Image-Super-Resolution

This repository gathers the code for image super resolution from the in-class CodaLab competition.

We use SwinIR, an image restoration toolbox (PyTorch) that provides training and testing codes for SwinIR, to train our model.

Reproducing Submission

We need to do some pre-preparation for training and testing on our custom dataset.

To reproduce my submission without retrainig, do the following steps:

1. Requirement
2. Repository Structure
3. Inference

Hardware

Ubuntu 18.04.5 LTS

Intel® Core™ i7-3770 CPU @ 3.40GHz × 8

GeForce GTX 1080/PCIe/SSE2

Requirement

All requirements should be:

$ virtualenv SwinIR --python=3.6
$ source ./SwinIR/bin/activate
$ cd Image-Super-Resolution
$ pip install -r requirements.txt

Official images can be downloaded from CodaLab competition

Repository Structure

The repository structure is:

Image-Super-Resolution(root)
  +-- data                   
  +-- models
  +-- utils
  +-- model_zoo                            # put model weight(.pth) here
  +-- options                              # training hyper-parameters setting
  |   +-- train_swinir_sr_classical.json  
  +-- testing_lr_images                    # testing data
  +-- training_hr_images                   # training data
  +-- inference.py
  +-- train.py
  +-- requirements.txt

Training

To train the model, run this command:

$ python train.py --opt options/train_swinir_sr_classical.json

Trained model will be saved in superresolution/swinir_sr_classical_patch48_x3/models

Inference

Please download this model if you want to reproduce my submission file, put it in model_zoo and run the following code.

To reproduce my submission file or test the model you trained, run:

$ python inference.py --task classical_sr --scale 3  --model_path model_zoo/model_final.pth --folder_lq testing_lr_images/testing_lr_images

Prediction file will be saved in results/swinir_classical_sr_x3

If you use different hardware, the inference result may be a little different.

Results

Our model achieves PSNR 28.3899dB

Reference

[1] SwinIR