AI
SpatialSpeak: QA-Native Reconstruction with Local and Global Context for Spatial Chain-of-Thought Reasoning
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 26/09/2026, 20:00:00 UTC
Briefing created: 29/09/2026, 03:23:42 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 03:18:24 UTC.
Briefings
SpatialSpeak is a two-stage framework that combines QA-native reconstruction pretraining with spatial chain-of-thought reasoning to improve multi-view spatial reasoning. The method uses marked-point 3D queries for local geometry and object-center queries for global context, sharing a text-based output interface. ReVSI experiments show the framework increased the gain from CoT-VC from 2.6 to 6.9 points, achieving a ReVSI score of 62.8, which exceeds the strongest baseline by 8.7 points.
Key points
- The framework introduces SpatialSpeak, a two-stage system combining QA-native reconstruction pretraining with spatial chain-of-thought reasoning.
- ReVSI experiments demonstrate that the framework increased the gain from CoT-VC from 2.6 to 6.9 points, achieving a ReVSI score of 62.8.
Context and limitations
The study was conducted on the ReVSI benchmark, with results reported on ReVSI, VSI-Bench, and SPAR-Bench. The evidence does not specify the exact model architecture used for the QA-RP or CoT-VC stages, nor does it provide details on the specific data splits or evaluation metrics beyond the benchmark scores.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 03:23:42 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeiahkoyd57fbcgnppw35hnvlznsdrrfmf6nhd2yqqqm246obrefcx4