u7d22\u8054\u5408\u4f20\u64ad\u6837\u5f0f-\u767d\u8272\u7248.bmp\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/\u626b\u7801_\u641c\u7d22\u8054\u5408\u4f20\u64ad\u6837\u5f0f-\u767d\u8272\u7248.bmp\" alt=\"\" \/><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u524d\u8a00 \u5728\u4e0a\u4e00\u7ae0\u8bfe\u7a0b\u3010\u8bfe\u7a0b\u603b\u7ed3\u3011Day13 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":2436,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"aside","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":"","theme-transparent-header-meta":"default","adv-header-id-meta":"","stick-header-meta":"default","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":"var(--ast-global-color-4)","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":[28],"tags":[65,66],"class_list":["post-2421","post","type-post","status-publish","format-aside","has-post-thumbnail","hentry","category-blog","tag-65","tag-66","post_format-post-format-aside"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.4 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u3010\u8bfe\u7a0b\u603b\u7ed3\u3011Day13\uff08\u4e0b\uff09\uff1a\u4eba\u8138\u8bc6\u522b\u548cMTCNN\u6a21\u578b - \u4e00\u8d77AI\u6280\u672f<\/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:\/\/17aitech.com\/?p=2421\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/17aitech.com\/?p=2421\",\"url\":\"https:\/\/17aitech.com\/?p=2421\",\"name\":\"\u3010\u8bfe\u7a0b\u603b\u7ed3\u3011Day13\uff08\u4e0b\uff09\uff1a\u4eba\u8138\u8bc6\u522b\u548cMTCNN\u6a21\u578b - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\/\/17aitech.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/17aitech.com\/?p=2421#primaryimage\"},\"image\":{\"@id\":\"https:\/\/17aitech.com\/?p=2421#primaryimage\"},\"thumbnailUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u5c01\u9762.jpeg\",\"datePublished\":\"2024-07-01T03:23:26+00:00\",\"dateModified\":\"2024-10-08T07:17:18+00:00\",\"author\":{\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/3d23bb6f7f115fcefc9ae7803a691739\"},\"breadcrumb\":{\"@id\":\"https:\/\/17aitech.com\/?p=2421#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/17aitech.com\/?p=2421\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/?p=2421#primaryimage\",\"url\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u5c01\u9762.jpeg\",\"contentUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u5c01\u9762.jpeg\",\"width\":896,\"height\":418},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/17aitech.com\/?p=2421#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/17aitech.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u3010\u8bfe\u7a0b\u603b\u7ed3\u3011Day13\uff08\u4e0b\uff09\uff1a\u4eba\u8138\u8bc6\u522b\u548cMTCNN\u6a21\u578b\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/17aitech.com\/#website\",\"url\":\"https:\/\/17aitech.com\/\",\"name\":\"\u4e00\u8d77AI\u6280\u672f\",\"description\":\"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca\",\"alternateName\":\"\u4e00\u8d77AI\u6280\u672f\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/17aitech.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/3d23bb6f7f115fcefc9ae7803a691739\",\"name\":\"Dongming\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/image\/\",\"url\":\"\/\/17aitech.com\/wp-content\/uploads\/member\/avatars\/238a0b923820dcc5.1732798681.jpg\",\"contentUrl\":\"\/\/17aitech.com\/wp-content\/uploads\/member\/avatars\/238a0b923820dcc5.1732798681.jpg\",\"caption\":\"Dongming\"},\"description\":\"\u89c1\u5929\u5730\uff0c\u89c1\u4f17\u751f\uff0c\u89c1\u81ea\u5df1\u3002\",\"sameAs\":[\"http:\/\/17aitech.com\"],\"url\":\"https:\/\/17aitech.com\/?page_id=33738&user=1\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u3010\u8bfe\u7a0b\u603b\u7ed3\u3011Day13\uff08\u4e0b\uff09\uff1a\u4eba\u8138\u8bc6\u522b\u548cMTCNN\u6a21\u578b - \u4e00\u8d77AI\u6280\u672f","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:\/\/17aitech.com\/?p=2421","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=2421","url":"https:\/\/17aitech.com\/?p=2421","name":"\u3010\u8bfe\u7a0b\u603b\u7ed3\u3011Day13\uff08\u4e0b\uff09\uff1a\u4eba\u8138\u8bc6\u522b\u548cMTCNN\u6a21\u578b - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=2421#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=2421#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u5c01\u9762.jpeg","datePublished":"2024-07-01T03:23:26+00:00","dateModified":"2024-10-08T07:17:18+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/3d23bb6f7f115fcefc9ae7803a691739"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=2421#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=2421"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=2421#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u5c01\u9762.jpeg","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/07\/\u5c01\u9762.jpeg","width":896,"height":418},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=2421#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u3010\u8bfe\u7a0b\u603b\u7ed3\u3011Day13\uff08\u4e0b\uff09\uff1a\u4eba\u8138\u8bc6\u522b\u548cMTCNN\u6a21\u578b"}]},{"@type":"WebSite","@id":"https:\/\/17aitech.com\/#website","url":"https:\/\/17aitech.com\/","name":"\u4e00\u8d77AI\u6280\u672f","description":"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca","alternateName":"\u4e00\u8d77AI\u6280\u672f","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/17aitech.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/17aitech.com\/#\/schema\/person\/3d23bb6f7f115fcefc9ae7803a691739","name":"Dongming","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/#\/schema\/person\/image\/","url":"\/\/17aitech.com\/wp-content\/uploads\/member\/avatars\/238a0b923820dcc5.1732798681.jpg","contentUrl":"\/\/17aitech.com\/wp-content\/uploads\/member\/avatars\/238a0b923820dcc5.1732798681.jpg","caption":"Dongming"},"description":"\u89c1\u5929\u5730\uff0c\u89c1\u4f17\u751f\uff0c\u89c1\u81ea\u5df1\u3002","sameAs":["http:\/\/17aitech.com"],"url":"https:\/\/17aitech.com\/?page_id=33738&user=1"}]}},"_links":{"self":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/2421","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=2421"}],"version-history":[{"count":8,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/2421\/revisions"}],"predecessor-version":[{"id":33239,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/2421\/revisions\/33239"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/media\/2436"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=2421"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=2421"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=2421"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}