VidHal: Benchmarking Hallucinations in Vision LLMs

TMLR Paper7249 Authors

Abstract: Vision Large Language Models (VLLMs) are widely acknowledged to be prone to hallucinations. Existing research addressing this problem has primarily been confined to image inputs, with sparse exploration of their video-based counterparts. Furthermore, current evaluation methods fail to capture nuanced errors in generated responses, which are often exacerbated by the rich spatiotemporal dynamics of videos. To address these two limitations, we introduce VidHal, a benchmark specially designed to evaluate video-based hallucinations in VLLMs. VidHal is constructed by bootstrapping video instances across a wide range of common temporal aspects. A defining feature of our benchmark lies in the careful creation of captions representing varying levels of hallucination associated with each video. To enable fine-grained evaluation, we propose a novel caption ordering task requiring VLLMs to rank captions by hallucinatory extent. We conduct extensive experiments on VidHal and comprehensively evaluated a broad selection of models, including both open-source and proprietary ones. Our results uncover significant limitations in existing VLLMs with respect to video-based hallucination generation. Through our benchmark, we aim to inspire further research on i) holistic understanding of VLLM capabilities, particularly regarding hallucination, and ii) advancing VLLMs to alleviate this problem.

Submission Type: Regular submission (no more than 12 pages of main content)

Assigned Action Editor: ~Ali_Etemad1

Submission Number: 7249