{"id":8452,"date":"2026-05-22T03:27:52","date_gmt":"2026-05-21T19:27:52","guid":{"rendered":"https:\/\/infernews.com\/blog\/semantic-generative-tuning-for-unified-multimodal-models\/"},"modified":"2026-05-22T03:27:52","modified_gmt":"2026-05-21T19:27:52","slug":"semantic-generative-tuning-for-unified-multimodal-models","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/semantic-generative-tuning-for-unified-multimodal-models\/","title":{"rendered":"SGT \u9ede\u6a23\u62c9\u8fd1\u591a\u6a21\u614b\u7406\u89e3\u8207\u751f\u6210"},"content":{"rendered":"<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/05\/und6dim-4c03b43532d3.jpg\" alt=\"Understanding capability gains across proxy task levels\"><\/figure>\n<p>SGT \u662f\u4e00\u500b\u7814\u7a76\u5473\u8f03\u91cd\u7684\u9805\u76ee\uff0c\u6838\u5fc3\u60f3\u6cd5\u662f\u7528\u300c\u5716\u50cf\u5206\u5272\u300d\u4f5c\u70ba\u751f\u6210\u5f0f\u5fae\u8abf\u7684\u4ee3\u7406\u4efb\u52d9\uff0c\u8b93\u540c\u4e00\u500b\u591a\u6a21\u614b\u6a21\u578b\u4e0d\u53ea\u770b\u5f97\u660e\uff0c\u4ea6\u756b\u5f97\u66f4\u6e96\u3002\u5b83\u91dd\u5c0d\u7684\u75db\u9ede\u5f88\u6e05\u695a\uff1a\u4e0d\u5c11\u7d71\u4e00\u591a\u6a21\u614b\u6a21\u578b\u6703\u628a\u7406\u89e3\u8207\u751f\u6210\u5206\u958b\u512a\u5316\uff0c\u7d50\u679c\u5169\u908a\u80fd\u529b\u672a\u5fc5\u771f\u6b63\u5354\u540c\u3002<\/p>\n<p>\u9019\u500b\u9805\u76ee\u7684\u4eae\u9ede\uff0c\u5728\u65bc\u5b83\u6c92\u6709\u518d\u628a\u91cd\u5fc3\u653e\u5728\u50cf\u7d20\u7d0b\u7406\uff0c\u800c\u662f\u6539\u7528\u8f03\u9ad8\u5c64\u6b21\u7684\u8a9e\u610f\u7d50\u69cb\u4f5c\u76e3\u7763\u3002\u7c21\u55ae\u8b1b\uff0c\u6a21\u578b\u4e0d\u662f\u53ea\u5b78\u984f\u8272\u548c\u908a\u7de3\uff0c\u800c\u662f\u5b78\u7269\u4ef6\u5340\u57df\u8207\u7a7a\u9593\u95dc\u4fc2\uff0c\u9019\u5c0d\u5716\u7247\u7406\u89e3\uff0c\u4ee5\u53ca\u6309\u4f4d\u7f6e\u751f\u6210\u5167\u5bb9\uff0c\u90fd\u66f4\u6709\u5e6b\u52a9\u3002<\/p>\n<p>\u5f9e README \u63d0\u4f9b\u7684\u7d50\u679c\u4f86\u770b\uff0cSGT \u66fe\u5728 BAGEL\uff087B+7B\uff09\u548c OmniGen2\uff083B+4B\uff09\u4e0a\u9a57\u8b49\uff0c\u986f\u793a\u5b83\u4e0d\u662f\u7d81\u6b7b\u55ae\u4e00\u67b6\u69cb\u7684\u505a\u6cd5\u3002\u7814\u7a76\u4ea6\u6bd4\u8f03\u4e86 edge\u3001depth\u3001segmentation \u4e09\u985e\u4ee3\u7406\u4efb\u52d9\uff0c\u7d50\u8ad6\u504f\u5411\u4e00\u81f4\uff1asegmentation \u5728\u7406\u89e3\u80fd\u529b\u63d0\u5347\u4e0a\u8868\u73fe\u6700\u597d\uff0c\u800c\u751f\u6210\u7684\u7a7a\u9593\u6e96\u78ba\u5ea6\u4ea6\u6709\u6539\u5584\uff1b\u81f3\u65bc\u6578\u5b78\u6216\u5716\u8868\u63a8\u7406\uff0c\u5c31\u672a\u898b\u660e\u986f\u5e6b\u52a9\u3002<\/p>\n<p>\u60f3\u4e86\u89e3\u9019\u500b\u9805\u76ee\uff0c\u53ef\u5148\u770b\u8ad6\u6587\u8207\u9805\u76ee\u9801\uff0c\u518d\u7559\u610f\u5176\u516c\u958b\u8cc7\u6599\u96c6 SAM-SGT\uff0c\u91cd\u9ede\u4e0d\u662f\u7acb\u5373\u90e8\u7f72\uff0c\u800c\u662f\u7406\u89e3\u9019\u5957\u8a13\u7df4\u601d\u8def\u5982\u4f55\u5957\u5165\u73fe\u6709\u591a\u6a21\u614b\u6a21\u578b\u3002\u5c0d\u7814\u7a76\u4eba\u54e1\u3001\u6a21\u578b\u8a13\u7df4\u5de5\u7a0b\u5e2b\uff0c\u6216\u95dc\u5fc3\u8996\u89ba\u7406\u89e3\u8207\u751f\u6210\u6574\u5408\u7684\u4eba\uff0c\u9019\u500b\u65b9\u5411\u7279\u5225\u503c\u5f97\u7559\u610f\u3002<\/p>\n<ul>\n<li>\u4ee5\u5716\u50cf\u5206\u5272\u4f5c\u751f\u6210\u5f0f\u5fae\u8abf\u4ee3\u7406\uff0c\u9023\u63a5\u7406\u89e3\u8207\u751f\u6210<\/li>\n<li>\u5df2\u5728 BAGEL \u8207 OmniGen2 \u9a57\u8b49\uff0c\u5177\u4e00\u5b9a\u67b6\u69cb\u901a\u7528\u6027<\/li>\n<li>\u6bd4\u8d77 edge\u3001depth\uff0csegmentation \u66f4\u80fd\u63d0\u5347\u8996\u89ba\u7406\u89e3\u8868\u73fe<\/li>\n<li>\u751f\u6210\u7d50\u679c\u7684\u7a7a\u9593\u5c0d\u4f4d\u66f4\u7a69\uff0c\u5c24\u5176\u9069\u5408\u4f4d\u7f6e\u654f\u611f\u63d0\u793a<\/li>\n<li>\u5206\u5272\u8cc7\u6599\u8d8a\u591a\uff0c\u8868\u73fe\u6709\u6301\u7e8c\u4e0a\u5347\u8da8\u52e2<\/li>\n<\/ul>\n<p>\u6574\u9ad4\u4f86\u8aaa\uff0cSGT \u5438\u5f15\u4e4b\u8655\u4e0d\u5728\u82b1\u5de7\u529f\u80fd\uff0c\u800c\u5728\u63d0\u51fa\u4e00\u689d\u9817\u52d9\u5be6\u7684\u8a13\u7df4\u8def\u7dda\uff1a\u7528\u9ad8\u5c64\u8a9e\u610f\u76e3\u7763\uff0c\u88dc\u56de\u591a\u6a21\u614b\u6a21\u578b\u5e38\u898b\u7684\u7406\u89e3\u8207\u751f\u6210\u843d\u5dee\u3002\u82e5\u4f60\u95dc\u6ce8 UMM \u5f8c\u8a13\u7df4\u65b9\u6cd5\uff0c\u9019\u500b\u9805\u76ee\u5f88\u503c\u5f97\u653e\u5165\u89c0\u5bdf\u540d\u55ae\u3002<\/p>\n<p><strong>GitHub\uff1a<\/strong> <a href=\"https:\/\/github.com\/song2yu\/SGT\" rel=\"noopener noreferrer\">https:\/\/github.com\/song2yu\/SGT<\/a><\/p>\n<p><strong>Paper\uff1a<\/strong> <a href=\"https:\/\/arxiv.org\/pdf\/2605.18714\" rel=\"noopener noreferrer\">https:\/\/arxiv.org\/pdf\/2605.18714<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>SGT\u7528\u5716\u50cf\u5206\u5272\u505a\u8a13\u7df4\u6a4b\u6a11\uff0c\u4ee4\u540c\u4e00\u6a21\u578b\u66f4\u8b58\u7747\u5716\u3001\u4ea6\u66f4\u64c5\u9577\u751f\u6210\u3002\u5c0d\u95dc\u6ce8\u591a\u6a21\u614b\u6a21\u578b\u767c\u5c55\u7684\u4eba\u4f86\u8aaa\uff0c\u9019\u662f\u9817\u6709\u610f\u601d\u7684\u4e00\u6b65\u3002<\/p>\n","protected":false},"author":8,"featured_media":8451,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[133,172],"tags":[],"class_list":["post-8452","post","type-post","status-publish","format-standard","hentry","category-133","category-172"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8452","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=8452"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8452\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/8451"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=8452"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=8452"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=8452"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}