LoST: Level of Semantics Tokenization for 3D Shapes
Niladri Shekhar Dutt, Zifan Shi, Paul Guerrero, Chun-Hao Paul Huang, [0.5ex] Duygu Ceylan, Niloy J. Mitra, Xuelin Chen, [1.5ex] University College London · 2026
Read LoST: Level of Semantics Tokenization for 3D Shapes by Niladri Shekhar Dutt, Zifan Shi, Paul Guerrero, Chun-Hao Paul Huang, [0.5ex] Duygu Ceylan, Niloy J. Mitra, Xuelin Chen, [1.5ex] University College London (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-17995.txt.
Description
Dutt and co-authors propose LoST, a way to tokenize 3D shapes by level of meaning, not detail. Early tokens decode into whole, plausible shapes with the main traits. Later tokens add fine shape and meaning details for each case. A new loss, RIDA, aligns shape latents with DINO semantic features. LoST beats past detail-based tokenizers on reconstruction and needs far fewer tokens for generation.
Catalog record
| CREATOR | Niladri Shekhar Dutt, Zifan Shi, Paul Guerrero, Chun-Hao Paul Huang, [0.5ex] Duygu Ceylan, Niloy J. Mitra, Xuelin Chen, [1.5ex] University College London |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 55804 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.17995