
{"id":10124,"date":"2026-07-10T17:23:20","date_gmt":"2026-07-10T09:23:20","guid":{"rendered":"https:\/\/infernews.com\/blog\/eccv-2026-video-oasis-rethinking-evaluation-for-video-understanding\/"},"modified":"2026-07-10T17:24:14","modified_gmt":"2026-07-10T09:24:14","slug":"eccv-2026-video-oasis-rethinking-evaluation-for-video-understanding","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/eccv-2026-video-oasis-rethinking-evaluation-for-video-understanding\/","title":{"rendered":"Video-Oasis \u60f3\u91cd\u505a\u5f71\u7247\u7406\u89e3\u8a55\u6e2c"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/07\/pasted-e2ff1c0ffd61.jpg\" alt=\"video native challenges\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u9ad8\u5206\u672a\u5fc5\u4ee3\u8868\u6a21\u578b\u771f\u4fc2\u7747\u5f97\u61c2\u5f71\u7247\uff0c\u5462\u500b\u9805\u76ee\u6b63\u6b63\u91dd\u5c0d\u5462\u500b\u843d\u5dee\u3002Video-Oasis \u5c6c\u65bc<strong>\u8cc7\u6599\u96c6\u8207\u8a55\u6e2c\u9805\u76ee<\/strong>\uff0c\u91cd\u9ede\u4e0d\u662f\u518d\u52a0\u4e00\u4efd\u984c\u5eab\uff0c\u800c\u662f\u91cd\u65b0\u6aa2\u67e5\u73fe\u6709 video benchmark \u5230\u5e95\u6709\u5e7e\u591a\u984c\u76ee\u771f\u7684\u9700\u8981 visual grounding \u8207 temporal reasoning\uff0c\u907f\u514d\u6a21\u578b\u53ea\u9760\u6587\u5b57\u7dda\u7d22\u3001\u55ae\u5e40\u756b\u9762\u6216\u975c\u614b\u80cc\u666f\u5c31\u7b54\u4e2d\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u666e\u904d\u505a\u6cd5\u662f\u628a\u4e0d\u540c\u5f71\u7247\u554f\u7b54 benchmark \u76f4\u63a5\u5408\u4f75\u6bd4\u8f03\uff0c\u4f5c\u8005\u8a8d\u70ba\u9019\u7a2e\u56fa\u5b9a\u7bc4\u5f0f\u5ffd\u7565\u4e86\u300c\u662f\u5426\u771f\u4fc2\u9700\u8981\u5f71\u7247\u300d\u9019\u500b\u524d\u63d0\u3002Video-Oasis \u5148\u6574\u7406 14 \u500b benchmark\u300124,416 \u500b QA samples\uff0c\u518d\u7528\u5171\u4eab\u7684 visual \u8207 temporal criteria \u5be9\u8996\u984c\u76ee\uff0c\u7d50\u679c\u6307\u51fa\u7d04 55% \u6a23\u672c\u53ef\u88ab non-video shortcuts \u89e3\u958b\uff0c\u4e4b\u5f8c\u518d\u8403\u53d6\u51fa 11,033 \u500b\u8f03\u5177\u4ee3\u8868\u6027\u7684 Video-Native \u6311\u6230\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5b83\u548c\u540c\u985e benchmark \u6700\u5927\u5206\u5225\uff0c\u5728\u65bc\u4e0d\u662f\u8ffd\u6c42\u8986\u84cb\u66f4\u591a\u984c\u578b\uff0c\u800c\u662f\u5148\u6e05\u7406\u8a55\u6e2c\u6c61\u67d3\u3002\u5b98\u65b9\u8cc7\u6599\u63d0\u5230\u4e94\u985e video-native challenges \u624d\u662f\u6838\u5fc3\u96e3\u9ede\uff0c\u800c\u73fe\u6642\u6a21\u578b\u5728\u9019\u90e8\u5206\u8868\u73fe\u4ecd\u7136\u504f\u5f31\uff0c\u6700\u4f73\u6a21\u578b Gemini-2.5 Pro \u53ea\u6709 46.7%\uff0c\u63a5\u8fd1 chance 25.63% \u4e4b\u4e0a\u4e0d\u9060\uff0c\u8aaa\u660e\u9019\u5957\u8a55\u6e2c\u66f4\u80fd\u62c9\u958b\u300c\u7b54\u5f97\u4e2d\u300d\u8207\u300c\u771f\u7406\u89e3\u300d\u4e4b\u9593\u7684\u5dee\u8ddd\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6db5\u84cb 14 \u500b benchmark\uff0c\u4efb\u52d9\u7531 perception \u5ef6\u4f38\u5230 reasoning\uff0c\u7247\u6bb5\u9577\u5ea6\u7531\u5e7e\u79d2\u5230\u6578\u5c0f\u6642<\/li>\n\n\n\n<li>\u4ee5 shared visual and temporal criteria \u91cd\u65b0\u5be9\u6838\u984c\u76ee\uff0c\u4e0d\u662f\u55ae\u7d14\u62fc\u63a5\u820a benchmark<\/li>\n\n\n\n<li>\u7d04 55% QA samples \u53ef\u7528 non-video shortcuts \u89e3\u7b54\uff0c\u771f\u6b63 video-native \u90e8\u5206\u7d04\u4f54 45%<\/li>\n\n\n\n<li>\u8a55\u6e2c\u6d41\u7a0b\u5efa\u57fa\u65bc lmms-eval\uff0c\u4e26\u652f\u63f4\u900f\u904e huggingface_hub \u4e0b\u8f09\u6a21\u578b<\/li>\n\n\n\n<li>README \u5df2\u63d0\u4f9b\u8cc7\u6599\u4e0b\u8f09\u3001\u5f71\u7247\u4fee\u5fa9\u8207\u76ee\u9304\u6574\u7406\u65b9\u5f0f\uff0c\u4f46\u5b8c\u6574\u7a0b\u5f0f\u78bc\u4ecd\u6a19\u793a\u70ba coming soon<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u90e8\u7f72\u7406\u89e3\u4e0a\uff0c\u5b83\u8f03\u50cf\u4e00\u500b\u7814\u7a76\u578b benchmark workflow\uff1a\u4f60\u8981\u5148\u6e96\u5099 Python 3.12\u3001CUDA-compatible GPUs\u3001torch\u3001vllm 0.11.0 \u8207 transformers 4.57.0\uff0c\u518d\u4e0b\u8f09\u5404 benchmark \u5f71\u7247\u3001\u7528 ffmpeg \u8173\u672c\u4fee\u5fa9\u640d\u6bc0\u6a94\u6848\uff0c\u4e4b\u5f8c\u900f\u904e\u5167\u5efa lmms-eval \u8dd1 vqa_total \u6216 v_oasis \u4efb\u52d9\u3002\u73fe\u968e\u6bb5\u8f03\u9069\u5408\u505a\u6a21\u578b\u8a55\u6e2c\u3001\u7814\u7a76\u6bd4\u8f03\uff0c\u6216\u8005\u5e6b\u5718\u968a\u6aa2\u67e5\u81ea\u5bb6 video model \u662f\u5426\u53ea\u662f\u5728 benchmark \u4e0a\u300c\u8b58\u8003\u8a66\u300d\uff0c\u672a\u5fc5\u9069\u5408\u4f5c\u70ba\u5373\u88dd\u5373\u7528\u7684\u61c9\u7528\u5de5\u5177\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9805\u76ee\u9810\u8a2d\u652f\u63f4\u53ef\u7531 huggingface_hub \u4e0b\u8f09\u7684\u6a21\u578b\uff0c\u793a\u4f8b\u63d0\u5230 <a href=\"https:\/\/huggingface.co\/nvidia\/Eagle2.5-8B\" target=\"_blank\" rel=\"noreferrer noopener\">Eagle2.5-8B<\/a>\uff1b\u6210\u7e3e\u8aaa\u660e\u4e2d\u5247\u9ede\u540d Gemini-2.5 Pro \u70ba\u76ee\u524d\u6700\u4f73\u8868\u73fe\u8005\u3002\u6574\u9ad4\u4f86\u770b\uff0cVideo-Oasis \u6700\u6709\u50f9\u503c\u7684\u5730\u65b9\u4e0d\u662f\u518d\u9020\u4e00\u500b\u6392\u884c\u699c\uff0c\u800c\u662f\u628a\u5f71\u7247\u7406\u89e3\u8a55\u6e2c\u88e1\u6700\u5bb9\u6613\u88ab\u5ffd\u7565\u7684\u6377\u5f91\u554f\u984c\u516c\u958b\u5316\uff0c\u4ee4\u5f8c\u7e8c\u6a21\u578b\u6bd4\u8f03\u66f4\u53ef\u4fe1\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/limgeuntaekk.github.io\/Video-Oasis\/\" rel=\"noopener noreferrer\" target=\"_blank\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/sejong-rcv\/Video-Oasis\" rel=\"noopener noreferrer\" target=\"_blank\"><strong>GitHub<\/strong><\/a> \u00b7 <a href=\"https:\/\/arxiv.org\/pdf\/2603.29616\" rel=\"noopener noreferrer\" target=\"_blank\"><strong>Paper<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u4e0d\u5c11\u5f71\u7247 benchmark \u5206\u6578\u770b\u4f3c\u6f02\u4eae\uff0c\u5176\u5be6\u672a\u5fc5\u771f\u4fc2\u8003\u5230\u5f71\u7247\u7406\u89e3\u3002Video-Oasis \u628a\u7126\u9ede\u653e\u56de\u8a55\u6e2c\u672c\u8eab\uff0c\u5c08\u9580\u62c6\u89e3\u6377\u5f91\u984c\u8207\u771f\u6b63\u9700\u8981\u6642\u5e8f\u8b49\u64da\u7684\u984c\u76ee\u3002<\/p>\n","protected":false},"author":8,"featured_media":10123,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,162,199,140,179,152,31,149,141],"tags":[],"class_list":["post-10124","post","type-post","status-publish","format-standard","hentry","category-133","category-ai-productions","category-dataset-","category-gemini","category-nvidia","category-python","category-video","category-149","category-141"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10124","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=10124"}],"version-history":[{"count":2,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10124\/revisions"}],"predecessor-version":[{"id":10126,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10124\/revisions\/10126"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/10123"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=10124"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=10124"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=10124"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}