MotionHalluc: Diagnosing Kinematic Hallucinations in Fine-Grained Motion Reasoning
For researchers and developers of large multimodal models, this work systematically identifies and measures a specific failure mode (motion hallucinations) in cross-video comparison tasks, providing a benchmark and a simple mitigation strategy.
The paper introduces MotionHalluc, a benchmark with 1540 questions over 553 video pairs to diagnose kinematic hallucinations in fine-grained motion reasoning. A training-free baseline (PPV) that injects explicit kinematic measurements reduces hallucinations by an average of 10.6% across models.
Motion instruction generation in cross-video comparison aims to produce corrective feedback that describes the differences between a query and a reference motion. However, existing models often generate instructions that exhibit motion hallucinations, failing to reflect actual kinematic differences between paired videos. To systematically investigate these hallucinations, we introduce MotionHalluc, a dedicated benchmark for evaluating motion hallucinations in paired-video comparison. MotionHalluc comprises 1540 fine-grained questions over 553 video pairs, evaluating hallucinations along three core dimensions: (1)directional hallucination, (2)attributional hallucination, and (3)temporal hallucination. Extensive evaluations of state-of-the-art large multimodal models demonstrate high susceptibility to these hallucinations. Furthermore, we provide Perceive-Parse-Verify (PPV) as a training-free measurements extraction and verification baseline that converts candidate instructions into executable measurement queries and supplies kinematic measurements at inference time. Our results show that this simple measurements injection yields an average 10.6% performance gain across models, suggesting that motion reasoning with explicit quantitative measurements is a key factor in reducing hallucinations in cross-video comparison. Our code and dataset will be made publicly available upon acceptance.