README.md

---
task_categories:
- robotics
---

<div align="center">
<h1> <img src="assets/icon.png" width="50" /> Humanoid-X </h1>
</div>
<h5 align="center">
    <a href="https://usc-gvl.github.io/UH-1/">🌐 Homepage</a> | <a href="https://huggingface.co/datasets/USC-GVL/Humanoid-X">⛁ Dataset</a> | <a href="https://huggingface.co/USC-GVL/UH-1">🤗 Models</a> | <a href="https://arxiv.org/abs/2412.14172">📑 Paper</a> | <a href="https://github.com/sihengz02/UH-1">💻 Code</a>
</h5>


This repo contains the officail dataset for the paper "[Learning from Massive Human Videos for Universal Humanoid Pose Control](https://arxiv.org/abs/2412.14172)"
If you like our project, please give us a star ⭐ on GitHub for latest update.

![Alt text](assets/teaser.png)

- In this repo, we fully release the text desciption data `texts.zip`, humanoid keypoints data `humanoid_keypoint.zip`, and humanoid actions data `humanoid_action.zip`.

- We only release part of the human poses data (charades subset, kinetics700 subset, and youtube subset) `human_pose.zip` due to license issues. Instead, we provide instructions on how to obtain other parts of human poses data: [HumanML3D/AMASS](https://github.com/EricGuo5513/HumanML3D), [Motion-X](https://github.com/IDEA-Research/Motion-X?tab=readme-ov-file#-dataset-download).

- We release the train, test, and valid set split as `train.txt`, `test.txt`, and `val.txt`.

- We will not release the original Internet videos to protect copyright.

# Dataset Statistics

![Alt text](assets/dataset.png)

# Dataset Collection Pipeline

![Alt text](assets/annotation.png)

# Citation

If you find our work helpful, please cite us:

```bibtex
@article{mao2024learning,
  title={Learning from Massive Human Videos for Universal Humanoid Pose Control},
  author={Mao, Jiageng and Zhao, Siheng and Song, Siqi and Shi, Tianheng and Ye, Junjie and Zhang, Mingtong and Geng, Haoran and Malik, Jitendra and Guizilini, Vitor and Wang, Yue},
  journal={arXiv preprint arXiv:2412.14172},
  year={2024}
}
```