CrossView: Can Vision-Language Models Reason Across Cameras?

2026-08-16

AI Summary

This study addresses the deficiency in cross-view reasoning capabilities within existing video understanding models by introducing CrossView, the first systematic multi-camera video question-answering benchmark encompassing autonomous driving and surveillance scenarios. By releasing an open-source dataset and evaluation framework, this work establishes a new standard for multi-view multimodal assessment. Experiments reveal that even state-of-the-art models like GPT-5.2 achieve suboptimal accuracy, with open-source alternatives exhibiting significant performance gaps. Notably, model performance correlates positively with enhanced joint multi-view processing capabilities, effectively exposing current bottlenecks in cross-view reasoning. These findings validate the benchmark’s efficacy and provide critical infrastructure for advancing research in multi-perspective video understanding.

Abstract

Video understanding benchmarks have long centered on single-camera settings, where modern multi-modal language models achieve strong performance across image and video tasks. Yet, the real world runs on multi-camera networks: autonomous vehicles, security systems, and robots all gather data across many simultaneous views. We argue that this is not simply "more" of the single-camera problem; it is fundamentally different. Multi-camera reasoning requires handling context that scales with the number of views, resolving occlusions visible from only a subset of cameras, judging which views matter, and integrating evidence across perspectives that may overlap or diverge. Current models struggle with exactly these challenges, yet no benchmark systematically targets them. We introduce CrossView, a multi-camera video question-answering benchmark spanning autonomous driving, security surveillance, egocentric/exocentric video, and robotics. Evaluation of proprietary models, such as GPT-5.2, and open-source models, like Qwen3-VL, reveals consistently low accuracy, with open-source models trailing by a wide margin. Performance scales strongly with a model's ability to jointly process multiple viewpoints, positioning CrossView as a rigorous benchmark for multi-camera video. We open-source our code and dataset at https://utaustin-swarmlab.github.io/CrossView.

Problem

• Multi-camera reasoning
• Video understanding benchmark
• Vision-Language Models
• Cross-view reasoning

Innovation

• Multi-camera reasoning
• Cross-view benchmark
• Video question answering
• Vision-language models
• Multi-view integration