MIA-Bench: Towards Better Instruction Following Evaluation of Multimodal LLMs

TL;DR

MIA-Bench is a comprehensive benchmark with 400 challenging image-prompt pairs designed to evaluate and improve the instruction-following capabilities of multimodal large language models, emphasizing adherence to complex instructions.

Contribution

The paper introduces MIA-Bench, a new benchmark for assessing instruction adherence in MLLMs, and explores supervised fine-tuning to enhance compliance without performance loss.

Findings

Significant performance variation among state-of-the-art MLLMs.

Fine-tuning improves instruction adherence.

Benchmark guides future MLLM development.

Abstract

We introduce MIA-Bench, a new benchmark designed to evaluate multimodal large language models (MLLMs) on their ability to strictly adhere to complex instructions. Our benchmark comprises a diverse set of 400 image-prompt pairs, each crafted to challenge the models' compliance with layered instructions in generating accurate responses that satisfy specific requested patterns. Evaluation results from a wide array of state-of-the-art MLLMs reveal significant variations in performance, highlighting areas for improvement in instruction fidelity. Additionally, we create extra training data and explore supervised fine-tuning to enhance the models' ability to strictly follow instructions without compromising performance on other tasks. We hope this benchmark not only serves as a tool for measuring MLLM adherence to instructions, but also guides future developments in MLLM training methods.