Hugging Face Daily Papers · September 3, 2026 · 3 min read

ZipTok3D: High-Fidelity 3D Tokenization with Compact Token Prefixes

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

ZipTok3D: High-Fidelity 3D Tokenization with Compact Token Prefixes</p>\n","updatedAt":"2026-09-03T07:56:25.451Z","author":{"_id":"66699aa8a33847217b5a49c7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/u8Z-6U8U7ARXOpdBDI7Qm.png","fullname":"Weijie Wang","name":"lhmd","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":15,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.614823043346405},"editors":["lhmd"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/u8Z-6U8U7ARXOpdBDI7Qm.png"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2609.01740","authors":[{"_id":"6a98f4bbfea818274321ff69","user":{"_id":"6954a846a754af29d8964715","avatarUrl":"/avatars/c68c3840490a835d484903c280b832a9.svg","isPro":false,"fullname":"LIN MINGDA","user":"FORTHLOTH","type":"user","name":"FORTHLOTH"},"name":"Mingda Lin","status":"claimed_verified","statusLastChangedAt":"2026-09-03T08:15:21.998Z","hidden":false},{"_id":"6a98f4bbfea818274321ff6a","user":{"_id":"66699aa8a33847217b5a49c7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/u8Z-6U8U7ARXOpdBDI7Qm.png","isPro":false,"fullname":"Weijie Wang","user":"lhmd","type":"user","name":"lhmd"},"name":"Weijie Wang","status":"claimed_verified","statusLastChangedAt":"2026-09-03T08:15:20.061Z","hidden":false},{"_id":"6a98f4bbfea818274321ff6b","name":"Zeyu Zhang","hidden":false},{"_id":"6a98f4bbfea818274321ff6c","name":"Bowen Cui","hidden":false},{"_id":"6a98f4bbfea818274321ff6d","name":"Yefei He","hidden":false},{"_id":"6a98f4bbfea818274321ff6e","name":"Haoyu Zhao","hidden":false},{"_id":"6a98f4bbfea818274321ff6f","name":"Yuanyu He","hidden":false},{"_id":"6a98f4bbfea818274321ff70","user":{"_id":"653862bdbe39573b3b247b44","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/653862bdbe39573b3b247b44/oMqOC4USbQlPX5HFLDXt6.jpeg","isPro":false,"fullname":"Donny Chen","user":"donydchen","type":"user","name":"donydchen"},"name":"Donny Y. Chen","status":"claimed_verified","statusLastChangedAt":"2026-09-03T08:15:17.772Z","hidden":false},{"_id":"6a98f4bbfea818274321ff71","user":{"_id":"64574380f3ef144c0e69d484","avatarUrl":"/avatars/a0a84757cb0bf09c24291803e1389b49.svg","isPro":false,"fullname":"Feng Chen","user":"chenfeng1271","type":"user","name":"chenfeng1271"},"name":"Feng Chen","status":"claimed_verified","statusLastChangedAt":"2026-09-03T08:17:48.487Z","hidden":false},{"_id":"6a98f4bbfea818274321ff72","name":"Bohan Zhuang","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/66699aa8a33847217b5a49c7/SgGJg0_RH5TQF7NpkXPOs.mp4"],"publishedAt":"2026-09-01T00:00:00.000Z","submittedOnDailyAt":"2026-09-03T00:00:00.000Z","title":"ZipTok3D: High-Fidelity 3D Tokenization with Compact Token Prefixes","submittedOnDailyBy":{"_id":"66699aa8a33847217b5a49c7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/u8Z-6U8U7ARXOpdBDI7Qm.png","isPro":false,"fullname":"Weijie Wang","user":"lhmd","type":"user","name":"lhmd"},"summary":"Compact token sequences are essential for efficient 3D generation. However, existing 3D tokenizers typically organize latent representations either over spatial regions or as fixed-size sets of global tokens, both suffering sharp reconstruction degradation when compressed to extremely low token budgets. In this paper, we present ZipTok3D, a 3D tokenizer designed for high-fidelity reconstruction from extremely short token sequences. Its key idea is to organize object geometry into progressively informative global-token prefixes and unfold these compact representations through iterative decoding. Specifically, nested dropout randomly truncates the latent sequence after encoding during training and requires each retained prefix to reconstruct the complete object, thereby prioritizing essential geometric information in the leading tokens. The decoder then repeatedly applies a parameter-shared Transformer block to recover fine-grained geometry from each prefix without a separate generative sampling stage. With the same token dimension, ZipTok3D achieves reconstruction quality comparable to the 32-token COD-VAE baseline using only one token on ShapeNet and four on TRELLIS, yielding 32times and 8times shorter token sequences, respectively.","upvotes":13,"discussionId":"6a98f4bcfea818274321ff73","projectPage":"https://forthloth.github.io/ziptok3d/","ai_summary":"ZipTok3D organizes 3D geometry into compact global-token prefixes with iterative decoding to achieve high-fidelity reconstruction from extremely short token sequences.","ai_keywords":["3D tokenizer","global-token prefixes","nested dropout","iterative decoding","parameter-shared Transformer","COD-VAE"],"ai_summary_model":"thinkingmachines/Inkling-Small","organization":{"_id":"61bac2af530e5c78d7b99667","name":"zju","fullname":"Zhejiang University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/5e1058e9fcf41d740b69966d/7G1xjlxwCdMEmKcxNR0n5.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"66699aa8a33847217b5a49c7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/u8Z-6U8U7ARXOpdBDI7Qm.png","isPro":false,"fullname":"Weijie Wang","user":"lhmd","type":"user"},{"_id":"696a1f07d93688663db6d872","avatarUrl":"/avatars/cab3e585c5ce21364c26ee84b2c65666.svg","isPro":false,"fullname":"yuzai","user":"henryyzhaoo","type":"user"},{"_id":"664c4774d1ba9237d352b0bd","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/664c4774d1ba9237d352b0bd/0lz318nQHS_7rL2x-QDbg.jpeg","isPro":false,"fullname":"yuke zhao","user":"hzeroyuke","type":"user"},{"_id":"653862bdbe39573b3b247b44","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/653862bdbe39573b3b247b44/oMqOC4USbQlPX5HFLDXt6.jpeg","isPro":false,"fullname":"Donny Chen","user":"donydchen","type":"user"},{"_id":"69537fb74aa43450652cd6bd","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/tKeEfCgBApNo9Aj8uDfn1.jpeg","isPro":false,"fullname":"JIAYUE LUO","user":"luoooooooooo","type":"user"},{"_id":"64574380f3ef144c0e69d484","avatarUrl":"/avatars/a0a84757cb0bf09c24291803e1389b49.svg","isPro":false,"fullname":"Feng Chen","user":"chenfeng1271","type":"user"},{"_id":"6954a846a754af29d8964715","avatarUrl":"/avatars/c68c3840490a835d484903c280b832a9.svg","isPro":false,"fullname":"LIN MINGDA","user":"FORTHLOTH","type":"user"},{"_id":"67fb6b6081692bf8e2bd49b1","avatarUrl":"/avatars/d2f28aa4e19c39cb39f5a301014e5739.svg","isPro":false,"fullname":"memory of fish","user":"fish456","type":"user"},{"_id":"68ea42733f01bb4a192c59c4","avatarUrl":"/avatars/cd9928c408e4abc139e97ed77a977548.svg","isPro":false,"fullname":"freestar","user":"FreeStar666","type":"user"},{"_id":"6a9931597d24e1ebe6a44c66","avatarUrl":"/avatars/96c2e31990519984cb3b7880b02591c9.svg","isPro":false,"fullname":"Lin","user":"forthlothhhhhh","type":"user"},{"_id":"69f0230ccb7c5205cf69cd71","avatarUrl":"/avatars/c81d7322d828808e78c70126d7da2848.svg","isPro":false,"fullname":"uppi7","user":"uppi7","type":"user"},{"_id":"6a993307e0676188bdcc27cf","avatarUrl":"/avatars/e1ee01dc3e39b4f490383904fb74c356.svg","isPro":false,"fullname":"lmd","user":"mannnnnnnn","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"61bac2af530e5c78d7b99667","name":"zju","fullname":"Zhejiang University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/5e1058e9fcf41d740b69966d/7G1xjlxwCdMEmKcxNR0n5.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2609/2609.01740.md","query":{}}">

Papers

arxiv:2609.01740

ZipTok3D: High-Fidelity 3D Tokenization with Compact Token Prefixes

Published on Sep 1

· Submitted by

Weijie Wang on Sep 3

Zhejiang University

Upvote

Authors:

Mingda Lin ,

Weijie Wang ,

Donny Y. Chen ,

Feng Chen ,

Abstract

ZipTok3D organizes 3D geometry into compact global-token prefixes with iterative decoding to achieve high-fidelity reconstruction from extremely short token sequences.

Generated by thinkingmachines/Inkling-Small

Compact token sequences are essential for efficient 3D generation. However, existing 3D tokenizers typically organize latent representations either over spatial regions or as fixed-size sets of global tokens, both suffering sharp reconstruction degradation when compressed to extremely low token budgets. In this paper, we present ZipTok3D, a 3D tokenizer designed for high-fidelity reconstruction from extremely short token sequences. Its key idea is to organize object geometry into progressively informative global-token prefixes and unfold these compact representations through iterative decoding. Specifically, nested dropout randomly truncates the latent sequence after encoding during training and requires each retained prefix to reconstruct the complete object, thereby prioritizing essential geometric information in the leading tokens. The decoder then repeatedly applies a parameter-shared Transformer block to recover fine-grained geometry from each prefix without a separate generative sampling stage. With the same token dimension, ZipTok3D achieves reconstruction quality comparable to the 32-token COD-VAE baseline using only one token on ShapeNet and four on TRELLIS, yielding 32times and 8times shorter token sequences, respectively.

View arXiv page View PDF Project page Add to collection