VidComposition: Can MLLMs Analyze Compositions in Compiled Videos?

TL;DR

VidComposition introduces a benchmark to evaluate multimodal models' ability to understand complex video compositions, revealing significant gaps between human and model performance in interpreting compiled videos.

Contribution

The paper presents a new benchmark with curated videos and annotations to assess MLLMs' understanding of video composition, highlighting current limitations.

Findings

Current MLLMs perform significantly worse than humans in video composition understanding.

The benchmark covers diverse aspects like camera movement, shot size, and character actions.

Evaluation of 33 models shows notable performance gaps, guiding future research.

Abstract

The advancement of Multimodal Large Language Models (MLLMs) has enabled significant progress in multimodal understanding, expanding their capacity to analyze video content. However, existing evaluation benchmarks for MLLMs primarily focus on abstract video comprehension, lacking a detailed assessment of their ability to understand video compositions, the nuanced interpretation of how visual elements combine and interact within highly compiled video contexts. We introduce VidComposition, a new benchmark specifically designed to evaluate the video composition understanding capabilities of MLLMs using carefully curated compiled videos and cinematic-level annotations. VidComposition includes 982 videos with 1706 multiple-choice questions, covering various compositional aspects such as camera movement, angle, shot size, narrative structure, character actions and emotions, etc. Our comprehensive evaluation of 33 open-source and proprietary MLLMs reveals a significant performance gap between human and model capabilities. This highlights the limitations of current MLLMs in understanding complex, compiled video compositions and offers insights into areas for further improvement. The leaderboard and evaluation code are available at https://yunlong10.github.io/VidComposition/