{"id":10763,"date":"2026-08-04T22:25:07","date_gmt":"2026-08-04T14:25:07","guid":{"rendered":"https:\/\/infernews.com\/blog\/eccv-2026-3dzip-spatial-aware-feature-diversity-guided-token-compression-for-3d\/"},"modified":"2026-08-04T23:15:45","modified_gmt":"2026-08-04T15:15:45","slug":"eccv-2026-3dzip-spatial-aware-feature-diversity-guided-token-compression-for-3d","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/eccv-2026-3dzip-spatial-aware-feature-diversity-guided-token-compression-for-3d\/","title":{"rendered":"3DZip \u628a 3D VLM token \u6e1b\u5c11\u5230 10 \u5206 \u4e4b\u4e00"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/logo_v10.png\" alt=\"3DZip logo\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u505a 3D Question Answering \u6642\uff0cprojection-based 3D vision-language models \u5f80\u5f80\u8981\u5148\u628a\u591a\u8996\u89d2 RGB-D \u7279\u5fb5\u6295\u5f71\u5230\u4e16\u754c\u5ea7\u6a19\uff0c\u7d50\u679c\u6bcf\u500b\u5834\u666f\u6703\u5806\u51fa\u5e7e\u5343\u500b token\uff0c\u63a8\u7406\u901f\u5ea6\u540c\u8a18\u61b6\u9ad4\u58d3\u529b\u90fd\u6703\u7acb\u5373\u8b8a\u6210\u74f6\u9838\u30023DZip \u5c6c\u65bc<strong>token compression \u6846\u67b6<\/strong>\uff0c\u8655\u7406\u7684\u6b63\u6b63\u662f\u9019\u500b\u554f\u984c\uff0c\u800c\u4e14\u505a\u6cd5\u4e0d\u662f\u518d\u8a13\u7df4\u4e00\u500b\u65b0\u6a21\u578b\uff0c\u800c\u662f\u76f4\u63a5\u63d2\u9032\u73fe\u6709\u6d41\u7a0b\uff0c\u5148\u6e1b\u91cd\u518d\u56de\u7b54\u554f\u984c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5b83\u7684\u5224\u65b7\u5f88\u6e05\u695a\uff1a3D token \u7684\u5197\u9918\u4e0d\u53ea\u4f86\u81ea\u7a7a\u9593\u4e0a\u592a\u63a5\u8fd1\uff0c\u9084\u5305\u62ec\u7269\u4ef6\u5c64\u7d1a\u5206\u4f48\u4e0d\u5e73\u5747\uff0c\u6240\u4ee5\u55ae\u9760 2D VLM \u5e38\u898b\u7684 attention \u6216\u8a9e\u610f\u76f8\u95dc\u6027\u58d3\u7e2e\uff0c\u672a\u5fc5\u4fdd\u5f97\u4f4f 3D \u5e7e\u4f55\u7d50\u69cb\u30023DZip \u7528\u4e09\u6b65\u8d70\u8655\u7406\uff0c\u5148\u505a voxelization \u6e05\u8d70\u9ede\u7d1a\u91cd\u8907\uff0c\u518d\u7528 Determinantal Point Process\uff08DPP\uff09\u6311\u51fa\u7279\u5fb5\u5920\u591a\u6a23\u7684 anchor tokens\uff0c\u6700\u5f8c\u5728\u7a7a\u9593\u9650\u5236\u4e0b\u5408\u4f75\u5176\u9918 token\uff0c\u91cd\u9ede\u662f\u4fdd\u6301 geometric coherence\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img data-dominant-color=\"e7e7e3\" data-has-transparency=\"false\" style=\"--dominant-color: #e7e7e3;\" loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"442\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/image-1.png\" alt=\"\" class=\"wp-image-10765 not-transparent\" srcset=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/image-1.png 1024w, https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/image-1-300x129.png 300w, https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/image-1-768x332.png 768w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">3DZip \u63d0\u4f9b <a href=\"https:\/\/github.com\/ZCMax\/LLaVA-3D\" target=\"_blank\" rel=\"noreferrer noopener\">LLaVA-3D<\/a> \u7684 inference \u8207 evaluation code\uff0c\u4ee3\u8868\u4f60\u53ef\u4ee5\u628a\u5b83\u7406\u89e3\u6210\u504f\u5411\u7814\u7a76\u9a57\u8b49\u3001\u6548\u80fd\u6bd4\u8f03\u540c\u65e2\u6709\u6a21\u578b\u52a0\u901f\u7684\u9805\u76ee\uff0c\u800c\u4e0d\u662f\u5373\u88dd\u5373\u7528\u7684\u5b8c\u6574\u7522\u54c1\u3002\u6838\u5fc3\u6f14\u7b97\u6cd5\u653e\u5728 <code>llava\/model\/multimodal_encoder\/video_encoder.py<\/code> \u7684 <code>3dzip<\/code> pooling branch\uff0c\u90e8\u7f72\u601d\u8def\u4e5f\u5f88\u76f4\u63a5\uff1a\u4ee5 LLaVA-3D \u70ba\u57fa\u790e\u6a21\u578b\uff0c\u628a\u58d3\u7e2e\u6d41\u7a0b\u63a5\u5230 multimodal encoder\uff0c\u518d\u7528\u57fa\u6e96\u6e2c\u8a66\u770b token \u6578\u3001\u901f\u5ea6\u540c\u56de\u7b54\u8cea\u7d20\u4e4b\u9593\u7684\u53d6\u6368\u3002<\/p>\n\n\n<div\n    \tclass=\"align wp-block-vpb-video\"    id='vpbpVideoPlayer-1'\n    data-attributes='{&quot;source&quot;:&quot;https:\\\/\\\/paper.pnu-cvsp.com\\\/3DZip\\\/static\\\/videos\\\/teaser.mp4&quot;,&quot;repeat&quot;:true,&quot;autoplay&quot;:true,&quot;muted&quot;:true,&quot;align&quot;:&quot;&quot;,&quot;poster&quot;:&quot;&quot;,&quot;controls&quot;:{&quot;play-large&quot;:true,&quot;restart&quot;:false,&quot;rewind&quot;:true,&quot;play&quot;:true,&quot;fast-forward&quot;:true,&quot;progress&quot;:true,&quot;current-time&quot;:true,&quot;duration&quot;:false,&quot;mute&quot;:true,&quot;volume&quot;:true,&quot;pip&quot;:false,&quot;airplay&quot;:false,&quot;captions&quot;:false,&quot;settings&quot;:true,&quot;download&quot;:false,&quot;fullscreen&quot;:true},&quot;captions&quot;:[],&quot;ratio&quot;:&quot;16:9&quot;,&quot;accentColor&quot;:&quot;&quot;,&quot;lazyLoad&quot;:false,&quot;width&quot;:&quot;100%&quot;,&quot;radius&quot;:&quot;0px&quot;,&quot;resetOnEnd&quot;:false,&quot;autoHideControl&quot;:true,&quot;isSetup&quot;:false}'\n><\/div>\n\n\n<p class=\"wp-block-paragraph\">3DZip \u5728\u4e09\u500b 3D question answering benchmarks \u4e0a\uff0c\u58d3\u5230<strong>128 tokens<\/strong>\u4e4b\u5f8c\u4ecd\u4fdd\u7559<strong>94.7%<\/strong>\u539f\u59cb\u8868\u73fe\uff0c\u63a8\u7406\u901f\u5ea6\u63d0\u5347\u5230<strong>1.92\u00d7<\/strong>\u3002\u9019\u985e\u6578\u5b57\u6700\u9069\u5408\u9700\u8981\u5728\u6709\u9650 GPU \u8a18\u61b6\u9ad4\u5167\u8dd1 3D VLM\u3001\u53c8\u4e0d\u60f3\u70ba\u52a0\u901f\u91cd\u65b0\u8a13\u7df4\u6574\u500b\u6a21\u578b\u7684\u7814\u7a76\u5718\u968a\uff1b\u4ee3\u50f9\u662f\u76ee\u524d\u516c\u958b\u91cd\u9ede\u4ecd\u96c6\u4e2d\u5728\u63a8\u7406\u8207\u8a55\u4f30\uff0cHugging Face \u6b0a\u91cd\u8207\u66f4\u5b8c\u6574\u7248\u672c\u4f8b\u5982 3DZip++ \u4ecd\u672a\u91cb\u51fa\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u514d\u8a13\u7df4\u8a2d\u8a08\uff0c\u91cd\u9ede\u5728\u65bc\u76f4\u63a5\u58d3\u7e2e projection-based 3D VLM \u7684 token \u6210\u672c<\/li>\n\n\n\n<li>\u8207 2D token compression \u4e0d\u540c\uff0c3DZip \u540c\u6642\u8655\u7406\u7a7a\u9593\u7d50\u69cb\u540c\u7279\u5fb5\u591a\u6a23\u6027<\/li>\n\n\n\n<li>\u5df2\u516c\u958b LLaVA-3D \u7684 inference \u8207 evaluation code\uff0c\u8f03\u9069\u5408\u7814\u7a76\u8207\u57fa\u6e96\u6bd4\u8f03<\/li>\n\n\n\n<li>128 tokens \u4ecd\u4fdd\u4f4f 94.7% \u539f\u59cb\u8868\u73fe\uff0c\u63a8\u7406\u901f\u5ea6\u53ef\u9054 1.92\u00d7<\/li>\n\n\n\n<li>\u73fe\u968e\u6bb5\u8f03\u50cf\u6f14\u7b97\u6cd5\u6a21\u7d44\uff0c\u672a\u898b\u5b8c\u6574\u7522\u54c1\u5316\u5c01\u88dd<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/paper.pnu-cvsp.com\/3DZip\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/cvsp-lab\/3DZip\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>3D \u554f\u7b54\u6a21\u578b\u5e38\u88ab\u6d77\u91cf token \u62d6\u6162\uff0c3DZip \u7528\u514d\u8a13\u7df4\u58d3\u7e2e\u65b9\u6cd5\u628a\u8ca0\u64d4\u5927\u5e45\u6e1b\u8f15\u3002\u901f\u5ea6\u63d0\u5347\u63a5\u8fd1\u5169\u500d\uff0c\u539f\u6709\u80fd\u529b\u5927\u81f4\u4fdd\u5f97\u4f4f\u3002<\/p>\n","protected":false},"author":8,"featured_media":10762,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,132,119,149,197],"tags":[],"class_list":["post-10763","post","type-post","status-publish","format-standard","hentry","category-133","category-3d","category-119","category-149","category-framework"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10763","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=10763"}],"version-history":[{"count":2,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10763\/revisions"}],"predecessor-version":[{"id":10768,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/10763\/revisions\/10768"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/10762"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=10763"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=10763"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=10763"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}