CIEM: Contrastive Instruction Evaluation Method for Better Instruction Tuning

TL;DR

This paper introduces CIEM, an automatic evaluation pipeline for hallucination in vision-language models, and CIT, a new instruction tuning method that reduces hallucinations by generating high-quality factual and contrastive data for model training.

Contribution

The paper presents a novel contrastive evaluation method (CIEM) and an instruction tuning approach (CIT) that together improve hallucination detection and reduction in large vision-language models.

Findings

CIT-tuned VLMs outperform baseline models in hallucination mitigation.

CIEM effectively identifies hallucination issues in existing VLMs.

CIT enhances model factual accuracy and robustness.

Abstract

Nowadays, the research on Large Vision-Language Models (LVLMs) has been significantly promoted thanks to the success of Large Language Models (LLM). Nevertheless, these Vision-Language Models (VLMs) are suffering from the drawback of hallucination -- due to insufficient understanding of vision and language modalities, VLMs may generate incorrect perception information when doing downstream applications, for example, captioning a non-existent entity. To address the hallucination phenomenon, on the one hand, we introduce a Contrastive Instruction Evaluation Method (CIEM), which is an automatic pipeline that leverages an annotated image-text dataset coupled with an LLM to generate factual/contrastive question-answer pairs for the evaluation of the hallucination of VLMs. On the other hand, based on CIEM, we further propose a new instruction tuning method called CIT (the abbreviation of Contrastive Instruction Tuning) to alleviate the hallucination of VLMs by automatically producing high-quality factual/contrastive question-answer pairs and corresponding justifications for model tuning. Through extensive experiments on CIEM and CIT, we pinpoint the hallucination issues commonly present in existing VLMs, the disability of the current instruction-tuning dataset to handle the hallucination phenomenon and the superiority of CIT-tuned VLMs over both CIEM and public datasets.