
{"id":9792,"date":"2026-06-30T04:01:50","date_gmt":"2026-06-29T20:01:50","guid":{"rendered":"https:\/\/infernews.com\/blog\/memobench\/"},"modified":"2026-06-30T04:01:50","modified_gmt":"2026-06-29T20:01:50","slug":"memobench","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/memobench\/","title":{"rendered":"MemoBench \u9ede\u6a23\u6e2c\u4e16\u754c\u6a21\u578b\u8a18\u61b6\u529b"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/06\/pasted-f62820816eef.jpg\" alt=\"Repository image for MemoBench-Team\/MemoBench\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">MemoBench \u662f\u4e00\u500b benchmark\uff0c\u5c6c\u65bc\u7528\u4f86\u8a55\u6e2c world generation models \u7684\u6578\u64da\u96c6\u52a0\u8a55\u6e2c\u5de5\u5177\u7d44\u3002\u5b83\u4e3b\u8981\u6aa2\u67e5\u6a21\u578b\u5728\u93e1\u982d\u79fb\u958b\u518d\u56de\u4f86\u4e4b\u5f8c\uff0c\u80fd\u5426\u628a\u540c\u4e00\u500b\u7269\u4ef6\u7684\u5916\u89c0\u3001\u4f4d\u7f6e\u8207\u72c0\u614b\u6e96\u78ba\u91cd\u5efa\uff0c\u800c\u4e0d\u662f\u53ea\u751f\u6210\u4e00\u6bb5\u770b\u4f3c\u6d41\u66a2\u7684\u5f71\u7247\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u73fe\u6709\u5f71\u7247\u751f\u6210\u6216\u4e16\u754c\u6a21\u578b\u8a55\u6e2c\uff0c\u5f88\u591a\u6642\u96c6\u4e2d\u5728\u756b\u8cea\u3001\u52d5\u4f5c\u6d41\u66a2\u5ea6\uff0c\u6216\u8005\u77ed\u6642\u9593\u4e00\u81f4\u6027\uff1b\u4f5c\u8005\u8a8d\u70ba\u9019\u985e\u7bc4\u5f0f\u672a\u5fc5\u80fd\u6e2c\u5230 <strong>visual memory<\/strong>\u3002MemoBench \u56e0\u6b64\u7528 <strong>V-D-R (Visible \u2192 Disappeared \u2192 Reappear)<\/strong> \u7d50\u69cb\u91cd\u7d44\u4efb\u52d9\uff1a\u5148\u898b\u5230\u76ee\u6a19\u7269\u4ef6\uff0c\u518d\u8b93\u5b83\u96e2\u958b\u8996\u91ce\uff0c\u6700\u5f8c\u8981\u6c42\u6a21\u578b\u5728\u91cd\u8fd4\u8996\u91ce\u6642\u4fdd\u6301\u7269\u4ef6\u6c38\u4e45\u6027\u8207\u5e7e\u4f55\u4e00\u81f4\u6027\uff0c\u9019\u6bd4\u55ae\u770b\u9996\u5c3e\u756b\u9762\u66f4\u63a5\u8fd1\u771f\u6b63\u7684\u8a18\u61b6\u6e2c\u8a66\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u8cc7\u6599\u90e8\u5206\u6709 360 \u6bb5\u7247\uff0c\u5206\u6210 196 \u6bb5 synthetic \u8207 164 \u6bb5 real-world clips\uff0c\u4e26\u63d0\u4f9b V-D-R \u908a\u754c\u3001GT camera poses\u3001\u76ee\u6a19\u7269\u4ef6\u6587\u5b57\u63cf\u8ff0\u3001VQA question banks \u7b49\u914d\u5957\u3002\u90e8\u7f72\u601d\u8def\u76f8\u7576\u6e05\u695a\uff1a\u74b0\u5883\u4ee5 Linux\u3001Python 3.11\u3001CUDA GPU \u70ba\u4e3b\uff0c\u8a55\u6e2c\u6d41\u7a0b\u5206\u6210\u81ea\u52d5\u6307\u6a19\u3001Object Revisit Score (ORS)\uff0c\u4ee5\u53ca VQA \u4e09\u6b65\uff0c\u8f38\u5165\u683c\u5f0f\u662f\u9010\u683c PNG \u5716\u7247\uff1bcamera controllability \u6703\u7528\u5230 MapAnything\uff0cORS \u6703\u7528\u5230 SAM-3\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5b83\u548c\u540c\u985e benchmark \u7684\u5dee\u7570\uff0c\u4e0d\u5728\u65bc\u7247\u6bb5\u6578\u91cf\u7279\u5225\u5927\uff0c\u800c\u5728\u65bc\u628a\u300c\u7269\u4ef6\u6d88\u5931\u5f8c\u518d\u51fa\u73fe\u300d\u8a2d\u6210\u6838\u5fc3\u58d3\u529b\u6e2c\u8a66\uff0c\u4e26\u540c\u6642\u8986\u84cb synthetic \u8207\u771f\u5be6\u5834\u666f\u300214 \u9805\u6307\u6a19\u4ea6\u4e0d\u662f\u53ea\u91cf\u756b\u9762\u597d\u4e0d\u597d\u770b\uff0c\u9084\u6703\u770b temporal consistency\u3001geometric fidelity\u3001object permanence\u3001camera controllability\uff0c\u4ee5\u53ca VQA-based reasoning\uff0c\u53d6\u5411\u660e\u986f\u504f\u5411\u8a3a\u65b7\u6a21\u578b\u7f3a\u53e3\uff0c\u800c\u4e0d\u662f\u53ea\u505a\u55ae\u4e00\u6392\u884c\u699c\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u9069\u5408\u6bd4\u8f03<\/strong> \u4e0d\u540c world generation models \u5728\u9577\u6642\u5e8f\u8a18\u61b6\u4e0a\u7684\u7a69\u5b9a\u5ea6<\/li>\n<li><strong>\u6e2c\u8a66\u6750\u6599\u5b8c\u6574<\/strong>\uff0c\u5305\u542b phase \u908a\u754c\u3001\u76f8\u6a5f\u59ff\u614b\u3001\u6587\u5b57\u63d0\u793a\u8207 VQA \u984c\u5eab<\/li>\n<li><strong>\u8a55\u6e2c\u89d2\u5ea6\u8f03\u7d30<\/strong>\uff0c\u628a\u4f4e\u968e\u756b\u8cea\u8207\u9ad8\u968e\u8a9e\u610f\u4e00\u81f4\u6027\u5206\u958b\u91cf\u5ea6<\/li>\n<li><strong>\u90e8\u7f72\u9580\u6abb\u4e0d\u7b97\u4f4e<\/strong>\uff0c\u9700\u8981 Linux\u3001CUDA GPU\uff0c\u4ea6\u4f9d\u8cf4 MapAnything \u8207 SAM-3<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u76f8\u95dc\u6a21\u578b\u65b9\u9762\uff0c\u5b98\u65b9\u5167\u5bb9\u63d0\u5230\u5df2\u57fa\u6e96\u6e2c\u8a66 8 \u500b\u6a21\u578b\uff0c\u4f46\u9019\u4efd\u5132\u5b58\u5eab\u6458\u8981\u672a\u5217\u51fa\u5b8c\u6574\u540d\u7a31\uff0c\u56e0\u6b64\u8f03\u7a69\u59a5\u7684\u7406\u89e3\u662f\uff1aMemoBench \u91cd\u9ede\u4e0d\u662f\u63d0\u4f9b\u65b0\u6a21\u578b\uff0c\u800c\u662f\u63d0\u4f9b\u4e00\u5957\u53ef\u91cd\u8986\u7684\u8a18\u61b6\u4e00\u81f4\u6027\u6e2c\u8a66\u57fa\u6e96\u3002\u5c0d\u7814\u7a76 world modeling\u3001\u5f71\u7247\u751f\u6210\u3001camera-controlled generation \u7684\u5718\u968a\u5c24\u5176\u6709\u53c3\u8003\u50f9\u503c\uff0c\u56e0\u70ba\u5b83\u80fd\u5e6b\u4f60\u5206\u8fa8\u6a21\u578b\u662f\u300c\u756b\u5f97\u975a\u300d\uff0c\u9084\u662f\u771f\u6b63\u8a18\u5f97\u4e4b\u524d\u898b\u904e\u4ec0\u9ebc\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/memobench-team.github.io\/\" rel=\"noopener noreferrer\" target=\"_blank\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/MemoBench-Team\/MemoBench\" rel=\"noopener noreferrer\" target=\"_blank\"><strong>GitHub<\/strong><\/a> \u00b7 <a href=\"https:\/\/arxiv.org\/pdf\/2606.27537\" rel=\"noopener noreferrer\" target=\"_blank\"><strong>Paper<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>MemoBench\u5c08\u9580\u6aa2\u67e5\u4e16\u754c\u751f\u6210\u6a21\u578b\u6703\u5426\u5728\u7269\u4ef6\u96e2\u93e1\u5f8c\u518d\u51fa\u73fe\u6642\u300c\u8a18\u932f\u6a23\u300d\u3002\u5b83\u7528V-D-R\u6d41\u7a0b\uff0c\u628a\u756b\u8cea\u3001\u4e00\u81f4\u6027\u8207\u8a9e\u610f\u5224\u65b7\u62c6\u958b\u91cf\u5ea6\u3002<\/p>\n","protected":false},"author":8,"featured_media":9791,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[178,179,30,168,152,197],"tags":[],"class_list":["post-9792","post","type-post","status-publish","format-standard","hentry","category-google","category-nvidia","category-image","category-linux","category-python","category-framework"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9792","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=9792"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9792\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/9791"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=9792"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=9792"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=9792"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}