---
language:
- zh
license: cc-by-nc-4.0
size_categories:
- 100K<n<1M
tags:
- sign-language
- pose
task_categories:
- video-text-to-text
---

# Summary
This is the **pose format** dataset proposed in our paper "[**Uni-Sign: Toward Unified Sign Language Understanding at Scale**](https://arxiv.org/abs/2501.15187)".
CSL-News is a large-scale Chinese Sign Language dataset designed for developing robust sign language understanding models. 

**Code**: [https://github.com/ZechengLi19/Uni-Sign](https://github.com/ZechengLi19/Uni-Sign)

# Download
Please refer to [**download script**](https://github.com/ZechengLi19/Uni-Sign/blob/main/download_scripts/download_CSL_News.py) to download CSL_News. 
You can also download each file by ```wget```, for instance:
```
wget https://huggingface.co/datasets/ZechengLi19/CSL-News_pose/resolve/main/archive_001.zip
wget https://huggingface.co/datasets/ZechengLi19/CSL-News_pose/resolve/main/archive_002.zip
wget https://huggingface.co/datasets/ZechengLi19/CSL-News_pose/resolve/main/archive_003.zip
...
```

# Usage
You can unzip each `archive_*.zip` file by ```unzip```, for instance:
```
unzip -j archive_001.zip -d ./CSL_News/pose_format
unzip -j archive_002.zip -d ./CSL_News/pose_format
unzip -j archive_003.zip -d ./CSL_News/pose_format
...
```
Each zip file contains multiple pickle files. Each pickle file stores a dictionary containing the keys: `keypoints`, `scores`, and `w_h`. 

``CSL_News_Labels.json`` and ``CSL_News_Labels.csv`` contains the text-annotations for CSL-News. They can easily be read by
```python
# Read CSL_News_Labels.json
import json
with open('CSL_News_Labels.json', 'r', encoding='utf-8') as f:
    data = json.load(f)
# Read CSL_News_Labels.csv
import pandas
data = pandas.read_csv("CSL_News_Labels.csv")
```

# Other format
The **RGB format** CSL-News dataset can be found [**here**](https://huggingface.co/datasets/ZechengLi19/CSL-News).

# License
CSL-News is released under the CC-BY-NC-4.0 license. The video samples in this dataset are collected from publicly available web videos. Users must ensure that their use of these video samples is strictly non-commercial.

# Why Non-Commercial?
The video samples in CSL-News are sourced from web videos, and their copyright belongs to the original content creators. While this dataset is provided for research purposes under the CC-BY-NC-4.0 license, commercial use of these videos may infringe upon the rights of the original creators. To respect their rights and ensure ethical use, we strictly enforce a non-commercial usage policy for CSL-News.

# Citation
```
@article{li2025uni-sign,
  title={Uni-Sign: Toward Unified Sign Language Understanding at Scale},
  author={Li, Zecheng and Zhou, Wengang and Zhao, Weichao and Wu, Kepeng and Hu, Hezhen and Li, Houqiang},
  journal={arXiv preprint arXiv:2501.15187},
  year={2025}
}
```