CSL & SLR500 Dataset

Jan 1, 2015

Download (*Important )

The CSL databases are released to universities and research institutes for research purpose only. To request the access right to the data resources, please follow the instructions below:

  1. Download the CSL Dataset Release Agreement;
  2. Read it carefully;
  3. Complete it appropriately. Note that the agreement should be signed by a full-time staff member (that is, the student is not acceptable).
  4. Please scan the signed agreement and send it to (ustc_vslrg At 126.com) and CC to Prof. Zhou (zhwg At ustc.edu.cn) and the full-time staff member who sign the agreement.

Introduction

We have two Chinese sign language datasets for isolated Sign Language Recognition and continuous Sign Language Recognition, respectively. Both datasets are collected with Kinect 2.0 by 50 signers. Each signer perfors 5 times for every word (sentence). The sign videos are recorded with 30fps. The distance between the signers and Kinect is about 1.5 meters. Each instance in both datasets contains RGB videos, depth videos, and 3D joints information of the signer.

Since the dataset is recorded with Microsoft Kinect, there are three data modalities available:

  1. RGB videos with resolution of 1280 x 720 pixels and frame rate of 30 fps.
  2. Depth videos with resolution of 512x 424 pixels and frame rate of 30 fps.
  3. Twenty-five skeleton joints locations of each frame.

Isolated SLR

The isolated SLR dataset contains 500 Chinese sign words. Each sign video is performed by 50 signers with 5 times. Hence, there are 250 instances for each sign word.

If you use this Chinese isolated SLR dataset in your research, please consider citing the following papers:

  • Hezhen Hu, Weichao Zhao, Wengang Zhou, and Houqiang Li, "SignBERT+: Hand-model-aware Self-supervised Pre-training for Sign Language Understanding," IEEE Trans. Pattern Anal. Mach. Intell. (TPAMI), 2023.
  • Weichao Zhao, Hezhen Hu, Wengang Zhou, Jiaxin Shi, and Houqiang Li, "BEST: BERT Pre-Training for Sign Language Recognition with Coupling Tokenization," Conference on Artificial Intelligence (AAAI), 2023.
  • Jie Huang, Wengang Zhou, Houqiang Li, and Weiping Li, "Attention based 3D-CNNs for Large-Vocabulary Sign Language Recognition, IEEE Trans. Circuits Syst. Video Technol. (TCSVT), 2018.

Continuous SLR

The corpus of continuous SLR dataset contains 100 Chinese sentence. There are 250 instances (50signers x 5times) for each sentence.

If you use this Chinese continuous SLR dataset in your research, please consider citing the following papers:

  • Junfu Pu, Wengang Zhou, and Houqiang Li, "Iterative Alignment Network for Continuous Sign Language Recognition," Conference on Computer Vision and Pattern Recognition (CVPR), 2019.
  • Hao Zhou, Wengang Zhou, and Houqiang Li, "Dynamic Pseudo Label Decoding for Continuous Sign Language Recognition," International Conference on Multimedia and Expo (ICME), 2019.
  • Jie Huang, Wengang Zhou, Qilin Zhang, Houqiang Li and Weiping Li, "Video-based Sign Language Recognition without Temporal Segmentation," AAAI Conference on Artificial Intelligence (AAAI), 2018.

Besides, you can refer to the following papers for continuous SLR published by our group:

  • Hezhen Hu, Weichao Zhao, Wengang Zhou, and Houqiang Li, "SignBERT+: Hand-model-aware Self-supervised Pre-training for Sign Language Understanding," IEEE Trans. Pattern Anal. Mach. Intell. (TPAMI), 2023.
  • Hezhen Hu, Junfu Pu, Wengang Zhou, and Houqiang Li, "Collaborative Multilingual Continuous Sign Language Recognition: A Unified Framework," IEEE Trans. Multimedia (TMM), 2022.
  • Junfu Pu, Wengang Zhou, Hezhen Hu, and Houqiang Li, "Boosting Continuous Sign Language Recognition via Cross Modality Augmentation," ACM International Conference on Multimedia (ACM MM), 2020.

Contact

If you have any questions about the dataset and our papers, please feel free to contact us: