{"id":9351,"date":"2026-06-18T06:58:52","date_gmt":"2026-06-17T22:58:52","guid":{"rendered":"https:\/\/infernews.com\/blog\/code-for-quot-self-evolving-visual-questioner-seeq-quot\/"},"modified":"2026-06-18T06:59:30","modified_gmt":"2026-06-17T22:59:30","slug":"code-for-quot-self-evolving-visual-questioner-seeq-quot","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/code-for-quot-self-evolving-visual-questioner-seeq-quot\/","title":{"rendered":"SeeQ \u8b93 VLM \u5b78\u8b58\u81ea\u5df1\u51fa\u8996\u89ba\u554f\u984c"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/06\/coverimage2.jpg\" alt=\"Cover Figure overview\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u73fe\u6709 Vision-Language Models\uff08VLMs\uff09\u591a\u6578\u6309\u300c\u88ab\u52d5\u7b54\u984c\u300d\u7bc4\u5f0f\u8a13\u7df4\uff1a\u4eba\u985e\u6216\u5916\u90e8\u6a21\u578b\u5148\u63d0\u4f9b\u554f\u984c\uff0c\u6a21\u578b\u518d\u5b78\u7fd2\u56de\u7b54\u3002\u8ad6\u6587\u8a8d\u70ba\u9019\u7a2e fixed inputs \u505a\u6cd5\u53d7\u5236\u65bc\u975c\u614b\u8cc7\u6599\u5206\u4f48\uff0cVisual Question Generation\uff08VQG\uff09\u4ea6\u5bb9\u6613\u5361\u5728\u6a19\u8a3b\u6210\u672c\u9ad8\u3001\u984c\u76ee\u6df1\u5ea6\u4e0d\u8db3\u9019\u5169\u500b\u74f6\u9838\uff0c\u6240\u4ee5 SeeQ \u63d0\u51fa Self-Evolving Visual Questioner\uff0c\u7528\u540c\u4e00\u500b VLM \u540c\u6642\u505a proposer \u8207 filter\uff0c\u81ea\u52d5\u5f9e\u672a\u6a19\u8a3b\u5716\u7247\u751f\u7522\u66f4\u96e3\u3001\u66f4\u8cbc\u8fd1\u756b\u9762\u5167\u5bb9\u7684\u554f\u984c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u500b\u9805\u76ee\u5c6c\u65bc<strong>\u6846\u67b6\u517c\u7814\u7a76\u578b\u5de5\u5177<\/strong>\uff0c\u91cd\u9ede\u4e0d\u662f\u518d\u505a\u4e00\u500b\u666e\u901a\u984c\u5eab\uff0c\u800c\u662f\u5efa\u7acb\u5b8c\u6574\u6d41\u6c34\u7dda\uff1a\u5148\u751f\u6210 seed questions\uff0c\u518d\u53cd\u8986\u6539\u5beb\uff0c\u63d0\u5347 visual search\u3001context \u8207 spatial reasoning \u8981\u6c42\uff0c\u4e4b\u5f8c\u518d\u7531\u6a21\u578b\u81ea\u884c\u904e\u6ffe\u3002\u4f5c\u8005\u540c\u6642\u52a0\u5165 exploration diversity \u63a7\u5236\uff0c\u76ee\u6a19\u662f\u907f\u514d\u8a13\u7df4\u4e00\u8def\u6536\u7a84\uff0c\u6700\u5f8c\u53ea\u5269\u55ae\u4e00\u98a8\u683c\u984c\u76ee\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5982\u679c\u4f60\u60f3\u8a66\uff0c\u8f03\u5408\u7406\u7684\u505a\u6cd5\u662f\u5148\u6e96\u5099\u5716\u7247\u5c0d\u61c9\u7684 JSON \u8f38\u5165\uff0c\u518d\u5206\u958b\u770b generation \u8207 evaluation \u5169\u90e8\u5206\u8f38\u51fa\u3002\u5009\u5eab\u5167\u6c92\u6709\u9644\u6a21\u578b\u6b0a\u91cd\u3001\u6578\u64da\u96c6\u8207\u5feb\u53d6\uff0c\u8a55\u6e2c\u4ea6\u6703\u7528\u5230 image-capable OpenAI evaluator \u8207 Qwen embedding models\uff0c\u6240\u4ee5\u8f03\u9069\u5408\u5df2\u7d93\u6709 VLM \u74b0\u5883\u3001\u60f3\u9a57\u8b49\u81ea\u52d5\u51fa\u984c\u6d41\u7a0b\u7684\u7814\u7a76\u8005\u6216\u591a\u6a21\u614b\u5718\u968a\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4ee5\u672a\u6a19\u8a3b\u5716\u7247\u958b\u59cb\uff0c\u81ea\u52d5\u751f\u6210\u3001\u6539\u5beb\u3001\u904e\u6ffe\u8996\u89ba\u554f\u984c<\/li>\n\n\n\n<li>\u4fdd\u7559 Agentic evaluation\uff0c\u5f9e visual search\u3001evidence coverage\u3001context\u3001spatial reasoning \u8a55\u5206<\/li>\n\n\n\n<li>\u53e6\u7528 Qwen embedding models \u6aa2\u67e5\u6574\u9ad4\u591a\u6a23\u6027\uff0c\u4e0d\u53ea\u770b\u55ae\u984c\u8cea\u7d20<\/li>\n\n\n\n<li>\u5f37\u8abf zero external supervision\uff0c\u4e0d\u4f9d\u8cf4\u4eba\u5de5\u6a19\u8a3b\u6216 GPT-4V \u9019\u985e\u5916\u90e8 teacher models<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u5275\u65b0\u9ede\u5728\u65bc\u5b83\u4e0d\u55ae\u6b62\u7528 VLM \u7522\u751f\u554f\u984c\uff0c\u9084\u628a\u300c\u63d0\u554f\u80fd\u529b\u300d\u7576\u6210\u53ef\u81ea\u6211\u589e\u5f37\u7684\u8a13\u7df4\u8a0a\u865f\uff0c\u4e26\u4e14\u628a questioner \u8207 answerer \u5169\u7a2e\u6a21\u5f0f\u4e00\u8d77\u8003\u616e\u3002\u6309\u8ad6\u6587\u8aaa\u6cd5\uff0c\u9019\u5957\u65b9\u6cd5\u5728\u591a\u500b backbone VLMs \u4e0a\u90fd\u80fd\u63d0\u5347\u554f\u984c\u8cea\u7d20\uff0c\u4ea6\u628a\u81ea\u52d5\u51fa\u984c\u7684\u96e3\u5ea6\u908a\u754c\u63a8\u9ad8\uff1b\u540c\u6a23\u9810\u7b97\u4e0b\uff0c\u6bd4\u76f4\u63a5\u7528\u975c\u614b\u4f86\u6e90\u8cc7\u6599\u8a13\u7df4\u66f4\u6709\u6548\uff0c\u800c\u6a21\u578b\u7684 answerer \u80fd\u529b\u4ea6\u672a\u6709\u660e\u986f\u72a7\u7272\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u76f8\u95dc\u6a21\u578b\u8207\u5143\u4ef6\u65b9\u9762\uff0c\u5009\u5eab\u5167\u5bb9\u986f\u793a\u751f\u6210\u6d41\u7a0b\u53ef\u914d\u5408 Qwen2.5 3B \u985e\u578b\u8a2d\u5b9a\uff0c\u8a55\u6e2c\u6703\u7528 OpenAI \u7684\u53ef\u770b\u5716\u8a55\u4f30\u5668\uff0c\u4ee5\u53ca Qwen embedding models\u3002\u82e5\u4f60\u95dc\u5fc3\u591a\u6a21\u614b\u8a13\u7df4\u3001\u5408\u6210\u6578\u64da\u3001\u6216\u60f3\u5efa\u7acb\u80fd\u81ea\u5df1\u767c\u554f\u518d\u81ea\u6211\u6539\u826f\u7684 Agentic workflow\uff0cSeeQ \u7684\u65b9\u6cd5\u8ad6\u6bd4\u55ae\u7d14\u770b\u5206\u6578\u66f4\u6709\u53c3\u8003\u50f9\u503c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>GitHub\uff1a<\/strong> <a href=\"https:\/\/github.com\/tianyi-lab\/SeeQ\" rel=\"noopener noreferrer\">https:\/\/github.com\/tianyi-lab\/SeeQ<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Paper\uff1a<\/strong> <a href=\"https:\/\/arxiv.org\/pdf\/2606.13929\" rel=\"noopener noreferrer\">https:\/\/arxiv.org\/pdf\/2606.13929<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u9019\u500b\u9805\u76ee\u805a\u7126 Visual Question Generation\uff0c\u60f3\u89e3\u6c7a\u9ad8\u8cea\u8996\u89ba\u63d0\u554f\u6578\u64da\u6602\u8cb4\u53c8\u96e3\u64f4\u5145\u7684\u554f\u984c\u3002\u5b83\u7528\u81ea\u6211\u6f14\u5316\u6d41\u7a0b\uff0c\u4ee4 VLM \u7531\u7b54\u984c\u8005\u8b8a\u6210\u6703\u4e3b\u52d5\u767c\u554f\u7684\u7cfb\u7d71\u3002<\/p>\n","protected":false},"author":8,"featured_media":9350,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[171,185,153,116,30,38,162,131,158,152,109,119,150,76,127,149,199,197],"tags":[],"class_list":["post-9351","post","type-post","status-publish","format-standard","hentry","category-171","category-qwen","category-openai","category-agentic","category-image","category-38","category-ai-productions","category-embedding","category-ide","category-python","category-rag","category-119","category-150","category-76","category-127","category-149","category-dataset-","category-framework"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9351","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=9351"}],"version-history":[{"count":1,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9351\/revisions"}],"predecessor-version":[{"id":9353,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9351\/revisions\/9353"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/9350"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=9351"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=9351"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=9351"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}