A Database for Talking Heads Quality Assessment (THQA) 🚀🚀🚀

🌟 Official Access to the THQA database in "THQA: A PERCEPTUAL QUALITY ASSESSMENT DATABASE FOR TALKING HEADS" 🌟

Authors: Yingjie Zhou, Zicheng Zhang, Wei Sun, Xiaohong Liu, Xiongkuo Min, Zhihua Wang, Xiao-Ping Zhang, and Guangtao Zhai

Institution: Shanghai Jiaotong University, PengCheng Laboratory, Toronto Metropolitan University

| Research Gate | Arxiv |

The THQA database is a quality assessment database designed specifically for the person independent driving methods. It compares 8 popular speech-driven methods and contains a total of 800 talking head videos. If there are any questions or suggestions for improvement against the database, please feel free to contact us by zyj2000@sjtu.edu.cn.

👀 Preview

👀 How to Use Our Work for Your Convenience

1. Open Platform to validate your talking head driving methods subjectively.
2. Open Image and Audio (code = thqa) Resources are provided which can be employed to driving heads.
3. More objective methods will be provided soon!!!
4. More Function is going to be explored.

👀 How to Join Us?

1. If you have more upadated talking head results, welcome to share them to us by email or PR.
2. We will carry out more subjective experiments on your results and update work.
3. Thank for your support. Hope this work can promote each other!!!

🚧 More Information and Convenience is Coming ...

[7.22.2025] Our zero-shot THQA method MI3S is accepted by Information Processing & Management.

[6.26.2025] Our latest THQA work is accepted by ICCV25, presenting the THQA-10K dataset as well as the FSCD quality assessment method.

[6.26.2025] We present the THQA-NTIRE dataset at CVPR NTIRE2025, hold XGC Quality Assessment Track 3: Talking Head.

[10.12.2024] The poster for ICIP2024 THQA has been released.

[9.25.2024] A derivative work ReLI-QA from the THQA dataset is accepted by VCIP2024.

[7.17.2024] A further research THQA-3D is accepted by ACM MM2024.

[6.14.2024] Release the original audios in Baidu YunPan.

[6.07.2024] Our paper is accepted by ICIP2024.

[4.22.2024] Poster our paper in arxiv.

[4.13.2024] Release the original images.

[3.01.2024] Poster our paper in Reseach Gate.

🌋 Distortions

💡 Links

The database is nearly 732MB, and 2 links are provided here.

Baidu YunPan | Onedrive

📑 Citation

@article{zhou2024thqa,
  title={THQA: A Perceptual Quality Assessment Database for Talking Heads},
  author={Zhou, Yingjie and Zhang, Zicheng and Sun, Wei and Liu, Xiaohong and Min, Xiongkuo and Wang, Zhihua and Zhang, Xiao-Ping and Zhai, Guangtao},
  journal={arXiv preprint arXiv:2404.09003},
  year={2024}
}