
{"id":14432,"date":"2026-11-03T15:04:04","date_gmt":"2026-11-03T07:04:04","guid":{"rendered":"https:\/\/infernews.com\/blog\/regap-bench\/"},"modified":"2026-11-03T15:05:01","modified_gmt":"2026-11-03T07:05:01","slug":"regap-bench","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/regap-bench\/","title":{"rendered":"ReGap-Bench\uff1a\u8996\u89ba\u8207\u63a8\u7406\u5206\u958b\u8a55\u6e2c"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/11\/ReGap-Bench.jpg\" alt=\"Teaser\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u6a21\u578b\u660e\u660e\u770b\u898b\u5716\u4e2d\u6240\u6709\u7269\u4ef6\uff0c\u6578\u6578\u6642\u537b\u6703\u91cd\u8907\u8a08\u7b97\uff1bReGap-Bench \u662f\u4e00\u500b\u91dd\u5c0d multimodal large language models\uff08MLLMs\uff09\u7684\u8cc7\u6599\u96c6\u8207\u8a55\u6e2c\u9805\u76ee\uff0c\u5c08\u9580\u6aa2\u67e5\u8a9e\u8a00\u63a8\u7406\u80fd\u5426\u6b63\u78ba\u5c0d\u61c9\u8996\u89ba\u4e2d\u7684\u7269\u4ef6\u3001\u5340\u57df\u548c\u4f4d\u7f6e\uff0c\u4e26\u5c07\u554f\u984c\u547d\u540d\u70ba reference gap\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5b83\u628a\u540c\u4e00\u5834\u666f\u3001\u554f\u984c\u548c\u7b54\u6848\u56fa\u5b9a\uff0c\u53ea\u66f4\u63db\u984d\u5916\u53c3\u7167\uff0c\u5f62\u6210 paired reference-intervention benchmark\u3002\u7de8\u865f\u8b93\u7269\u4ef6\u53ef\u4ee5\u88ab\u7d22\u5f15\uff0c\u9670\u5f71\u6a19\u793a\u6709\u754c\u5340\u57df\uff0c\u7db2\u683c\u7dda\u5247\u5354\u52a9\u5b9a\u4f4d\u3001\u8ffd\u8e64\u908a\u754c\u548c\u6bd4\u8f03\u4f4d\u79fb\uff1b\u53c3\u7167\u662f\u5426\u914d\u5408\u4efb\u52d9\uff0c\u6703\u76f4\u63a5\u5f71\u97ff\u63a8\u7406\u7d50\u679c\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6db5\u84cb shape counting\u3001angle ranking\u3001size ranking\u3001clock ranking \u548c transformation prediction \u4e94\u9805\u4efb\u52d9<\/li>\n<li>\u5171 200 \u500b\u5834\u666f\u53ca 3,400 \u5f35\u5716\u7247\uff0c\u8a2d\u6709 plain\u3001numbering\u3001shading \u548c grid \u689d\u4ef6<\/li>\n<li>task-matched references \u53ef\u4ee4\u6240\u6709\u53d7\u6e2c\u6a21\u578b\u5728\u6240\u6709\u4efb\u52d9\u63d0\u5347\uff0c\u6700\u9ad8\u9054 25%<\/li>\n<li>mismatched references \u5e6b\u52a9\u6709\u9650\uff0c\u751a\u81f3\u6703\u4ee4\u8868\u73fe\u4e0b\u964d<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u6e2c\u8a66\u53ef\u7528 OpenAI\u3001Google Gemini API\uff0c\u4ea6\u53ef\u900f\u904e vLLM \u57f7\u884c Qwen3.5\uff1b\u5f8c\u8005\u9700\u8981 GPU\uff0cAPI \u6a21\u578b\u5247\u53ea\u9700\u8a2d\u5b9a\u76f8\u61c9\u91d1\u9470\u3002Python 3.10 \u74b0\u5883\u548c requirements \u5df2\u5217\u660e\uff0c\u8f38\u51fa\u6703\u4fdd\u5b58\u5230 outputs\uff0c\u65b9\u4fbf\u6bd4\u8f03\u540c\u4e00\u6a21\u578b\u5728\u4e0d\u540c\u53c3\u7167\u689d\u4ef6\u4e0b\u7684\u5dee\u7570\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">ReGap-Bench \u7684\u50f9\u503c\u4e0d\u5728\u65bc\u518d\u52a0\u4e00\u500b\u7e3d\u5206\uff0c\u800c\u662f\u628a\u8996\u89ba\u611f\u77e5\u548c\u8a9e\u8a00\u63a8\u7406\u4e4b\u9593\u7684\u63a5\u5408\u554f\u984c\u55ae\u7368\u62c9\u51fa\u4f86\u3002\u7d50\u679c\u652f\u6301\u4e00\u500b\u52d9\u5be6\u5224\u65b7\uff1a\u6539\u5584\u6a21\u578b\u672a\u5fc5\u53ea\u9760\u66f4\u5f37\u7684\u8996\u89ba\u7de8\u78bc\u5668\uff0c\u70ba\u4efb\u52d9\u5efa\u7acb\u5408\u9069\u7684\u5916\u90e8\u8996\u89ba\u53c3\u7167\uff0c\u4e5f\u53ef\u80fd\u6210\u70ba\u4e0d\u9700\u91cd\u65b0\u8a13\u7df4\u7684 visual skill\uff1b\u4e0d\u904e\u53c3\u7167\u932f\u914d\u6703\u53cd\u904e\u4f86\u5e72\u64fe\u6a21\u578b\uff0c\u56e0\u6b64\u5b83\u66f4\u9069\u5408\u7814\u7a76\u5718\u968a\u3001MLLM \u8a55\u6e2c\u4eba\u54e1\u548c\u9700\u8981\u5206\u6790\u591a\u6b65\u8996\u89ba\u63a8\u7406\u5931\u8aa4\u7684\u958b\u767c\u8005\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/regap-bench.github.io\/\" rel=\"noopener\" target=\"_blank\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/regap-bench\/ReGap-Bench\" rel=\"noopener\" target=\"_blank\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6a21\u578b\u770b\u898b\u7269\u4ef6\uff0c\u4e0d\u4ee3\u8868\u80fd\u9010\u6b65\u8ffd\u8e64\u5b83\u5011\u3002ReGap-Bench \u7528\u53ef\u914d\u5c0d\u7684\u8996\u89ba\u53c3\u7167\uff0c\u6e2c\u51fa\u591a\u6a21\u614b\u63a8\u7406\u7a76\u7adf\u5361\u5728\u54ea\u88cf\u3002<\/p>\n","protected":false},"author":8,"featured_media":14431,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","wpai_meta_description":"","footnotes":""},"categories":[133,171,76,119,127,185,178,153,140,137,30,152,199,201],"tags":[],"class_list":["post-14432","post","type-post","status-publish","format-standard","hentry","category-133","category-171","category-76","category-119","category-127","category-qwen","category-google","category-openai","category-gemini","category-api","category-image","category-python","category-dataset-","category-skill-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/14432","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=14432"}],"version-history":[{"count":1,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/14432\/revisions"}],"predecessor-version":[{"id":14433,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/14432\/revisions\/14433"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/14431"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=14432"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=14432"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=14432"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}