{"id":11252,"date":"2026-08-21T00:33:00","date_gmt":"2026-08-20T16:33:00","guid":{"rendered":"https:\/\/infernews.com\/blog\/semcomp-bench\/"},"modified":"2026-08-21T00:33:00","modified_gmt":"2026-08-20T16:33:00","slug":"semcomp-bench","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/semcomp-bench\/","title":{"rendered":"SemComp-Bench\uff1a\u5f71\u7247\u751f\u6210\u8a55\u6e2c\u7531\u300c\u4f3c\u6a23\u300d\u8d70\u5411\u771f\u6b63\u5b8c\u6210\u4efb\u52d9"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/pasted-9d1634079f36.jpg\" alt=\"Repository image for Kelly372\/SemComp-Bench\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e00\u6bb5\u5f71\u7247\u756b\u9762\u6d41\u66a2\u3001\u7269\u4ef6\u5916\u89c0\u76f8\u8fd1\uff0c\u4e0d\u4ee3\u8868\u5b83\u771f\u7684\u5b8c\u6210\u4e86\u6307\u4ee4\u3002SemComp-Bench\uff08Benchmarking Semantic Task Completion in Video Generation\uff09\u628a\u8a55\u6e2c\u7126\u9ede\u653e\u5728\u300c\u7d50\u679c\u6709\u5426\u505a\u5230\u300d\u548c\u300c\u662f\u5426\u4ecd\u7136\u4fdd\u7559\u8207\u4efb\u52d9\u76f8\u95dc\u7684\u8a9e\u7fa9\u300d\uff1bGitHub \u9805\u76ee\u5247\u662f\u4e00\u5957\u7528\u4f86\u5efa\u7acb SemComp-Data \u7684\u8cc7\u6599\u8655\u7406\u7ba1\u7dda\uff0c\u8655\u7406\u5f71\u7247\u7be9\u9078\u3001\u72c0\u614b\u5b9a\u4f4d\u3001\u6307\u4ee4\u6574\u7406\u548c\u7d50\u679c\u6a19\u8a3b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u7531\u539f\u59cb\u5f71\u7247\u5230\u53ef\u8a55\u6e2c\u8cc7\u6599\uff0c\u6d41\u7a0b\u5206\u6210 9 \u500b\u968e\u6bb5\uff1a\u5148\u6309\u6a19\u984c\u904e\u6ffe\u53ca\u5206\u985e\u4efb\u52d9\uff0c\u518d\u5b9a\u4f4d reference frame \u8207 outcome state\uff0c\u6aa2\u67e5\u756b\u9762\u8cea\u7d20\u548c\u72c0\u614b\u9806\u5e8f\uff0c\u7522\u751f\u4e2d\u82f1\u96d9\u8a9e\u7684\u7c21\u77ed\u53ca\u8a73\u7d30\u6307\u4ee4\uff0c\u6700\u5f8c\u62bd\u53d6 outcome-centric clips\u3001\u6a19\u8a3b semantic alignment types\uff0c\u4e26\u63cf\u8ff0\u7d50\u679c\u72c0\u614b\u3002\u9019\u7a2e\u505a\u6cd5\u628a\u8a55\u6e2c\u6240\u9700\u7684\u53c3\u8003\u756b\u9762\u3001\u6307\u4ee4\u548c\u5b8c\u6210\u7d50\u679c\u653e\u5728\u540c\u4e00\u6bb5\u771f\u5be6\u5f71\u7247\u8108\u7d61\u4e2d\uff0c\u8f03\u9069\u5408\u6aa2\u67e5\u6a21\u578b\u662f\u5426\u771f\u7684\u505a\u5230\u6307\u5b9a\u6539\u8b8a\uff0c\u800c\u4e0d\u53ea\u662f\u7522\u751f\u770b\u4f3c\u5408\u7406\u7684\u756b\u9762\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9805\u76ee\u5c6c\u65bc\u5f71\u7247\u751f\u6210\u8a55\u6e2c\u7684\u8cc7\u6599\u96c6\u5efa\u69cb\u5de5\u5177\uff0c\u5be6\u969b\u89e3\u6c7a\u7684\u662f\u628a\u96f6\u6563\u5f71\u7247\u6574\u7406\u6210\u53ef\u9a57\u8b49\u3001\u53ef\u91cd\u8907\u8a55\u5206\u7684\u4efb\u52d9\u6a23\u672c\u3002SemComp-Bench \u76ee\u524d\u63d0\u4f9b 1,273 \u500b\u7d50\u69cb\u5316\u6a23\u672c\u30016 \u500b\u771f\u5be6\u4e16\u754c\u9818\u57df\u300160 \u500b SemComp-Core cases\uff0c\u5e73\u5747 outcome-centric clip \u7d04 4.03 \u79d2\uff0c\u4e26\u914d\u6709\u5169\u7a2e\u6307\u4ee4\u3001\u56db\u985e reference alignment types\uff0c\u4ee5\u53ca 27 \u500b\u8a55\u6e2c\u53d6\u6a23\u756b\u9762\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4f7f\u7528\u8005\u9700\u8981 Python 3.10 \u6216\u66f4\u65b0\u7248\u672c\u3001ffmpeg\u3001ffprobe\uff0c\u4ee5\u53ca\u4f9b\u7b2c 2 \u81f3\u7b2c 5 \u548c\u7b2c 7 \u81f3\u7b2c 9 \u968e\u6bb5\u4f7f\u7528\u7684 multimodal model service\uff1b\u7b2c 6 \u968e\u6bb5\u7684 ImageBind inference \u5efa\u8b70\u4f7f\u7528\u652f\u63f4 CUDA \u7684\u74b0\u5883\u3002\u539f\u59cb\u5f71\u7247\u3001\u6a21\u578b\u6b0a\u91cd\u3001\u670d\u52d9\u6191\u8b49\u548c\u57f7\u884c\u8f38\u51fa\u5747\u4e0d\u96a8\u5132\u5b58\u5eab\u63d0\u4f9b\uff0c\u56e0\u6b64\u8f03\u9069\u5408\u7814\u7a76\u5718\u968a\u6309\u81ea\u5df1\u7684\u5f71\u7247\u53ca\u6a21\u578b\u670d\u52d9\u91cd\u5efa\u8cc7\u6599\uff0c\u800c\u4e0d\u662f\u4e0b\u8f09\u5f8c\u5373\u6642\u53d6\u5f97\u5b8c\u6574\u6578\u64da\u96c6\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4ee5 outcome achievement \u914d\u5408 semantic grounding\uff0c\u907f\u514d\u53ea\u7528\u756b\u8cea\u6216 prompt alignment \u5224\u65b7\u6210\u529f<\/li>\n<li>\u900f\u904e reference frame\u3001outcome state \u548c\u77ed\u7247\u5efa\u7acb\u5b8c\u6574\u8a55\u6e2c\u4e09\u5143\u7d44<\/li>\n<li>1 \u81f3 7 \u968e\u6bb5\u901a\u5e38\u6703\u8f38\u51fa snapshot\u3001excluded set\uff0c\u6280\u8853\u5931\u6557\u53e6\u6709 error.parquet<\/li>\n<li>\u53ef\u7528 tests\/ \u7684\u96e2\u7dda\u56de\u6b78\u6e2c\u8a66\u6aa2\u67e5\u8655\u7406\u6d41\u7a0b<\/li>\n<li>splitting\/ \u542b\u6539\u7de8\u81ea Panda-70M \u548c ImageBind \u7684\u5143\u4ef6\uff0c\u975e\u5546\u696d\u6388\u6b0a\u9650\u5236\u9700\u8981\u5148\u5be9\u95b1<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u5c0d\u7814\u7a76\u751f\u6210\u5f71\u7247\u3001\u88fd\u4f5c\u8a55\u6e2c\u6578\u64da\uff0c\u6216\u9700\u8981\u5206\u6790\u4efb\u52d9\u5b8c\u6210\u7387\u7684\u5718\u968a\u800c\u8a00\uff0c\u9019\u5957\u7ba1\u7dda\u63d0\u4f9b\u4e86\u6e05\u695a\u7684\u91cd\u5efa\u5165\u53e3\uff1b\u4f46\u5b83\u4f9d\u8cf4\u5916\u90e8\u591a\u6a21\u614b\u6a21\u578b\u670d\u52d9\u548c\u672c\u5730\u5a92\u9ad4\u8cc7\u6e90\uff0c\u8cc7\u6599\u5efa\u7acb\u6210\u672c\u8207\u6388\u6b0a\u5be9\u67e5\u4ecd\u662f\u63a1\u7528\u524d\u5fc5\u9808\u8a08\u7b97\u7684\u90e8\u5206\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/semcomp-bench.github.io\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/Kelly372\/SemComp-Bench\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a> \u00b7 <a href=\"https:\/\/huggingface.co\/datasets\/Kelly372\/SemComp-Bench\" rel=\"noopener noreferrer\"><strong>\u6578\u64da\u96c6<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>SemComp-Bench \u4e0d\u53ea\u770b\u751f\u6210\u5f71\u7247\u662f\u5426\u903c\u771f\uff0c\u9084\u6aa2\u67e5\u6307\u5b9a\u7d50\u679c\u6709\u5426\u5b8c\u6210\uff0c\u4ee5\u53ca\u95dc\u9375\u8a9e\u7fa9\u662f\u5426\u4ecd\u7136\u4fdd\u7559\u3002<\/p>\n","protected":false},"author":8,"featured_media":11251,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[133,179,31,152,119,149,199],"tags":[],"class_list":["post-11252","post","type-post","status-publish","format-standard","hentry","category-133","category-nvidia","category-video","category-python","category-119","category-149","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11252","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=11252"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11252\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/11251"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=11252"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=11252"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=11252"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}