Direct Preference Optimization of Video Large Multimodal Models from Language Model Reward
This addresses the problem of accurate factuality assessment in video QA for researchers and developers, representing an incremental advancement by adapting existing DPO techniques to video LMMs with a novel proxy approach.
The paper tackles the challenge of providing informative feedback for video instruction-following tasks, particularly in detecting hallucinations, by introducing a framework that uses detailed video captions as a proxy for video content to enable language models to score video QA predictions. The result shows robust alignment with GPT-4V's reward mechanism and significant performance improvements on video QA tasks through DPO.
Preference modeling techniques, such as direct preference optimization (DPO), has shown effective in enhancing the generalization abilities of large language model (LLM). However, in tasks involving video instruction-following, providing informative feedback, especially for detecting hallucinations in generated responses, remains a significant challenge. Previous studies have explored using large large multimodal models (LMMs) as reward models to guide preference modeling, but their ability to accurately assess the factuality of generated responses compared to corresponding videos has not been conclusively established. This paper introduces a novel framework that utilizes detailed video captions as a proxy of video content, enabling language models to incorporate this information as supporting evidence for scoring video Question Answering (QA) predictions. Our approach demonstrates robust alignment with OpenAI GPT-4V model's reward mechanism, which directly takes video frames as input. Furthermore, we show that applying this tailored reward through DPO significantly improves the performance of video LMMs on video QA tasks.