Datasets:
license: mit
task_categories:
- image-to-video
language:
- en
tags:
- diffusion
- generation
- human
- animation
size_categories:
- 10K<n<100K
Data for Human Image Animation
🎏 Introduction
TL; DR: With the rapid developments in generative models, including the diffusion-based or the flow-based models, the human-centric tasks, like pose-driven human image animation, audio-driven action generation, diffusion-based pose estimation, human optical estimation, etc., have attracted a lot of attention from lots of works.
We pay attention to the quality of the training data of human data for these tasks. However, due to the lack of high-quality datasets, especially for the human image animation, we find it is hard to collect videos from existing public datasets, while these videos have these characteristics:
- High-resolution: the resolution of the vertical video is larger than 1080 * 576.
- High-dynamic: the video is vivid and suitable to learn human motions.
- Dancing-style: In this stage, we focus on the human animation task and mainly collect videos like TikTok styles.
⚔️ What we do
We collect a large number of videos from the internet. After filtering low-quality, limited motion, and bad frames, we get 25,000 videos in this repo. Now we provide a visualization to these data and the corresponding pose data, you can check each training video in our work.
Notice: we do not allow any commercial usage of these videos and you must delete them within 24 hours after downloading.
Tips: If you find that your data is being infringed upon, please contact us immediately to request its removal.
📍 Citation
If you find this guidance helpful, please consider citing:
@article{zhao2025dynamictrl,
title={DynamiCtrl: Rethinking the Basic Structure and the Role of Text for High-quality Human Image Animation},
author={Haoyu, Zhao and Zhongang, Qi and Cong, Wang and Qingping, Zheng and Guansong, Lu and Fei, Chen and Hang, Xu and Zuxuan, Wu},
year={2025},
journal={arXiv:2503.21246},
}