MVPBench: A Multi-Video Perception Evaluation Benchmark for Multi-Modal Video Understanding

TL;DR

MVPBench is a comprehensive benchmark designed to evaluate multi-modal models' ability to understand and analyze multiple videos simultaneously, addressing a gap in existing single-video or static image benchmarks.

Contribution

This paper introduces MVPBench, the first benchmark specifically targeting multi-video perception with diverse subtasks and extensive question-answering tests, to evaluate and improve multi-video understanding models.

Findings

Current models show limited ability to process multi-video inputs effectively.

MVPBench covers 14 diverse subtasks across various visual domains.

Extensive evaluations highlight significant challenges in multi-video comprehension.

Abstract

The rapid progress of Large Language Models (LLMs) has spurred growing interest in Multi-modal LLMs (MLLMs) and motivated the development of benchmarks to evaluate their perceptual and comprehension abilities. Existing benchmarks, however, are limited to static images or single videos, overlooking the complex interactions across multiple videos. To address this gap, we introduce the Multi-Video Perception Evaluation Benchmark (MVPBench), a new benchmark featuring 14 subtasks across diverse visual domains designed to evaluate models on extracting relevant information from video sequences to make informed decisions. MVPBench includes 5K question-answering tests involving 2.7K video clips sourced from existing datasets and manually annotated clips. Extensive evaluations reveal that current models struggle to process multi-video inputs effectively, underscoring substantial limitations in their multi-video comprehension. We anticipate MVPBench will drive advancements in multi-video perception.