VideoHallucer: Evaluating Intrinsic and Extrinsic Hallucinations in Large Video-Language Models
This work addresses the critical issue of hallucinations in video-language models for researchers and developers, providing a benchmark and analysis to guide improvements, though it is incremental as it builds on existing evaluation methods.
The paper tackles the problem of hallucinations in large video-language models by introducing VideoHallucer, the first comprehensive benchmark for evaluating intrinsic and extrinsic hallucinations, revealing that most current models have significant hallucination issues and achieving a 5.38% average improvement in hallucination resistance with their self-PEP framework.
Recent advancements in Multimodal Large Language Models (MLLMs) have extended their capabilities to video understanding. Yet, these models are often plagued by "hallucinations", where irrelevant or nonsensical content is generated, deviating from the actual video context. This work introduces VideoHallucer, the first comprehensive benchmark for hallucination detection in large video-language models (LVLMs). VideoHallucer categorizes hallucinations into two main types: intrinsic and extrinsic, offering further subcategories for detailed analysis, including object-relation, temporal, semantic detail, extrinsic factual, and extrinsic non-factual hallucinations. We adopt an adversarial binary VideoQA method for comprehensive evaluation, where pairs of basic and hallucinated questions are crafted strategically. By evaluating eleven LVLMs on VideoHallucer, we reveal that i) the majority of current models exhibit significant issues with hallucinations; ii) while scaling datasets and parameters improves models' ability to detect basic visual cues and counterfactuals, it provides limited benefit for detecting extrinsic factual hallucinations; iii) existing models are more adept at detecting facts than identifying hallucinations. As a byproduct, these analyses further instruct the development of our self-PEP framework, achieving an average of 5.38% improvement in hallucination resistance across all model architectures.