README.md

CSPT

Official source codes for our TIP 2022 paper "Contrastive Self-Supervised Pre-Training for Video Quality Assessment".

Self-supervised training

Step 1. Prepare the training/validation data and txt files.

• each item in data files should be like "vid_X1_X2.png" (vid denotes the video name, X1 denotes the distortion number, X2 denotes the frame number)
• each item in .txt files should be like "vid.mp4" (vid denotes the video name)

Step 2. Uncertainty-based ranking to split target domain into subdomains by running:

$ python ./source/main.py

CSPT-pretrained weights

We only provide pre-trained model with resnet-50 backbone here. You can use this model to finetune on your own data.

Download link: CSPT-resnet50

Password: 6xwg

Environment

• Python 3.9.7
• Pytorch 1.11.0 Torchvision 0.12.0
• Cuda 11.3 Cudnn 8.2.1
• cv2

Citation

If you find this work useful for your research, please cite our paper:

@article{chen2021contrastive,
  title={Contrastive Self-Supervised Pre-Training for Video Quality Assessment},
  author={Chen, Pengfei and Li, Leida and Wu, Jinjian and Dong, Weisheng and Shi, Guangming},
  journal={IEEE Transactions on Image Processing},
  volume={31},
  pages={458--471},
  year={2021},
  publisher={IEEE}
}