SpatialBrain stores spatial-reasoning experience with a fixed backbone
Original title: SpatialBrain: Spatial Intelligence Growing from Experience
On five benchmarks it beats direct answers from the same backbone on four and ties on the fifth.
SpatialBrain keeps its vision-language backbone fixed and retains experience about when to observe, reconstruct, measure or change reference frames. 3DFusionMemory stores queryable spatial relations, a skill library holds tool procedures, and Spatial Cognition links task conditions to when to use them. Completed trajectories become process records between task blocks without transferring scene coordinates or reference answers.
- Scene evidence is kept separate from experience about how to reason.
- Later decisions draw on earlier execution without reference answers.
- Component comparisons and coverage curves describe how resources are used.
From the original article
1 min read at huggingface.co
Spatial reasoning requires more than access to geometric tools: a reasoner must decide when direct observation is sufficient and when to reconstruct, measure, or change reference frames. We present SpatialBrain, a framework that retains experience about these decisions while keeping its vision-language backbone fixed.
By Yifan Zhou, Ziyang Gong, Xiaojun Ye · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.