DiSR splits spatial reasoning into 3D perception by frozen expert models and language-based reasoning by a LoRA-tuned LLM, reporting state-of-the-art results on 3DSRBench and SPAR-Bench with 0.33M training samples.
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Disentangling 3D Modeling from Spatial Reasoning
DiSR splits spatial reasoning into 3D perception by frozen expert models and language-based reasoning by a LoRA-tuned LLM, reporting state-of-the-art results on 3DSRBench and SPAR-Bench with 0.33M training samples.