Saikat Dutta, Nisarg A. Shah, Anurag Mittal
Accepted at NTIRE workshop, collocated with CVPR 2021 ArXiv | Paper | Poster
Create a conda environment with Pytorch-1.1, CuPy-6.0, OpenCV, SciPy.
conda create -n myenv
conda activate myenv
conda install pytorch==1.1.0 torchvision==0.3.0 cudatoolkit=10.0 -c pytorch
conda install -c anaconda cupy
conda install -c anaconda scipy
conda install -c conda-forge opencv
We use REDS STSR dataset for training and validation. Get the dataset by registering here.
Unzip the dataset under REDS/ directory.
---REDS/
|---val/
|---val_sharp_bicubic/
|---X4/
|---val_sharp/
Start your training by loading PWCNet (checkpoints/pretrained/network-chairs-things.pytorch) and RSDN (checkpoints/pretrained/RSDN.pth).
python REDS_val.py
If this project helps in your research, please cite our paper:
@inproceedings{dutta2021efficient,
title={Efficient space-time video super resolution using low-resolution flow and mask upsampling},
author={Dutta, Saikat and Shah, Nisarg A and Mittal, Anurag},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={314--323},
year={2021}
}
The following repositories were used to develop this project :
[1] QVI
[2] RSDN
[3] PWCNet
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent