{"id":11230,"date":"2026-08-19T22:14:21","date_gmt":"2026-08-19T14:14:21","guid":{"rendered":"https:\/\/infernews.com\/blog\/v-rae\/"},"modified":"2026-08-19T22:18:39","modified_gmt":"2026-08-19T14:18:39","slug":"v-rae","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/v-rae\/","title":{"rendered":"V-RAE \u91cd\u6574\u5f71\u7247\u6f5b\u7a7a\u9593\uff0c\u751f\u6210\u66f4\u5feb\u66f4\u6e96"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/pasted-181092468a42.jpg\" alt=\"V-RAE method\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u505a\u5f71\u7247\u751f\u6210\u6642\uff0c\u6f5b\u7a7a\u9593\u4e00\u65e6\u53c8\u5927\u53c8\u96dc\uff0c\u8a13\u7df4\u901f\u5ea6\u3001\u91cd\u5efa\u54c1\u8cea\u540c\u5f8c\u7e8c\u751f\u6210\u90fd\u6703\u4e00\u9f4a\u53d7\u62d6\u7d2f\u3002V-RAE\u653e\u55ba\u5462\u500b\u4f4d\u7f6e\u5207\u5165\uff1a\u5b83\u5c6c\u65bc\u5f71\u7247\u8868\u793a\u81ea\u7de8\u78bc\u5668\u6a21\u578b\uff0c\u5c07 frozen vision foundation model \u7684\u8868\u5fb5\u518d\u58d3\u6210\u66f4\u7dca\u6e4a\u7684 generative latents\uff0c\u8655\u7406\u7684\u662f\u5f71\u7247\u8868\u793a\u592a\u5197\u9577\u3001\u4f46\u53c8\u4e0d\u80fd\u5931\u53bb\u8a9e\u610f\u540c\u52d5\u614b\u9023\u7e8c\u6027\u7684\u554f\u984c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">V-RAE\u4e0d\u662f\u91cd\u65b0\u8a13\u7df4\u6574\u500b\u8996\u89ba\u9aa8\u5e79\uff0c\u800c\u662f\u63a5\u5728 DINOv3\u3001SigLIP2\u3001V-JEPA2.1\u3001EUPE \u9019\u985e frozen encoder \u4e4b\u4e0a\uff0c\u7528 lightweight temporal pooling module \u6e1b\u5c11\u6642\u9593\u7dad\u5ea6\u4e0a\u7684\u91cd\u8907\u8cc7\u8a0a\uff0c\u518d\u4ea4\u7531 video decoder \u91cd\u5efa\u9023\u7e8c\u52d5\u4f5c\u3002\u9019\u7a2e\u505a\u6cd5\u7684\u53d6\u6368\u5728\u65bc\uff0c\u5b83\u66f4\u4f9d\u8cf4\u73fe\u6210\u8996\u89ba\u8868\u5fb5\u7684\u54c1\u8cea\uff0c\u4f46\u63db\u4f86\u8f03\u8f15\u91cf\u7684\u5f71\u7247 latent \u58d3\u7e2e\u6d41\u7a0b\uff0c\u4ea6\u4ee4 semantic latents \u53ef\u4ee5\u8b8a\u6210 directly decodable predictive state space\u3002<\/p>\n\n\n<figure class=\"wp-block-embed-youtube wp-block-embed is-type-video is-provider-youtube wp-embed-aspect-16-9 wp-has-aspect-ratio\"><div class=\"lyte-wrapper\" title=\"V-RAE\uff1a\u91cd\u6784\u89c6\u9891\u6f5c\u5728\u7a7a\u95f4\u4ee5\u5b9e\u73b0\u9ad8\u6548\u751f\u6210 2026-08-16\" style=\"width:853px;max-width:100%;margin:5px auto;\"><div class=\"lyMe\" id=\"WYL_cL2DINeFNdo\" itemprop=\"video\" itemscope itemtype=\"https:\/\/schema.org\/VideoObject\"><div><meta itemprop=\"thumbnailUrl\" content=\"https:\/\/infernews.com\/blog\/wp-content\/plugins\/wp-youtube-lyte\/lyteCache.php?origThumbUrl=https%3A%2F%2Fi.ytimg.com%2Fvi%2FcL2DINeFNdo%2Fhqdefault.jpg\" \/><meta itemprop=\"embedURL\" content=\"https:\/\/www.youtube.com\/embed\/cL2DINeFNdo\" \/><meta itemprop=\"duration\" content=\"PT2M38S\" \/><meta itemprop=\"uploadDate\" content=\"2026-08-16T10:37:44Z\" \/><\/div><div id=\"lyte_cL2DINeFNdo\" data-src=\"https:\/\/infernews.com\/blog\/wp-content\/plugins\/wp-youtube-lyte\/lyteCache.php?origThumbUrl=https%3A%2F%2Fi.ytimg.com%2Fvi%2FcL2DINeFNdo%2Fhqdefault.jpg\" class=\"pL\"><div class=\"tC\"><div class=\"tT\" itemprop=\"name\">V-RAE\uff1a\u91cd\u6784\u89c6\u9891\u6f5c\u5728\u7a7a\u95f4\u4ee5\u5b9e\u73b0\u9ad8\u6548\u751f\u6210 2026-08-16<\/div><\/div><button tabindex=\"0\" class=\"play\"><\/button><div class=\"ctrl\"><div class=\"Lctrl\"><\/div><div class=\"Rctrl\"><\/div><\/div><\/div><noscript><a href=\"https:\/\/youtu.be\/cL2DINeFNdo\" rel=\"nofollow\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/plugins\/wp-youtube-lyte\/lyteCache.php?origThumbUrl=https%3A%2F%2Fi.ytimg.com%2Fvi%2FcL2DINeFNdo%2F0.jpg\" alt=\"V-RAE\uff1a\u91cd\u6784\u89c6\u9891\u6f5c\u5728\u7a7a\u95f4\u4ee5\u5b9e\u73b0\u9ad8\u6548\u751f\u6210 2026-08-16\" width=\"853\" height=\"460\" \/><br \/>Watch this video on YouTube<\/a><\/noscript><meta itemprop=\"description\" content=\"V-RAE\uff1a\u91cd\u6784\u89c6\u9891\u6f5c\u5728\u7a7a\u95f4\u4ee5\u5b9e\u73b0\u9ad8\u6548\u751f\u6210 2026-08-16 2608.13556 https:\/\/arxiv.org\/pdf\/2608.13556 \u65f6\u95f4\u6233\uff1a 00:00 \u89c6\u9891\u751f\u6210\u7684\u6f5c\u5728\u7a7a\u95f4\u6311\u6218 00:27 V-RAE\u67b6\u6784\u6982\u89c8 00:53 \u6838\u5fc3\u6280\u672f\uff1a\u65f6\u95f4\u6c60\u5316\u4e0e\u89e3\u7801 01:19 \u65b0\u6307\u6807\uff1atFVD\u7684\u6d1e\u5bdf 01:47 \u5b9e\u9a8c\u7ed3\u679c\u4e0e\u6027\u80fd\u63d0\u5347 02:11 \u603b\u7ed3\u4e0e\u672a\u6765\u5c55\u671b \u89c6\u9891\u751f\u6210\u6a21\u578b\u603b\u662f\u8bad\u7ec3\u592a\u6162\u3001\u8d28\u91cf\u4e0d\u9ad8\uff1f\u8fd9\u7bc7\u8bba\u6587V-RAE\u7ed9\u51fa\u4e86\u7b54\u6848\u3002\u901a\u8fc7\u5229\u7528\u51bb\u7ed3\u7684\u89c6\u89c9\u57fa\u7840\u6a21\u578b\u4f5c\u4e3a\u6f5c\u5728\u7a7a\u95f4\uff0cV-RAE\u4e0d\u4ec5\u5b9e\u73b0\u4e86\u9ad8\u8d28\u91cf\u7684\u89c6\u9891\u91cd\u6784\uff0c\u66f4\u8ba9\u751f\u6210\u6a21\u578b\u7684\u6536\u655b\u901f\u5ea6\u63d0\u5347\u4e866\u500d\u3002\u672c\u89c6\u9891\u5c06\u5e26\u4f60\u6df1\u5165\u7406\u89e3V-RAE\u7684\u6838\u5fc3\u67b6\u6784\u3001\u521b\u65b0\u7684tFVD\u8bc4\u4f30\u6307\u6807\uff0c\u4ee5\u53ca\u5b83\u5982\u4f55\u6539\u53d8\u89c6\u9891\u751f\u6210\u7684\u672a\u6765\u3002\"><\/div><\/div><div class=\"lL\" style=\"max-width:100%;width:853px;margin:5px auto;\"><\/div><figcaption><\/figcaption><\/figure>\n\n\n<p class=\"wp-block-paragraph\">\u9805\u76ee\u63d0\u4f9b\u4e86\u8a13\u7df4\u3001\u8a55\u4f30\u8207\u91cd\u5efa\u793a\u4f8b\u6240\u9700\u7684\u7a0b\u5f0f\u7d50\u69cb\uff0c\u5b89\u88dd\u689d\u4ef6\u5beb\u660e\u8981\u7528 Linux\u3001NVIDIA GPUs\u3001CUDA \u76f8\u5bb9\u9a45\u52d5\u3001FFmpeg\uff0c\u4ee5\u53ca Python 3.10 \u6216\u4ee5\u4e0a\u3002\u53ef\u914d\u5408\u5df2\u91cb\u51fa\u7684 checkpoints \u8207\u5c0d\u61c9 frozen encoder \u505a\u91cd\u5efa\u6e2c\u8a66\uff0c\u4f46\u80fd\u5426\u81ea\u7531\u4e0b\u8f09\u3001\u4e0b\u8f09\u7bc4\u570d\u662f\u5426\u5b8c\u6574\uff0c\u4ecd\u8981\u4ee5\u7576\u524d\u767c\u4f48\u9801\u9762\u70ba\u6e96\uff0c\u4e0d\u9069\u5b9c\u76f4\u63a5\u5047\u8a2d\u4efb\u4f55\u4eba\u90fd\u53ef\u7121\u9650\u5236\u53d6\u5f97\u5168\u90e8\u6a21\u578b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u7d50\u679c V-RAE\u5728 K600 reconstruction \u53d6\u5f97 2.13 rFVD\uff0c\u6578\u503c\u512a\u65bc\u6587\u4e2d\u6bd4\u8f03\u7684\u5927\u578b pretrained video VAEs\uff1bclass-conditional generation \u5247\u5728 UCF101 \u8207 K600 \u5206\u5225\u9054\u5230 117.86 \u8207 19.16 gFVD\uff0c\u4e26\u63d0\u5230\u53ef\u5feb\u6700\u591a 6 \u500d\u6536\u6582\u3002\u4f5c\u8005\u4ea6\u63d0\u51fa tFVD\uff0c\u4ee4\u5b83\u8207\u4eba\u985e\u5224\u65b7\u7684\u4e00\u81f4\u6027\u63d0\u5347\uff0c\u5728 UCF101 \u8207 K600 \u7684 Pearson correlation \u5206\u5225\u9054\u5230 r = 0.621 \u8207 r = 0.919\uff0c\u9019\u9ede\u5c0d\u5f71\u7247\u751f\u6210\u8a55\u6e2c\u6709\u76f4\u63a5\u610f\u7fa9\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u63a5\u5728 frozen vision foundation model \u5f8c\u9762\u505a\u58d3\u7e2e\uff0c\u907f\u514d\u7531\u96f6\u5efa\u7acb\u6574\u5957\u5f71\u7247\u8868\u5fb5<\/li>\n\n\n\n<li>\u7528 temporal pooling \u6e1b\u5c11\u6642\u9593\u5197\u9918\uff0c\u540c\u6642\u4fdd\u4f4f semantic structure<\/li>\n\n\n\n<li>\u540c\u6642\u8986\u84cb reconstruction\u3001class-conditional generation \u8207 predictive modeling \u5834\u666f<\/li>\n\n\n\n<li>\u5009\u5eab\u5df2\u5305\u542b training\u3001evaluation\u3001sampling \u6240\u9700\u7d50\u69cb\uff0c\u4f46\u90e8\u7f72\u524d\u63d0\u504f\u5411\u7814\u7a76\u7d1a Linux + NVIDIA GPU \u74b0\u5883<\/li>\n\n\n\n<li>\u9069\u5408\u7814\u7a76\u5f71\u7247\u751f\u6210\u3001\u4e16\u754c\u72c0\u614b\u5efa\u6a21\u3001\u9577\u5e8f\u5217\u8868\u793a\u5b78\u7fd2\u7684\u5718\u968a\u53c3\u8003\u5176 latent \u8a2d\u8a08<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">V-RAE\u8f03\u9069\u5408\u6709\u5f71\u7247\u6a21\u578b\u5be6\u9a57\u80fd\u529b\u7684\u7814\u7a76\u5718\u968a\u3001\u505a VideoDiT \u985e\u751f\u6210\u6d41\u7a0b\u7684\u4eba\uff0c\u4ee5\u53ca\u60f3\u628a\u5f71\u7247 latent \u62ff\u53bb\u505a\u9810\u6e2c\u72c0\u614b\u7a7a\u9593\u5efa\u6a21\u7684\u9805\u76ee\u3002\u5c0d\u65bc\u600e\u6a23\u628a\u5f37\u5927\u7684\u8996\u89ba\u8868\u5fb5\u8f49\u6210\u66f4\u53ef\u751f\u6210\u3001\u53ef\u91cd\u5efa\u3001\u53ef\u8a55\u6e2c\u7684\u5f71\u7247 latent\uff0c\u5df2\u7d93\u7d66\u51fa\u4e00\u689d\u76f8\u7576\u5177\u9ad4\u7684\u8def\u7dda\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/v-rae.github.io\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/V-RAE\/V-RAE\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a> \u00b7 <a href=\"https:\/\/huggingface.co\/Guomh0707\/V-RAE-Models\" rel=\"noopener noreferrer\"><strong>\u6a21\u578b<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>V-RAE\u628a\u5f71\u7247\u751f\u6210\u524d\u6700\u96e3\u8655\u7406\u7684\u6642\u9593\u5197\u9918\u58d3\u7d30\uff0c\u540c\u6642\u4fdd\u4f4f\u8a9e\u610f\u7d50\u69cb\u3002\u5c0d\u60f3\u505a\u91cd\u5efa\u3001\u751f\u6210\u540c\u9810\u6e2c\u5efa\u6a21\u7684\u5718\u968a\uff0c\u5462\u500b\u65b9\u5411\u5e7e\u6709\u53c3\u8003\u50f9\u503c\u3002<\/p>\n","protected":false},"author":8,"featured_media":11229,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,179,31,168,152,76,127,141],"tags":[],"class_list":["post-11230","post","type-post","status-publish","format-standard","hentry","category-133","category-nvidia","category-video","category-linux","category-python","category-76","category-127","category-141"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11230","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=11230"}],"version-history":[{"count":1,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11230\/revisions"}],"predecessor-version":[{"id":11232,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11230\/revisions\/11232"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/11229"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=11230"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=11230"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=11230"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}