3ViewSense: Spatial and Mental Perspective Reasoning from Orthographic Views in Vision-Language Models

Current Large Language Models have achieved Olympiad-level logic, yet Vision-Language Models paradoxically falter on elementary spatial tasks like block counting. This capability mismatch reveals a critical "spatial intelligence gap," where models fail to construct coherent 3D mental representations from 2D observations. We uncover this gap via diagnostic analyses showing the bottleneck is a missing view-consistent spatial interface rather than insufficient visual features or weak reasoning. To bridge this, we introduce 3ViewSense, a framework that grounds spatial reasoning in Orthographic Views. Drawing on engineering cognition, we propose a "Simulate-and-Reason" mechanism that decomposes complex scenes into canonical orthographic projections to resolve geometric ambiguities. By aligning egocentric perceptions with these allocentric references, our method facilitates explicit mental rotation and reconstruction. Empirical results on spatial reasoning benchmarks demonstrate that our method significantly outperforms existing baselines, with consistent gains on occlusion-heavy counting and view-consistent spatial reasoning. The framework also improves the stability and consistency of spatial descriptions, offering a scalable path toward stronger spatial intelligence in multimodal systems.

Large language models can now solve very difficult math and logic problems, even at the level of Olympiad competitions. Yet when vision is added, they can still fail at simple spatial tasks, such as counting a small pile of blocks. This paper studies this gap and finds that the issue is not only seeing or reasoning, but also how visual information is mentally organized during reasoning. We propose 3ViewSense, a method inspired by engineering drawings. Instead of answering directly from a single image, it teaches a vision-language model to mentally imagine the scene from front, side, and top views, then combine these views to reason about the 3D layout. This view-based mental reasoning helps the model handle hidden blocks, object positions, and perspective changes more reliably.