MMRefine: Unveiling the Obstacles to Robust Refinement in Multimodal Large Language Models

TL;DR

MMRefine is a benchmark designed to evaluate and analyze the error refinement capabilities of Multimodal Large Language Models across diverse scenarios and error types, aiming to improve reasoning during inference.

Contribution

The paper introduces MMRefine, a novel benchmark for assessing error refinement in MLLMs, including a detailed analysis framework and insights into current bottlenecks.

Findings

Identification of key bottlenecks in MLLMs refinement performance

Analysis of error types affecting refinement success

Benchmark results highlighting areas for future improvement

Abstract

This paper introduces MMRefine, a MultiModal Refinement benchmark designed to evaluate the error refinement capabilities of Multimodal Large Language Models (MLLMs). As the emphasis shifts toward enhancing reasoning during inference, MMRefine provides a framework that evaluates MLLMs' abilities to detect and correct errors across six distinct scenarios beyond just comparing final accuracy before and after refinement. Furthermore, the benchmark analyzes the refinement performance by categorizing errors into six error types. Experiments with various open and closed MLLMs reveal bottlenecks and factors impeding refinement performance, highlighting areas for improvement in effective reasoning enhancement. Our code and dataset are publicly available at https://github.com/naver-ai/MMRefine.