{"id":14856,"date":"2023-03-31T14:26:56","date_gmt":"2023-03-31T06:26:56","guid":{"rendered":"http:\/\/139.9.1.231\/?p=14856"},"modified":"2023-03-31T14:26:58","modified_gmt":"2023-03-31T06:26:58","slug":"sotatowards-an-end-to-end-framework-for-flow-guided-video-inpainting","status":"publish","type":"post","link":"http:\/\/139.9.1.231\/index.php\/2023\/03\/31\/sotatowards-an-end-to-end-framework-for-flow-guided-video-inpainting\/","title":{"rendered":"\u89c6\u9891P\u56fe\u65b0SOTA\uff1aTowards An End-to-End Framework for Flow-Guided Video Inpainting"},"content":{"rendered":"\n

Github: https:\/\/github.com\/MCG-NKU\/E2FGVI\/<\/a><\/em><\/strong><\/p>\n\n\n\n

E2FGVI\u7531\u5357\u5f00\u5927\u5b66\u548c\u6d77\u601d<\/strong>\u5408\u4f5c\u5b8c\u6210\u3002\u4e00\u4f5cLi Zhen\u4e3a\u5357\u5f00\u5927\u5b66\u535a\u58eb\u751f\uff0c\u5171\u540c\u4e00\u4f5cLu ChengZe\u4e5f\u6765\u81ea\u5357\u5f00\u3002\u901a\u8baf\u4f5c\u8005\u4e3a\u5357\u5f00\u5927\u5b66\u8ba1\u7b97\u673a\u5b66\u9662\u6559\u6388\u7a0b\u660e\u660e\uff0c\u4e3b\u8981\u7814\u7a76\u65b9\u5411\u662f\u8ba1\u7b97\u673a\u89c6\u89c9\u548c\u56fe\u5f62\u5b66\u3002<\/p>\n\n\n\n

Paper: Towards An End-to-End Framework for Flow-Guided Video Inpainting<\/a><\/p>\n\n\n\n

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

\u5728CVPR 2022\u6536\u5f55\u7684\u8bba\u6587\u4e2d\uff0c\u5c31\u51fa\u73b0\u4e86\u8fd9\u4e48\u4e00\u4e2aP\u56fe\u795e\u5668\uff0c\u5b83\u53ef\u4ee5\u5206\u5206\u949f\u7ed9\u4f60\u4e0a\u6f14\u5404\u79cd\u4eba\u50cf\u6d88\u5931\u5927\u6cd5<\/strong>\uff0c\u4e0d\u7559\u4efb\u4f55\u75d5\u8ff9\u3002<\/p>\n\n\n\n

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

\u53bb\u6c34\u5370\u3001\u586b\u8865\u7f3a\u5931<\/strong>\u66f4\u662f\u4e0d\u5728\u8bdd\u4e0b\uff0c\u5e76\u4e14\u5404\u79cd\u5206\u8fa8\u7387<\/strong>\u7684\u89c6\u9891\u90fd\u80fdhold\u4f4f\u3002<\/p>\n\n\n\n

\u6b63\u5982\u4f60\u6240\u89c1\uff0c\u8fd9\u4e2a\u6a21\u578b\u5982\u6b64\u4e1d\u6ed1\u7684\u8868\u73b0\u8ba9\u5b83\u5728\u4e24\u4e2a\u57fa\u51c6\u6570\u636e\u96c6\u4e0a\u90fd\u5b9e\u73b0\u4e86SOTA\u6027\u80fd<\/strong>\u3002<\/p>\n\n\n\n

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

\u540c\u65f6\u5b83\u7684\u63a8\u7406\u65f6\u95f4\u548c\u8ba1\u7b97\u590d\u6742<\/strong>\u8868\u73b0\u4e5f\u5f88\u62a2\u773c\uff1a\u524d\u8005\u6bd4\u6b64\u524d\u7684\u65b9\u6cd5\u5feb\u4e86\u8fd115\u500d<\/strong>\uff0c\u53ef\u4ee5\u5728Titan XP GPU\u4e0a\u4ee5\u6bcf\u5e270.12\u79d2\u7684\u901f\u5ea6\u5904\u7406432 \u00d7 240\u7684\u89c6\u9891\uff1b\u540e\u8005\u5219\u662f\u5728\u6240\u6709\u6bd4\u8f83\u7684SOTA\u65b9\u6cd5\u4e2d\u5b9e\u73b0\u4e86\u6700\u4f4e\u7684FLOPs\u5206\u6570<\/strong>\u3002<\/p>\n\n\n\n

\u6539\u5584\u5149\u6d41\u6cd5<\/strong><\/p>\n\n\n\n

\u76ee\u524d\u5f88\u591a\u89c6\u9891\u4fee\u590d\u7b97\u6cd5\u5229\u7528\u7684\u90fd\u662f\u5149\u6d41\u6cd5<\/strong>\u00a0\uff08Optical flow\uff09\u3002\u4e5f\u5c31\u662f\u5229\u7528\u56fe\u50cf\u5e8f\u5217\u4e2d\u50cf\u7d20\u5728\u65f6\u95f4\u57df\u4e0a\u7684\u53d8\u5316\u4ee5\u53ca\u76f8\u90bb\u5e27\u4e4b\u95f4\u7684\u76f8\u5173\u6027\uff0c\u627e\u5230\u4e0a\u4e00\u5e27\u8ddf\u5f53\u524d\u5e27\u4e4b\u95f4\u5b58\u5728\u7684\u5bf9\u5e94\u5173\u7cfb\uff0c\u4ece\u800c\u8ba1\u7b97\u51fa\u76f8\u90bb\u5e27\u4e4b\u95f4\u7269\u4f53\u7684\u8fd0\u52a8\u4fe1\u606f\u3002<\/p>\n\n\n\n

\u8fd9\u4e2a\u65b9\u6cd5\u7684\u7f3a\u70b9\u5f88\u660e\u663e\uff1a\u8ba1\u7b97\u91cf\u5927\u3001\u8017\u65f6\u957f<\/strong>\uff0c\u4e5f\u5c31\u662f\u6548\u7387\u4f4e\u3002<\/p>\n\n\n\n

\u4e3a\u6b64\uff0c\u7814\u7a76\u4eba\u5458\u8bbe\u8ba1\u4e86\u4e09\u4e2a\u53ef\u8bad\u7ec3\u6a21\u5757\uff0c\u5206\u522b\u4e3a\u6d41\u5b8c\u6210 \uff08flow completion\uff09\u3001\u7279\u5f81\u4f20\u64ad<\/strong> \uff08feature propagation\uff09\u548c\u5185\u5bb9\u5e7b\u60f3\uff08content hallucination\uff09\uff0c\u63d0\u51fa\u4e86\u4e00\u4e2a\u6d41\u5f15\u5bfc\uff08flow-guided\uff09\u7684\u7aef\u5230\u7aef<\/strong>\u89c6\u9891\u4fee\u590d\u6846\u67b6\uff1aE2FGVI<\/strong>\u3002<\/p>\n\n\n\n

\u8fd9\u4e09\u4e2a\u6a21\u5757\u4e0e\u4e4b\u524d\u57fa\u4e8e\u5149\u6d41\u7684\u65b9\u6cd5\u7684\u4e09\u4e2a\u9636\u6bb5\u76f8\u5bf9\u5e94\uff0c\u4e0d\u8fc7\u53ef\u4ee5\u8fdb\u884c\u8054\u5408\u4f18\u5316\uff0c\u4ece\u800c\u5b9e\u73b0\u66f4\u9ad8\u6548\u7684\u4fee\u590d\u8fc7\u7a0b\u3002<\/p>\n\n\n\n

\u5177\u4f53\u6765\u8bf4\uff0c\u5bf9\u4e8e\u6d41\u5b8c\u6210\u6a21\u5757\uff0c\u8be5\u65b9\u6cd5\u76f4\u63a5\u5728mask viedo\u4e2d\u4e00\u6b65\u5b8c\u6210\u64cd\u4f5c\uff0c\u800c\u4e0d\u662f\u50cf\u6b64\u524d\u65b9\u6cd5\u91c7\u7528\u591a\u4e2a\u590d\u6742\u7684\u6b65\u9aa4\u3002\u5bf9\u4e8e\u7279\u5f81\u4f20\u64ad\u6a21\u5757\uff0c\u4e0e\u6b64\u524d\u7684\u50cf\u7d20\u7ea7\u4f20\u64ad\u76f8\u6bd4\uff0c\u8be5\u65b9\u6cd5\u4e2d\u7684\u6d41\u5f15\u5bfc\u4f20\u64ad\u8fc7\u7a0b\u5728\u7279\u5f81\u7a7a\u95f4\u4e2d\u501f\u52a9\u53ef\u53d8\u5f62\u5377\u79ef\u8fdb\u884c\u3002<\/p>\n\n\n\n

\u901a\u8fc7\u66f4\u591a\u53ef\u5b66\u4e60\u7684\u91c7\u6837\u504f\u79fb\u548c\u7279\u5f81\u7ea7\u64cd\u4f5c\uff0c\u4f20\u64ad\u6a21\u5757\u91ca\u653e\u4e86\u6b64\u524d\u4e0d\u80fd\u51c6\u786e\u8fdb\u884c\u6d41\u4f30\u8ba1\u7684\u538b\u529b\u3002<\/p>\n\n\n\n

\u5bf9\u4e8e\u5185\u5bb9\u5e7b\u60f3\u6a21\u5757\uff0c\u7814\u7a76\u4eba\u5458\u5219\u63d0\u51fa\u4e86\u4e00\u79cd\u65f6\u95f4\u7126\u70b9Transformer\u6765\u6709\u6548\u5730\u5efa\u6a21\u7a7a\u95f4\u548c\u65f6\u95f4\u7ef4\u5ea6\u4e0a\u7684\u957f\u7a0b\u4f9d\u8d56\u5173\u7cfb\u3002\u540c\u65f6\u8be5\u6a21\u5757\u8fd8\u8003\u8651\u4e86\u5c40\u90e8\u548c\u975e\u5c40\u90e8\u65f6\u95f4\u90bb\u57df\uff0c\u4ece\u800c\u83b7\u5f97\u66f4\u5177\u65f6\u95f4\u76f8\u5173\u6027\u7684\u4fee\u590d\u7ed3\u679c\u3002<\/p>\n\n\n\n

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

\u5b9a\u91cf\u5b9e\u9a8c\uff1a<\/strong><\/p>\n\n\n\n

\u7814\u7a76\u4eba\u5458\u5728\u6570\u636e\u96c6YouTube VOS\u548cDAVIS\u4e0a\u8fdb\u884c\u4e86\u5b9a\u91cf\u5b9e\u9a8c\uff0c\u5c06\u4ed6\u4eec\u7684\u65b9\u6cd5\u4e0e\u4e4b\u524d\u7684\u89c6\u9891\u4fee\u590d\u65b9\u6cd5\u8fdb\u884c\u4e86\u6bd4\u8f83\u3002<\/p>\n\n\n\n

\u5982\u4e0b\u8868\u6240\u793a\uff0cE2FGVI\u5728\u5168\u90e8\u56db\u4e2a\u91cf\u5316\u6307\u6807\u4e0a\u90fd\u8fdc\u8fdc\u8d85\u8fc7\u4e86\u8fd9\u4e9bSOTA\u7b97\u6cd5<\/strong>\uff0c\u80fd\u591f\u751f\u6210\u53d8\u5f62\u66f4\u5c11\uff08PSNR\u548cSSIM\uff09\u3001\u89c6\u89c9\u4e0a\u66f4\u5408\u7406\uff08VFID\uff09\u548c\u65f6\u7a7a\u4e00\u81f4\u6027\u66f4\u4f73\uff08Ewarp\uff09\u7684\u4fee\u590d\u89c6\u9891\uff0c\u9a8c\u8bc1\u4e86\u8be5\u65b9\u6cd5\u7684\u4f18\u8d8a\u6027\u3002<\/p>\n\n\n\n

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

\u6b64\u5916\uff0cE2FGVI\u4e5f\u5177\u6709\u6700\u4f4e\u7684FLOPs\u503c\uff08\u8ba1\u7b97\u590d\u6742\u5ea6\uff09\uff0c\u5c3d\u7ba1\u8bad\u7ec3\u662f\u5728432 \u00d7 240\u5206\u8fa8\u7387\u7684\u89c6\u9891\u4e0a\u8fdb\u884c\uff0c\u5b83\u7684HQ\u7248\u672c\u505a\u5230\u4e86\u652f\u6301\u4efb\u610f\u5206\u8fa8\u7387\u3002<\/p>\n\n\n\n

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

\u5b9a\u6027\u5b9e\u9a8c\uff1a<\/strong><\/p>\n\n\n\n

\u7814\u7a76\u4eba\u5458\u9996\u5148\u9009\u62e9\u4e86\u4e09\u79cd\u6700\u6709\u4ee3\u8868\u6027\u7684\u65b9\u6cd5\uff0c\u5305\u62ecCAP\u3001FGVC\uff08\u57fa\u4e8e\u5149\u6d41\u6cd5\uff09\u548cFuseformer\uff08\u5165\u9009ICCV 2021\uff09\uff0c\u8fdb\u884c\u5bf9\u8c61\u79fb\u9664\uff08\u4e0b\u56fe\u524d\u4e09\u884c\uff09\u548c\u7f3a\u5931\u8865\u5168\uff08\u4e0b\u56fe\u540e\u4e24\u884c\uff09\u7684\u6548\u679c\u6bd4\u8f83\u3002<\/p>\n\n\n\n

\u53ef\u4ee5\u53d1\u73b0\uff0c\u524d\u4e09\u79cd\u65b9\u6cd5\u5f88\u96be\u5728\u906e\u6321\u533a\u57df\u6062\u590d\u51fa\u5408\u7406\u7684\u7ec6\u8282\u3001\u64e6\u9664\u4eba\u7269\u4e5f\u4f1a\u9020\u6210\u6a21\u7cca\uff0c\u4f46E2FGVI\u53ef\u4ee5\u751f\u6210\u76f8\u5bf9\u771f\u5b9e\u7684\u7eb9\u7406\u548c\u7ed3\u6784\u4fe1\u606f\u3002<\/p>\n\n\n\n

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

\u6b64\u5916\uff0c\u5b83\u4eec\u8fd8\u9009\u7528\u4e865\u79cd\u65b9\u6cd5\u8fdb\u884c\u4e86\u7528\u6237\u7814\u7a76\uff0c\u7ed3\u679c\u5927\u90e8\u5206\u4eba\u90fd\u5bf9E2FGVI\u4fee\u590d\u540e\u7684\u6548\u679c\u66f4\u6ee1\u610f\u3002<\/p>\n\n\n\n

\u7efc\u4e0a\uff0c\u7814\u7a76\u4eba\u5458\u4e5f\u8868\u793a\uff0c\u5e0c\u671b\u4ed6\u4eec\u63d0\u51fa\u7684\u65b9\u6cd5\u53ef\u4ee5\u6210\u4e3a\u89c6\u9891\u4fee\u590d\u9886\u57df\u65b0\u7684\u5f3a\u5927\u57fa\u7ebf<\/p>\n","protected":false},"excerpt":{"rendered":"

Github: https:\/\/github.com\/MCG-NKU\/E2FGVI\/ E2FGVI\u7531\u5357\u5f00\u5927\u5b66\u548c … \u7ee7\u7eed\u9605\u8bfb\u89c6\u9891P\u56fe\u65b0SOTA\uff1aTowards An End-to-End Framework for Flow-Guided Video Inpainting<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[37,31],"tags":[],"_links":{"self":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/14856"}],"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=14856"}],"version-history":[{"count":4,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/14856\/revisions"}],"predecessor-version":[{"id":14864,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/14856\/revisions\/14864"}],"wp:attachment":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/media?parent=14856"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/categories?post=14856"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/tags?post=14856"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}