From Correspondence to Actions: Human-Like Multi-Image Spatial Reasoning in Multi-modal Large Language Models
This work addresses a specific problem in AI for improving spatial reasoning in multimodal models, representing an incremental advancement with novel training objectives.
The paper tackled the challenge of multi-image spatial reasoning in multimodal large language models by proposing HATCH, a training framework that incorporates cross-view correspondence and stepwise viewpoint transformation, resulting in consistent outperformance of baselines and competitive results against larger models.
While multimodal large language models (MLLMs) have made substantial progress in single-image spatial reasoning, multi-image spatial reasoning, which requires integration of information from multiple viewpoints, remains challenging. Cognitive studies suggest that humans address such tasks through two mechanisms: cross-view correspondence, which identifies regions across different views that correspond to the same physical locations, and stepwise viewpoint transformation, which composes relative viewpoint changes sequentially. However, existing studies incorporate these mechanisms only partially and often implicitly, without explicit supervision for both. We propose Human-Aware Training for Cross-view correspondence and viewpoint cHange (HATCH), a training framework with two complementary objectives: (1) Patch-Level Spatial Alignment, which encourages patch representations to align across views for spatially corresponding regions, and (2) Action-then-Answer Reasoning, which requires the model to generate explicit viewpoint transition actions before predicting the final answer. Experiments on three benchmarks demonstrate that HATCH consistently outperforms baselines of comparable size by a clear margin and achieves competitive results against much larger models, while preserving single-image reasoning capabilities.