Mitigating Fine-Grained Hallucination by Fine-Tuning Large Vision-Language Models with Caption Rewrites

TL;DR

This paper introduces ReCaption, a framework that reduces fine-grained hallucinations in large vision-language models by rewriting captions with ChatGPT and fine-tuning the models, leading to improved accuracy and generation quality.

Contribution

It proposes a novel approach combining caption rewriting and fine-tuning to specifically target and mitigate fine-grained hallucinations in LVLMs, with a new evaluation method for detailed hallucination assessment.

Findings

ReCaption significantly reduces fine-grained hallucinations across different LVLMs.

The approach improves the quality of generated text in vision-language tasks.

The proposed evaluation method effectively measures fine-grained hallucination levels.

Abstract

Large language models (LLMs) have shown remarkable performance in natural language processing (NLP) tasks. To comprehend and execute diverse human instructions over image data, instruction-tuned large vision-language models (LVLMs) have been introduced. However, LVLMs may suffer from different types of object hallucinations. Nevertheless, LVLMs are evaluated for coarse-grained object hallucinations only (i.e., generated objects non-existent in the input image). The fine-grained object attributes and behaviors non-existent in the image may still be generated but not measured by the current evaluation methods. In this paper, we thus focus on reducing fine-grained hallucinations of LVLMs. We propose \textit{ReCaption}, a framework that consists of two components: rewriting captions using ChatGPT and fine-tuning the instruction-tuned LVLMs on the rewritten captions. We also propose a fine-grained probing-based evaluation method named \textit{Fine-Grained Object Hallucination Evaluation} (\textit{FGHE}). Our experiment results demonstrate that ReCaption effectively reduces fine-grained object hallucination for different LVLM options and improves their text generation quality. The code can be found at https://github.com/Anonymousanoy/FOHE.