{"id":941,"date":"2023-08-30T15:21:17","date_gmt":"2023-08-30T07:21:17","guid":{"rendered":"http:\/\/www.tamanegi.xyz\/?p=941"},"modified":"2023-08-30T15:21:17","modified_gmt":"2023-08-30T07:21:17","slug":"omni-scale-feature-learning%ef%bc%9apart-2-experiments","status":"publish","type":"post","link":"http:\/\/tamanegi.xyz\/?p=941","title":{"rendered":"Omni-Scale Feature Learning\uff1aPart.2. Experiments"},"content":{"rendered":"

Experiments<\/h1>\n

Datasets&Settings<\/h2>\n

\u5927\u6570\u636e\u96c6<\/h4>\n

Market1501
\nCUHK03\uff1a767\/700 \u6d4b\u8bd5\u65b9\u5f0f\uff0c\u8be6\u89c1https:\/\/blog.csdn.net\/HeavenerWen\/article\/details\/1062482573 \u6d4b\u8bd5\u534f\u8bae 2
\nDukeMTMC-reID (Duke)
\nMSMT17<\/p>\n

\u5c0f\u6570\u636e\u96c6<\/h4>\n

VIPeR GRID
\n\u5c0f\u6570\u636e\u96c6\u9700\u8981\u9884\u5148\u5728\u5927\u6570\u636e\u96c6\u4e0a Pre-training\u4e4b\u540e\u624d\u80fd\u5b66\u4e60\u3002<\/p>\n

Implementation details<\/h2>\n

OSNet \u9876\u90e8\u9884\u52a0\u4e00\u4e2a\u5206\u7c7b\u5c42\uff08linear FC+softmax\uff09\u6bcf\u4e2a\u4eba\u7684\u8eab\u4efd\u88ab\u89c6\u4e3a\u4e00\u4e2a\u72ec\u7279\u7684\u7c7b\u522b\u3002
\nPerson matching \u57fa\u4e8eFC \u5c42\u63d0\u53d6\u7684 512 \u7ef4\u7279\u5f81\u8ba1\u7b97\u7684\u6b27\u6c0f\u8ddd\u79bb ($l_2 distance$)
\nbatch_size:64 weight_decay=5e-4
\nlearning_rate=0.065\uff0c\u5728 150\u3001225\u3001300 \u8f6e\u8870\u51cf 0.1
\nSGD \u8dd1 350epochs<\/p>\n

Big Re-ID datasets Results<\/h2>\n

\u975e\u5e38\u597d<\/p>\n

Small Re-ID datasets Results<\/h2>\n

\u5c0f\u6570\u636e\u96c6\u76f8\u6bd4\u5927\u6570\u636e\u96c6\u800c\u8a00\uff0c\u8bad\u7ec3\u56fe\u50cf\u53ea\u6709\u51e0\u767e\u4e2a\uff0c\u56e0\u6b64\u9700\u8981\u5728\u5927\u6570\u636e\u96c6\u4e0a\u8fdb\u884c\u9884\u8bad\u7ec3\u5e76\u5fae\u8c03\u3002<\/p>\n

Ablation experiments<\/h2>\n

\u50cf\u4e00\u4e2a\u6218\u795e\uff0c\u4ec5\u4ec5\u662f\u8bba\u6587\u4e2d\u5c55\u793a\u7684\u5c31\u8dd1\u4e86 10 \u4e2a\u5bf9\u6bd4\u6a21\u578b\u548c1 \u4e2a\u57fa\u7ebf\u6a21\u578b\u3002
\n[1]\u7ed3\u8bba \uff1a\u6a21\u578b\u51cf\u5c11\u4e86\u4e09\u500d\u4ee5\u4e0a\uff0c\u6027\u80fd\u53ea\u964d\u4f4e\u4e86 0.4%
\n[2]\u7ed3\u8bba \uff1a\u591a\u5c3a\u5ea6\u4e0d\u7ba1\u7528\u4ec0\u4e48\u65b9\u6cd5\u8054\u5408\u90fd\u6bd4\u5355\u5c3a\u5ea6\u6548\u679c\u597d
\n[3]Issue\uff1aChannel-wise Gates \u6bd4Stream-wise Gates\u66f4\u9002\u5408
\n[4]\u7ed3\u8bba\uff1a\u5728\u5904\u7406\u6bcf\u4e2a\u8f93\u5165\u56fe\u50cf\u65f6\uff0c\u9002\u5e94\u6027\u5730\u8fdb\u884c\u5c3a\u5ea6\u878d\u5408\u662f\u81f3\u5173\u91cd\u8981\u7684\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

Experiments Datasets&Settings \u5927\u6570\u636e\u96c6 Market1501 CUHK0 […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[105],"tags":[],"_links":{"self":[{"href":"http:\/\/tamanegi.xyz\/index.php?rest_route=\/wp\/v2\/posts\/941"}],"collection":[{"href":"http:\/\/tamanegi.xyz\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/tamanegi.xyz\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/tamanegi.xyz\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/tamanegi.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=941"}],"version-history":[{"count":1,"href":"http:\/\/tamanegi.xyz\/index.php?rest_route=\/wp\/v2\/posts\/941\/revisions"}],"predecessor-version":[{"id":942,"href":"http:\/\/tamanegi.xyz\/index.php?rest_route=\/wp\/v2\/posts\/941\/revisions\/942"}],"wp:attachment":[{"href":"http:\/\/tamanegi.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=941"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/tamanegi.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=941"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/tamanegi.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=941"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}