More Images, More Problems? A Controlled Analysis of VLM Failure Modes

TL;DR

This paper introduces MIMIC, a benchmark for evaluating multi-image understanding in LVLMs, identifies key failure modes, and proposes data and optimization strategies that significantly improve model performance.

Contribution

The work presents MIMIC, a comprehensive benchmark for multi-image capabilities, and introduces novel data-generation and attention-masking methods to address identified weaknesses.

Findings

LVLMs struggle with cross-image information aggregation.

Proposed remedies improve multi-image task performance.

Enhanced models outperform previous state-of-the-art.

Abstract

Large Vision Language Models (LVLMs) have demonstrated remarkable capabilities, yet their proficiency in understanding and reasoning over multiple images remains largely unexplored. While existing benchmarks have initiated the evaluation of multi-image models, a comprehensive analysis of their core weaknesses and their causes is still lacking. In this work, we introduce MIMIC (Multi-Image Model Insights and Challenges), a new benchmark designed to rigorously evaluate the multi-image capabilities of LVLMs. Using MIMIC, we conduct a series of diagnostic experiments that reveal pervasive issues: LVLMs often fail to aggregate information across images and struggle to track or attend to multiple concepts simultaneously. To address these failures, we propose two novel complementary remedies. On the data side, we present a procedural data-generation strategy that composes single-image annotations into rich, targeted multi-image training examples. On the optimization side, we analyze layer-wise attention patterns and derive an attention-masking scheme tailored for multi-image inputs. Experiments substantially improved cross-image aggregation, while also enhancing performance on existing multi-image benchmarks, outperforming prior state of the art across tasks. Data and code will be made available at https://github.com/anurag-198/MIMIC.