OctLLM unifies 3D and language with sparse octree tokens
OctLLM encodes 3D geometry explicitly with sparse octree tokens, avoiding shape compression losses and backbone fine-tuning, and sets new SOTA results while preserving general language ability.
2026-10-05 ~ 2026-10-05 · 2 related posts
- OctLLM Uses Sparse Octree Tokens to Hit 3D SOTA While Preserving Language Ability — Ran Dan · 2026-10-05
- OctLLM encodes 3D geometry as explicit octree token sequences without sacrificing language ability — _akhaliq · 2026-10-05