VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding
This addresses the issue of overestimating AI reasoning abilities in video understanding for researchers and developers, though it is incremental as it builds on existing evaluation methods.
The paper tackles the problem of evaluating whether vision-language models truly comprehend visual content by proposing negative-control tests with synthetic videos depicting physically impossible or logically inconsistent events, and finds that leading models often miss these violations, but reinforcement learning fine-tuning on the dataset improves recognition without reducing standard benchmark performance.
Vision-Language Models (VLMs) have achieved strong results in video understanding, yet a key question remains: do they truly comprehend visual content or only learn shallow correlations between vision and language? Real visual understanding, especially of physics and common sense, is essential for AI systems that interact with the physical world. Current evaluations mostly use real-world videos similar to training data, so high benchmark scores may not reflect real reasoning ability. To address this, we propose negative-control tests using videos that depict physically impossible or logically inconsistent events. We introduce VideoHallu, a synthetic dataset of physics- and commonsense-violating scenes generated with Veo2, Sora, and Kling. It includes expert-annotated question-answer pairs across four categories of violations. Tests of leading VLMs (Qwen-2.5-VL, Video-R1, VideoChat-R1) show that, despite strong results on benchmarks such as MVBench and MMVU, they often miss these violations, exposing gaps in visual reasoning. Reinforcement learning fine-tuning on VideoHallu improves recognition of such violations without reducing standard benchmark performance. Our data is available at https://github.com/zli12321/VideoHallu.git.