Multi event video dataset improves long take generation
MEV: A Multi-Event Video Dataset for Long-Take Generation
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Many current video generation models struggle to create long videos that show a sequence of different events smoothly. The authors introduce MEV, a new dataset designed with videos that feature several clear, separate events labeled precisely with descriptions. This dataset helps models learn how to connect multiple events coherently in one long video. The authors also developed a detailed way to annotate videos so each event and the full video have meaningful captions. Using MEV, models fine-tuned with event boundaries produce better transitions and more consistent long videos.
What this means in practice
- •For video generation developers: Create longer, multi-event videos with smoother transitions using MEV's event-labeled data to fine-tune models.
- •For video editing software teams: Incorporate event-level video segmentation and captioning from MEV to enable finer-grained editing and content manipulation.
Authors
Peiyuan Zhu, Shaoan Xie, Yifan Shen, Wen Tian, Zijian Li, Guangyi Chen, Kun Zhang
Abstract
Current datasets for text-to-video generation are largely limited to short clips depicting a single event or coarse-grained captions. As a result, they provide limited supervision for modeling long-take temporal dependencies, often leading to temporally inconsistent generations when describing multiple events. In this paper, we introduce MEV, a multi-event video dataset designed for long-take video generation. MEV provides high-quality real-world videos where subjects can perform a sequence of distinct, non-overlapping events, featuring precise temporal boundaries and fine-grained procedural annotations. This event-centric discretization reduces confounding factors and enables models to support temporally localized conditioning and build long-take videos from atomic event units. Furthermore, we introduce a comprehensive annotation pipeline that segments videos into temporally grounded events, and generates structured captions at both the global level and the event level. In addition, under an event-centric evaluation protocol, we conduct multi-event generation experiments by fine-tuning on MEV with temporally grounded event conditioning, leading to improved long-take event transitions and more coherent content. We believe this work provides a publicly available resource for multi-event fine-tuning and event-centric evaluation, together with practical insights into learning from temporally structured video-text data.