GHOST: Geometry-Hierarchical Online Streaming Token Eviction for Efficient 3D Reconstruction
Leyang Chen, Junyi Wu, Zhiteng Li, Yulun Zhang · 2026
Read GHOST: Geometry-Hierarchical Online Streaming Token Eviction for Efficient 3D Reconstruction by Leyang Chen, Junyi Wu, Zhiteng Li, Yulun Zhang (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-2605-15852.txt.
Description
Leyang Chen and three colleagues present GHOST, a training-free way to manage the key-value cache in streaming 3D reconstruction from long monocular video. Older methods either cut the cache to fixed anchor frames, hurting quality, or use attention scores blind to 3D structure. GHOST ranks token value in tiers, shields special tokens from eviction, and sets layer budgets by cosine similarity, guided by the model's own geometry outputs. Tests show it keeps reconstruction quality while halving the cache and speeding inference 1.75 times over state-of-the-art methods.
Catalog record
| CREATOR | Leyang Chen, Junyi Wu, Zhiteng Li, Yulun Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 50225 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.15852