Awesome

UCDA

This the the repository includes the official codes for our ICCV2021 paper "Unsupervised Curriculum Domain Adaptation for No-Reference Video Quality Assessment".

Training

We use the 10-fold cross-validation in our experiments. To reach to the comparable performance you may need to train a few times. Note that in the given training codes, no validation and test sets are used, you should split your own training/validation/test sets to save the best model during training.

Step 1. Conduct the domain adaptation between source and target domains by running:

$ python ./Sources/first_uda.py

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

$ python ./Sources/ranking.py

Step 3. Conduct the domain adaptation between subdomains by running:

$ python ./Sources/second_uda.py

Environment

• Python 3.6.5
• Pytorch 1.0.1
• Cuda 9.0 Cudnn 7.1

Citation

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

@InProceedings{Chen_2021_ICCV,
    author    = {Chen, Pengfei and Li, Leida and Wu, Jinjian and Dong, Weisheng and Shi, Guangming},
    title     = {Unsupervised Curriculum Domain Adaptation for No-Reference Video Quality Assessment},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2021},
    pages     = {5178-5187}
}