Mementos: A Comprehensive Benchmark for Multimodal Large Language Model Reasoning over Image Sequences

TL;DR

Mementos introduces a new benchmark with 4,761 image sequences to evaluate multimodal large language models' ability to reason over dynamic visual information, revealing current models' struggles with sequential image understanding.

Contribution

This paper presents Mementos, the first comprehensive benchmark for assessing MLLMs' reasoning over image sequences, highlighting their limitations and influencing factors.

Findings

MLLMs often hallucinate or misrepresent objects and behaviors in sequences.

GPT-4V and Gemini show significant challenges in dynamic scene understanding.

Key factors include object-behavior correlation and co-occurring behaviors.

Abstract

Multimodal Large Language Models (MLLMs) have demonstrated proficiency in handling a variety of visual-language tasks. However, current MLLM benchmarks are predominantly designed to evaluate reasoning based on static information about a single image, and the ability of modern MLLMs to extrapolate from image sequences, which is essential for understanding our ever-changing world, has been less investigated. To address this challenge, this paper introduces Mementos, a new benchmark designed to assess MLLMs' sequential image reasoning abilities. Mementos features 4,761 diverse image sequences with varying lengths. We also employ a GPT-4 assisted method to evaluate MLLM reasoning performance. Through a careful evaluation of nine recent MLLMs on Mementos, including GPT-4V and Gemini, we find that they struggle to accurately describe dynamic information about given image sequences, often leading to hallucinations/misrepresentations of objects and their corresponding behaviors. Our quantitative analysis and case studies identify three key factors impacting MLLMs' sequential image reasoning: the correlation between object and behavioral hallucinations, the influence of cooccurring behaviors, and the compounding impact of behavioral hallucinations. Our dataset is available at https://github.com/umd-huang-lab/Mementos.