ConsiSpace: Learning Geometric Consistency Matters for Video Spatial Reasoning
2026-07-20 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors developed a method called ConsiSpace to help AI models better understand spatial relationships in long videos, especially when the viewpoint changes. Their approach organizes spatial information in a way that keeps it consistent across different views and uses special training techniques to improve reasoning stability. This method was tested on three benchmarks and showed significant improvements compared to previous models. Essentially, the authors focused on making spatial reasoning in videos more reliable by emphasizing geometric consistency.
video spatial reasoningmultimodal large language modelsgeometry-consistent memoryself-supervised learningreinforcement learningcross-view stabilityspatial evidencelong-video question answeringspatial relationssupervised fine-tuning
Authors
Ting Huang, Zhenyu Zhang, Wenyuan Huang, Jian Yang, Hao Tang
Abstract
Video spatial reasoning is essential for navigation-oriented perception and long-video question answering, where models must infer spatial relations across long horizons under changing viewpoints. However, existing multimodal large language models (MLLMs) remain largely semantic-centric, and often fail to reliably aggregate consistent spatial evidence from redundant video observations, leading to inefficient or unstable reasoning. To address these issues, we propose ConsiSpace, a geometry-consistency-aware framework for geometry-sensitive video spatial reasoning that turns spatial consistency into both an evidence organization principle and an explicit post-SFT learning signal. We build a geometry-consistent memory (GCM) including implicit evidence tokens and explicit geometric cues, and leverage efficient organization strategies to compactly preserve task-related spatial evidence. Furthermore, we utilize unified consistency self-supervised reinforcement learning (UC-SSRL) after supervised fine-tuning to improve cross-view stability, with answer-, metric-, and topology-consistency rewards. Extensive experiments on three spatial-reasoning benchmarks, VSI-Bench, OSI-Bench, and MMSI-Video-Bench, show consistent gains, improving the average score by 12.6 points over the strongest baselines.