How and What to Imagine? Visual Thinking in Unified Multimodal Models for Cross-View Spatial Reasoning
2026-05-26 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors studied how vision-language models can better understand scenes from different viewpoints by using images to help their reasoning, rather than just language. They introduced a technique called View Dropout to make models rely more on intermediate 'thinking images' when answering questions. They tested different kinds of visual thinking images and found that panoramic views combined with their method worked best for generalizing to new, real-world data. This shows a way to improve how models handle spatial reasoning across views.
vision-language modelsspatial reasoningmultimodal modelsView Dropoutthinking imagespanoramic renderingout-of-domain generalizationintermediate representationssynthetic scenes
Authors
Qian Yang, Ankur Sikarwar, Huy Le, Le Zhang, Zhuan Shi, Perouz Taslakian, Aishwarya Agrawal
Abstract
Cross-view spatial reasoning remains a weak spot for vision-language models (VLMs): they often reason in language and lose the fine-grained geometry needed for the task. Thinking with images aims to address this by generating an intermediate thinking image, but recent work shows that models often ignore the visual evidence in these traces. We therefore ask how to make visual thinking matter, and what kind of visual thinking works best. We study these questions in unified multimodal models (UMMs), which natively support interleaved image-text generation. For the first question, we propose View Dropout (VDrop), a training-time intervention that hides parts of one input view from the answer span while keeping them visible to the thinking-image tokens. This encourages the model to use the thinking image when answering, instead of relying only on the input views. Once the thinking image is used for answer prediction, we study which type of visual thinking is most effective. We frame this as a learnability-informativeness tradeoff and compare three thinking-image variants: top-down, panoramic, and point-matching renderings. Trained on synthetic scenes and evaluated on five real-world out-of-domain benchmarks, panoramic visual thinking with VDrop is the only configuration that is both informative and learnable, and it achieves the best out-of-domain generalization.