RLHF-V: Towards Trustworthy MLLMs via Behavior Alignment from Fine-grained Correctional Human Feedback

TL;DR

RLHF-V improves multimodal large language models by using fine-grained human feedback to significantly reduce hallucinations, enhancing trustworthiness with high data and computational efficiency.

Contribution

The paper introduces RLHF-V, a novel method that leverages segment-level human feedback for behavior alignment, significantly reducing hallucinations in MLLMs.

Findings

Reduces hallucination rate by 34.8% with only 1.4k data samples.

Outperforms models trained on larger datasets in trustworthiness.

Achieves state-of-the-art trustworthiness among open-source MLLMs.

Abstract

Multimodal Large Language Models (MLLMs) have recently demonstrated impressive capabilities in multimodal understanding, reasoning, and interaction. However, existing MLLMs prevalently suffer from serious hallucination problems, generating text that is not factually grounded in associated images. The problem makes existing MLLMs untrustworthy and thus impractical in real-world (especially high-stakes) applications. To address the challenge, we present RLHF-V, which enhances MLLM trustworthiness via behavior alignment from fine-grained correctional human feedback. Specifically, RLHF-V collects human preference in the form of segment-level corrections on hallucinations, and performs dense direct preference optimization over the human feedback. Comprehensive experiments on five benchmarks in both automatic and human evaluation show that, RLHF-V can enable substantially more trustworthy MLLM behaviors with promising data and computation efficiency. Remarkably, using 1.4k annotated data samples, RLHF-V significantly reduces the hallucination rate of the base MLLM by 34.8%, outperforming the concurrent LLaVA-RLHF trained on 10k annotated data. The final model achieves state-of-the-art performance in trustworthiness among open-source MLLMs, and shows better robustness than GPT-4V in preventing hallucinations aroused from over-generalization. We open-source our code, model, and data at https://github.com/RLHF-V/RLHF-V.