Pytorch Implementation of ICCV21 poster "Multi-Scale Separable Network for Ultra-High-Definition Video Deblurring" by Senyou Deng, Wenqi Ren*, Yanyang Yan, Tao Wang, Fenglong Song, and Xiaochun Cao.

The framework of our UHDVD model:

Some examples from our 4KRD dataset:
Please download training datasets (4KRD/GoPro/DVD/REDS) into './datas/XXXX'. 4KRD dataset here
Running the following command to obtain files: 'train_blur_list.txt' and 'train_sharp_list.txt' in './datas/XXXX'.

python txt_list.py

4KRD pretrained models are stored in './checkpoints/4KRD'.

For requires, run following commands.

pip install -r requirements.txt

For training, run following commands.

python train_video.py -se start_epoch -g GPU_number

For testing, put test samples as './test_set/XXXX/XXXX/blurry', then run following commands.

python test_video.py -g GPU_number

The results will be saved at './test_set/XXXX/XXXX/deblur_4KRD'.

If you think this work is useful for your research, please cite the following paper.

@inproceedings{deng2021multi,
  title={Multi-Scale Separable Network for Ultra-High-Definition Video Deblurring},
  author={Deng, Senyou and Ren, Wenqi and Yan, Yanyang and Wang, Tao and Song, Fenglong and Cao, Xiaochun},
  booktitle={the IEEE/CVF International Conference on Computer Vision (ICCV)},
  pages={14030--14039},
  year={2021}
}