{"id":11866,"date":"2026-09-04T01:30:23","date_gmt":"2026-09-03T17:30:23","guid":{"rendered":"https:\/\/infernews.com\/blog\/acm-mm-2026-beyond-visual-similarity-entity-aligned-retrieval-for-knowledge-base\/"},"modified":"2026-09-04T01:30:23","modified_gmt":"2026-09-03T17:30:23","slug":"acm-mm-2026-beyond-visual-similarity-entity-aligned-retrieval-for-knowledge-base","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/acm-mm-2026-beyond-visual-similarity-entity-aligned-retrieval-for-knowledge-base\/","title":{"rendered":"KBMR \u8996\u89ba\u554f\u7b54\u6aa2\u7d22\u5e36\u5411\u5be6\u9ad4\u8a9e\u7fa9"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/09\/pasted-fda1a06e8e1c.jpg\" alt=\"KBMR method overview\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u9762\u5c0d\u4eba\u7269\u3001\u5730\u9ede\u6216\u7269\u4ef6\uff0cKB-VQA \u82e5\u53ea\u6309\u5716\u7247\u5916\u89c0\u641c\u5c0b\uff0c\u5bb9\u6613\u56e0\u8996\u89d2\u3001\u5e74\u4ee3\u548c\u98a8\u683c\u8b8a\u5316\u6435\u932f\u8cc7\u6599\u3002KBMR \u5c6c\u65bc\u9762\u5411 Knowledge-Based Visual Question Answering\uff08KB-VQA\uff09\u7684\u591a\u6a21\u614b\u6a21\u578b\u6aa2\u7d22\u5668\uff0c\u5be6\u969b\u8655\u7406\u7684\u662f\u300c\u756b\u9762\u76f8\u4f3c\u4f46\u5be6\u9ad4\u4e0d\u540c\u300d\u7684\u8b49\u64da\u641c\u5c0b\u554f\u984c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">KBMR \u5148\u7531 EVA-CLIP-8B \u7be9\u9078\u5019\u9078\uff0c\u518d\u4ee5 MLLM-based Semantic Discriminator \u70ba\u67e5\u8a62\u8207\u5019\u9078\u7522\u751f\u9023\u7e8c\u7684 entity-consistency weights\u3002Qwen2-VL-7B \u4ee5\u9019\u4e9b\u8a9e\u7fa9\u8a0a\u865f\u8a13\u7df4 embedding retriever\uff0c\u4e26\u900f\u904e symmetric KL distillation\uff0c\u4ee4 cosine similarity \u5f97\u51fa\u7684 retriever posterior \u8207 discriminator weights \u5f62\u6210\u7684 semantic prior \u5c0d\u9f4a\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u91cd\u9ede\u6574\u7406\u70ba\uff1a<br \/>\n&#8211; \u7531 surface-level visual matching \u6539\u70ba entity-aligned semantic retrieval\u3002<br \/>\n&#8211; \u9023\u7e8c\u6b0a\u91cd\u53ef\u652f\u63f4\u8f03\u53ef\u9760\u7684 hard-negative mining \u548c soft supervision\u3002<br \/>\n&#8211; \u66ff\u63db\u539f\u6709 first-stage retriever\uff0c\u6bcb\u9808\u6539\u52d5\u4e0b\u6e38 reasoning \u6216 reranking\u3002<br \/>\n&#8211; \u4e0d\u540c KB-VQA pipeline \u7684\u7aef\u5230\u7aef\u7b54\u6848\u6e96\u78ba\u7387\u6700\u9ad8\u63d0\u5347 9.4 \u500b\u767e\u5206\u9ede\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u7a2e\u53d6\u6368\u9069\u5408\u5df2\u6709 KB-VQA\u3001\u5916\u90e8\u77e5\u8b58\u5eab\u548c\u5f8c\u7e8c\u63a8\u7406\u6d41\u7a0b\u7684\u7814\u7a76\u5718\u968a\uff0c\u56e0\u70ba KBMR \u4e3b\u6253 plug-and-play \u66ff\u63db\u6aa2\u7d22\u5668\uff0c\u800c\u4e0d\u662f\u91cd\u65b0\u8a2d\u8a08\u6574\u689d\u7cfb\u7d71\u3002KBMR \u5217\u51fa Python 3.10+\u3001EVA-CLIP-8B \u53ca Qwen2-VL-7B \u7b49\u6e96\u5099\u9805\u76ee\uff0c\u4ea6\u5305\u62ec\u8a13\u7df4\u548c\u8a55\u4f30\u7ae0\u7bc0\uff1b\u76ee\u524d\u5df2\u63d0\u4f9b\u5b8c\u6574\u5b89\u88dd\u6307\u4ee4\u3001\u6a21\u578b\u53d6\u5f97\u65b9\u5f0f\u53ca\u7368\u7acb\u63a8\u7406\u6d41\u7a0b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/github.com\/realHarryX\/KBMR\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>KBMR \u91dd\u5c0d KB-VQA \u5bb9\u6613\u6435\u932f\u76f8\u4f3c\u5716\u7247\u7684\u554f\u984c\uff0c\u4ee5 MLLM \u5c07\u6aa2\u7d22\u7126\u9ede\u7531\u5916\u89c0\u5339\u914d\u8f49\u5411\u5be6\u9ad4\u8a9e\u7fa9\u3002<\/p>\n","protected":false},"author":8,"featured_media":11865,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[133,185,131,152,119,149],"tags":[],"class_list":["post-11866","post","type-post","status-publish","format-standard","hentry","category-133","category-qwen","category-embedding","category-python","category-119","category-149"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11866","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=11866"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11866\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/11865"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=11866"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=11866"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=11866"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}