
{"id":12683,"date":"2026-09-28T18:41:40","date_gmt":"2026-09-28T10:41:40","guid":{"rendered":"https:\/\/infernews.com\/blog\/fomo-forking-moment-in-generative-trajectory-as-a-perceptual-distance\/"},"modified":"2026-09-28T18:41:40","modified_gmt":"2026-09-28T10:41:40","slug":"fomo-forking-moment-in-generative-trajectory-as-a-perceptual-distance","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/fomo-forking-moment-in-generative-trajectory-as-a-perceptual-distance\/","title":{"rendered":"FoMo \u4e0d\u7528\u4eba\u624b\u6a19\u8a3b\uff0c\u8b93\u5f71\u50cf\u66f4\u8cbc\u8fd1\u611f\u77e5"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/09\/FoMo_generated_pairs.jpg\" alt=\"FoMo: data generation from the forking moment of the diffusion trajectory, and ranked-BCE training\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u6bd4\u8f03\u539f\u5716\u8207\u4fee\u6539\u7248\u672c\u6642\uff0c\u50cf\u7d20\u5dee\u7570\u672a\u5fc5\u7b49\u65bc\u4eba\u773c\u611f\u53d7\u5230\u7684\u5dee\u7570\u3002FoMo \u4ee5 reference-based perceptual distance metrics\uff08\u53c3\u8003\u5f0f\u611f\u77e5\u8ddd\u96e2\u6a21\u578b\uff09\u8655\u7406\u9019\u500b\u554f\u984c\uff0c\u900f\u904e FLUX.1-dev \u7684 img2img \u8ecc\u8de1\u88fd\u9020\u6709\u6b21\u5e8f\u7684\u5f71\u50cf\u8b8a\u9ad4\uff0c\u4e26\u4ee5\u6b64\u8a13\u7df4\u5f71\u50cf\u54c1\u8cea\u8a55\u4f30\uff08image quality assessment\uff0cIQA\uff09\u6a21\u578b\uff0c\u5c6c\u65bc\u958b\u6e90\u6a21\u578b\u9805\u76ee\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5b83\u7684\u95dc\u9375\u505a\u6cd5\uff0c\u662f\u5728 diffusion model \u7684\u53bb\u566a\u6d41\u7a0b\u4e2d\uff0c\u65bc\u4e0d\u540c denoising steps \u91cd\u65b0\u63a5\u56de\u751f\u6210\u8ecc\u8de1\u3002\u91cd\u65b0\u63a5\u5165\u5f97\u8d8a\u65e9\uff0c\u751f\u6210\u5716\u8207\u539f\u5716\u7684\u611f\u77e5\u504f\u96e2\u901a\u5e38\u8d8a\u5927\uff1b\u9019\u500b forking moment \u4fbf\u63d0\u4f9b\u4e86\u5bc6\u96c6\u800c\u6709\u6392\u5e8f\u7684\u76e3\u7763\u8a0a\u865f\u3002\u9805\u76ee\u7528 480k \u7d44\u5f71\u50cf\u8cc7\u6599\u914d\u5408 pairwise ranking objective\uff0c\u6e1b\u5c11\u4f9d\u8cf4\u4eba\u624b\u9010\u5f35\u6a19\u8a3b\u7684\u6210\u672c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9805\u76ee\u63d0\u4f9b\u4e03\u500b\u6a21\u5f0f\uff0c\u6db5\u84cb LPIPS\u3001DISTS\u3001DreamSim\uff0c\u4ee5\u53ca\u4ee5 frozen DINOv3\u3001CLIP\u3001MAE backbone \u914d\u5408 prediction head \u7684\u7248\u672c\u3002DreamSim \u900f\u904e LoRA adapters \u5fae\u8abf\uff0c\u800c dinov3\u3001clip\u3001mae \u6703\u628a\u5169\u5f35\u5716\u7247\u6309\u6b21\u5e8f\u8b80\u5165\u540c\u4e00\u7d44 token\uff1b\u9810\u8a2d\u6703\u8a08\u7b97\u5169\u500b\u65b9\u5411\u7684\u5e73\u5747\u5206\uff0c\u8655\u7406\u8f38\u51fa\u53ef\u80fd\u4e0d\u5c0d\u7a31\u7684\u554f\u984c\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>lower score \u4ee3\u8868\u5169\u5f35\u5716\u7247\u66f4\u76f8\u4f3c\uff0craw \u8f38\u51fa\u4e5f\u662f\u516c\u958b\u8a55\u4f30\u63a1\u7528\u7684\u6578\u503c<\/li>\n<li>output=&#8217;loss&#8217; \u53ef\u63d0\u4f9b\u4ee5 identity pair \u70ba\u9328\u9ede\u7684 perceptual loss<\/li>\n<li>CNN \u985e\u6a21\u578b\u4f7f\u7528\u56fa\u5b9a 256px crop\uff0cDreamSim \u4f7f\u7528\u56fa\u5b9a 224px crop<\/li>\n<li>DINOv3\u3001CLIP\u3001MAE \u652f\u63f4 64 \u81f3 256px \u96a8\u6a5f\u88c1\u526a\uff0c\u4e26\u4ee5 zero-padding \u548c attention masking \u7d44\u6279<\/li>\n<li>\u9805\u76ee\u53ef\u900f\u904e Python \u5957\u4ef6 fomo-iqa \u8f09\u5165\u9810\u8a13\u7df4\u6a21\u578b\uff0c\u76f4\u63a5\u6bd4\u8f03\u53c3\u8003\u5716\u8207\u6e2c\u8a66\u5716<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">FoMo \u9069\u5408\u9700\u8981\u81ea\u52d5\u5224\u65b7\u751f\u6210\u5716\u7247\u3001\u7de8\u8f2f\u7248\u672c\u6216\u91cd\u5efa\u7d50\u679c\u76f8\u4f3c\u7a0b\u5ea6\u7684\u7814\u7a76\u8005\u8207\u5f71\u50cf\u5de5\u5177\u5718\u968a\u3002\u5b83\u7684\u9650\u5236\u4e5f\u5f88\u6e05\u695a\uff1a\u8a13\u7df4\u8a0a\u865f\u4f86\u81ea FLUX.1-dev \u7684\u751f\u6210\u8ecc\u8de1\uff0c\u4e0d\u80fd\u76f4\u63a5\u7b49\u540c\u4eba\u985e\u504f\u597d\uff1b\u4e0d\u540c backbone \u7684\u8f38\u51fa\u5f62\u5f0f\u3001\u88c1\u526a\u7b56\u7565\u8207\u5c0d\u7a31\u8655\u7406\u4ea6\u6703\u5f71\u97ff\u6bd4\u8f03\u7d50\u679c\uff0c\u56e0\u6b64\u61c9\u6309\u5de5\u4f5c\u6d41\u7a0b\u9078\u64c7\u6a21\u578b\u4e26\u81ea\u884c\u9a57\u8b49\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/huggingface.co\/papers\/2609.25716\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/JHLew\/FoMo\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>FoMo \u501f\u52a9 diffusion model \u751f\u6210\u7531\u6dfa\u5165\u6df1\u7684\u5f71\u50cf\u8b8a\u9ad4\uff0c\u514d\u9664\u5927\u91cf\u4eba\u5de5\u6a19\u8a3b\uff0c\u8a13\u7df4\u51fa\u53ef\u6bd4\u8f03\u5169\u5f35\u5716\u7247\u611f\u77e5\u5dee\u7570\u7684\u6a21\u578b\u3002<\/p>\n","protected":false},"author":8,"featured_media":12682,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[133,127,30,120,38,212],"tags":[],"class_list":["post-12683","post","type-post","status-publish","format-standard","hentry","category-133","category-127","category-image","category-120","category-38","category-lora"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/12683","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=12683"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/12683\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/12682"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=12683"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=12683"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=12683"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}