Revisiting Temporal Alignment for Video Restoration CVPR-2022

Kun Zhou, Wenbo Li, Liying Lu, Xiaoguang Han, Jiangbo Lu

[paper]

---

We have provided the source code of our video super-resolution, video deblurring, and video denoising models.

We provide our results at Google Cloud.

The pre-trained models are uploaded in the google cloud.

Some in-the-wild testing sequences are available here.

---

File Structure

Click to expand

libs

DcNv2

utils

common.py
core.py
model_opr.py

models

VDB

config.py
network.py
validate.py
sequence_test.py
load_VDB_Data.py
VideoDeblur.py

VDN

config.py
network.py
validate.py
validate_davis.py
sequence_test.py

VSR_REDS

config.py
network.py
validate.py

VSR_VIMEO90K

config.py
network.py
validate.py
sequence_test.py

---

Usage

The DCNv2 should be installed correctly by running:
mask.sh in ./libs/DCNv2_latest/
For evaluating the results of each model, you can run the corresponding "validate.py".
Also you can run the sequence_test.py for testing your own video sequences.

Citing

If you find this code useful for your research, please consider citing the following paper:

@inproceedings{zhou2021rta, 
  title={Revisiting Temporal Alignment for Video Restoration},
  author={Kun Zhou and Wenbo Li and Liying Lu and Xiaoguang Han and Jiangbo Lu}, 
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition} 
  year={2022} 
} 

---

License

Our code is for research purposes only.