DiCoBench: Benchmarking Multi-Image Fine-Grained Perception via Differential and Commonality Visual Cues

2026-06-25

AI Summary

Existing multimodal large language model benchmarks predominantly rely on explicit textual prompts or low-resolution images, making them inadequate for evaluating a model’s autonomous perception of implicit visual cues at high resolution. To address this gap, this work introduces the first systematic benchmark specifically designed for cross-image, high-resolution fine-grained perception, comprising 765 near-2K multi-image samples across eight perception tasks organized into two tracks: discriminative and commonality-based visual reasoning. Evaluation is conducted via multiple-choice questions paired with human-curated fine-grained annotations. Comprehensive assessment of 18 state-of-the-art models reveals substantial performance gaps compared to human accuracy (98.3%), highlighting a critical deficiency in current models’ ability to interpret micro-scale visual details.

Abstract

Recent advancements in Multimodal Large Language Models (MLLMs) have demonstrated impressive fine-grained perception capabilities. However, existing benchmarks predominantly rely on explicit textual cues or low-resolution inputs, failing to evaluate a model's ability to autonomously perceive implicit visual cues in high-resolution. To bridge this gap, we introduce DiCoBench, a comprehensive, multi-image high-resolution benchmark designed for cross-image fine-grained perception. DiCoBench consists of 765 meticulously curated samples categorized into two progressive tracks: Differential Visual Cues and Commonality Visual Cues, covering 8 distinct perception tasks. By formulating the benchmark as a multiple-choice question task and utilizing high-resolution imagery (approaching 2K), we eliminate evaluation metric bias and pose a substantial challenge to current state-of-the-art MLLMs. Our extensive evaluation of 18 diverse MLLMs reveals a striking performance gap compared to human accuracy (98.3\%), with top-performing models struggling significantly with micro-scale detail capture. We believe DiCoBench will serve as a challenging testbed to drive future research in autonomous, high-resolution multi-image perception.

Problem

• fine-grained perception
• multimodal large language models
• high-resolution imagery
• visual cues
• multi-image benchmark

Innovation

• fine-grained perception
• multi-image benchmark
• differential visual cues
• commonality visual cues
• high-resolution MLLM evaluation