Video Action Differencing

James Burgess, Xiaohan Wang, Yuhui Zhang, Anita Rau, Alejandro Lozano, Lisa Dunlap, Trevor Darrell, Serena Yeung-Levy

Keywords: Video, Actions, Differencing, Zero-shot, benchmark

TL;DR: A new task and benchmark for comparing how an action is performed between two videos, with a zero-shot method

Abstract: How do two individuals differ when performing the same action? In this work, we introduce Video Action Differencing, the novel task of identifying subtle differences between videos of the same action, which has numerous applications, such as coaching and skill acquisition. To enable development on this new task, we first create VidDiffBench, a benchmark dataset containing 557 video pairs, with human annotations of 4,719 fine-grained action differences and 2,075 timestamps indicating where these differences occur. Our experiments demonstrate that VidDiffBench poses a significant challenge for state-of-the-art large multimodal models (LMMs), such as GPT-4o, Gemini 1.5 Pro, and Qwen2-VL. By analyzing the failure cases of LMMs on VidDiffBench, we highlight two key challenges for this task: frame-by-frame alignment and fine-grained frame comparison. To overcome these, we propose VidDiff, an agent-based system that breaks the task into three stages: action difference proposal, keyframe localization, and difference verification, each stage utilizing specialized foundation models. The VidDiff method outperforms these baseline LMMs. We release both the dataset and code to encourage and support future research in this domain.

Primary Area: datasets and benchmarks

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 14085