Benchmarking Multi-Image Understanding in Vision and Language Models: Perception, Knowledge, Reasoning, and Multi-Hop Reasoning

TL;DR

This paper introduces MIRB, a comprehensive benchmark for evaluating vision and language models' ability to understand and reason across multiple images, highlighting current limitations and gaps in multi-image reasoning capabilities.

Contribution

We present MIRB, the first benchmark specifically designed for multi-image understanding in vision and language models, covering perception, knowledge, reasoning, and multi-hop reasoning.

Findings

Open-source VLMs approach GPT-4V in single-image tasks.

Significant performance gap exists in multi-image reasoning tasks.

GPT-4V still struggles with the MIRB benchmark.

Abstract

The advancement of large language models (LLMs) has significantly broadened the scope of applications in natural language processing, with multi-modal LLMs extending these capabilities to integrate and interpret visual data. However, existing benchmarks for visual language models (VLMs) predominantly focus on single-image inputs, neglecting the crucial aspect of multi-image understanding. In this paper, we introduce a Multi-Image Relational Benchmark MIRB, designed to evaluate VLMs' ability to compare, analyze, and reason across multiple images. Our benchmark encompasses four categories: perception, visual world knowledge, reasoning, and multi-hop reasoning. Through a comprehensive evaluation of a wide range of open-source and closed-source models, we demonstrate that while open-source VLMs were shown to approach the performance of GPT-4V in single-image tasks, a significant performance gap remains in multi-image reasoning tasks. Our findings also reveal that even the state-of-the-art GPT-4V model struggles with our benchmark, underscoring the need for further research and development in this area. We believe our contribution of MIRB could serve as a testbed for developing the next-generation multi-modal models.