Deep Blind Video Super-resolution(DBVSR)

This repository is an official PyTorch implementation of the paper "Deep Blind Video Super-resolution".
The code is built on Ubuntu 16.04 environment (Python3.6, PyTorch_0.4.1, CUDA8.0, cuDNN5.1) with Tesla V100/1080Ti GPUs.

Dependencies

• ubuntu16.04
• python 3.6(Recommend to use Anaconda)
• pyTorch0.4.1
• numpy
• skimage
• imageio
• matplotlib
• tqdm
• cv2

Get started

Trainset:

We use the REDS dataset to train our models. You can download it from official website

We regroup the REDS training and validation sets same as EDVR do:
trainset: the original training (except 4 clips) and validation sets, total 266 clips
validationset: 000, 011, 015 and 020 clips from the original training set, total 4 clips

Models

All the models(X4) can be downloaded from GoogleDrive.

Quicktest with benchmark

After download our models in paper, place the folder models_in_paper to the path ./DBVSR You can test our super-resolution algorithm with REDS4 dataset. Please organize the testset in testset folder like this:

        |--REDS
           |--test
              |--HR
                  |--000
                      |--00000000.png
                             :
                             :
                      |--00000099.png
                  |--011
                  |--015
                  |--020
              |--LR
                  |--000
                      |--00000000.png
                             :
                             :
                      |--00000099.png
                  |--011
                  |--015
                  |--020

please check the data root of test sets in code ./code/option/template.py, line 9(args.dir_data_test).(for dbvsr)
please check the data root of test sets in code ./code/option/template.py, line 26(args.dir_data_test).(for baseline_lr)
please check the data root of test sets in code ./code/option/template.py, line 43(args.dir_data_test).(for baseline_hr)

Then, run the following commands:

cd code
python main.py --test_only

And generated results can be found in ./experiment/dbvsr_test/results/ for dbvsr results
And generated results can be found in ./experiment/baseline_lr_test/results/ for baseline_lr results
And generated results can be found in ./experiment/baeline_hr_test/results/ for baseline_hr results

• To test other benchmarks, you can modify the option(dir_data_test) of the command above.
• To change the save root, you can modify the option(save) of the command above.

How to train

If you have downloaded the trainset, please make sure that the trainset has been organized as follows:

       |--REDS
           |--train
              |--HR
                  |--001
                      |--00000000.png
                      |--00000001.png
                             :
                             :
                      |--00000099.png
                  |--002
                      :
                      :
                  |--239
              |--LR
                  |--001
                      |--00000000.png
                      |--00000001.png
                             :
                             :
                      |--00000099.png
                  |--002
                      :
                      :
                  |--239

Then, please check the data root of train sets in code ./code/option/template.py, line 6(args.dir_data).(for dbvsr)
please check the data root of train sets in code ./code/option/template.py, line 23(args.dir_data).(for baseline_lr)
please check the data root of train sets in code ./code/option/template.py, line 40(args.dir_data).(for baseline_hr)

The command for training is as follow:

cd code
python main.py

The pretrain_model of pwc-net and fcnet can be found in ./pretrain.