Q-GeoMem: Question-Guided Geometric Memory for Video Spatial Reasoning

2026-05-26

AI Summary

Existing video spatial reasoning methods often employ generic memory mechanisms that introduce redundant or irrelevant geometric information, thereby impairing long-term reasoning capabilities. This work proposes a question-guided geometric memory framework that efficiently accumulates and integrates spatial evidence relevant and novel to the given question. The framework injects geometric features conditioned on camera poses and employs a dual-memory architecture comprising a fine-grained contextual memory bank and a semantic-geometric evidence bank. A Q-Former module evaluates the relevance and novelty of incoming evidence relative to the question, enabling adaptive read/write operations and dynamic memory capacity control. Experiments on VSI-Bench and VSTI-Bench demonstrate state-of-the-art performance, validating the effectiveness and novelty of the proposed approach.

Abstract

Video spatial reasoning requires accumulating viewpoint-dependent evidence over time while retaining information useful to the question being asked. Existing spatial video-language models improve geometric perception and long-range context modeling, but often treat memory as a generic temporal cache, which can introduce redundant or irrelevant geometry and weaken long-horizon reasoning. We propose \textbf{\ours}, a question-guided geometric memory framework for video spatial reasoning. \ours injects camera-conditioned geometry into visual tokens and maintains two complementary memories: a Fine-Grained Context Bank for recent dense features and camera states, and a Semantic-Geometric Evidence Bank for compact long-range evidence. Each candidate frame is scored by the product of Q-Former-based question relevance and novelty with respect to the retained bank; this score is stored and reused during reading, while a capacity-based replacement rule keeps the bank compact. During reasoning, both memories are read before update and adaptively fused with the current frame representation. Experiments on VSI-Bench and VSTI-Bench show that \ours achieves state-of-the-art performance among evaluated spatial reasoning models, validating the effectiveness of question-guided geometric memory. Ablations further verify the contribution of the proposed evidence scoring mechanism.

Problem

• video spatial reasoning
• geometric memory
• question-guided
• long-horizon reasoning
• spatial video-language models

Innovation

• question-guided memory
• geometric reasoning
• video spatial reasoning
• evidence scoring
• complementary memory banks