Home

Awesome

SRGAN-PyTorch

Overview

This repository contains an op-for-op PyTorch reimplementation of Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network.

Table of contents

Download weights

Download all available model weights.

# Download `SRGAN_x4-SRGAN_ImageNet.pth.tar` weights to `./results/pretrained_models`
$ bash ./scripts/download_weights.sh SRGAN_x4-SRGAN_ImageNet
# Download `SRResNet_x4-SRGAN_ImageNet.pth.tar` weights to `./results/pretrained_models`
$ bash ./scripts/download_weights.sh SRResNet_x4-SRGAN_ImageNet
# Download `DiscriminatorForVGG_x4-SRGAN_ImageNet.pth.tar` weights to `./results/pretrained_models`
$ bash ./scripts/download_weights.sh DiscriminatorForVGG_x4-SRGAN_ImageNet

Download datasets

These train images are randomly selected from the verification part of the ImageNet2012 classification dataset.

$ bash ./scripts/download_datasets.sh SRGAN_ImageNet

It is convenient to download some commonly used test data sets here.

$ bash ./scripts/download_datasets.sh Set5

How Test and Train

Both training and testing only need to modify yaml file.

Set5 is used as the test benchmark in the project, and you can modify it by yourself.

If you need to test the effect of the model, download the test dataset.

$ bash ./scripts/download_datasets.sh Set5

Test srgan_x4

$ python3 test.py --config_path ./configs/test/SRGAN_x4-SRGAN_ImageNet-Set5.yaml

Test srresnet_x4

$ python3 test.py --config_path ./configs/test/SRResNet_x4-SRGAN_ImageNet-Set5.yaml

Train srresnet_x4

First, the dataset image is split into several small images to reduce IO and keep the batch image size uniform.

$ python3 ./scripts/split_images.py

Then, run the following commands to train the model

$ python3 train_net.py --config_path ./configs/train/SRResNet_x4-SRGAN_ImageNet.yaml

Resume train srresnet_x4

Modify the ./configs/train/SRResNet_x4-SRGAN_ImageNet.yaml file.

$ python3 train_net.py --config_path ./configs/train/SRResNet_x4-SRGAN_ImageNet.yaml

Train srgan_x4

$ python3 train_gan.py --config_path ./configs/train/SRGAN_x4-SRGAN_ImageNet.yaml

Resume train srgan_x4

Modify the ./configs/train/SRGAN_x4-SRGAN_ImageNet.yaml file.

$ python3 train_gan.py --config_path ./configs/train/SRGAN_x4-SRGAN_ImageNet.yaml

Result

Source of original paper results: https://arxiv.org/pdf/1609.04802v5.pdf

In the following table, the psnr value in () indicates the result of the project, and - indicates no test.

Set5ScaleSRResNetSRGAN
PSNR432.05(32.16)29.40(30.67)
SSIM40.9019(0.8938)0.8472(0.8627)
Set14ScaleSRResNetSRGAN
PSNR428.49(28.57)26.02(27.12)
SSIM40.8184(0.7815)0.7397(0.7321)
BSD100ScaleSRResNetSRGAN
PSNR427.58(27.56)25.16(26.22)
SSIM40.7620(0.7367)0.6688(0.6867)
# If you do not train the model yourself, you can download the model weights and test them.
$ bash ./scripts/download_weights.sh SRGAN_x4-SRGAN_ImageNet
$ python3 ./inference.py

Input:

<span align="center"><img width="240" height="360" src="figure/comic.png"/></span>

Output:

<span align="center"><img width="240" height="360" src="figure/sr_comic.png"/></span>

Build `srresnet_x4` model successfully.
Load `srresnet_x4` model weights `SRGAN-PyTorch/results/pretrained_models/SRGAN_x4-SRGAN_ImageNet.pth.tar` successfully.
SR image save to `./figure/sr_comic.png`

Contributing

If you find a bug, create a GitHub issue, or even better, submit a pull request. Similarly, if you have questions, simply post them as GitHub issues.

I look forward to seeing what the community does with these models!

Credit

Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network

Christian Ledig, Lucas Theis, Ferenc Huszar, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, Wenzhe Shi <br>

Abstract <br> Despite the breakthroughs in accuracy and speed of single image super-resolution using faster and deeper convolutional neural networks, one central problem remains largely unsolved: how do we recover the finer texture details when we super-resolve at large upscaling factors? The behavior of optimization-based super-resolution methods is principally driven by the choice of the objective function. Recent work has largely focused on minimizing the mean squared reconstruction error. The resulting estimates have high peak signal-to-noise ratios, but they are often lacking high-frequency details and are perceptually unsatisfying in the sense that they fail to match the fidelity expected at the higher resolution. In this paper, we present SRGAN, a generative adversarial network (GAN) for image super-resolution (SR). To our knowledge, it is the first framework capable of inferring photo-realistic natural images for 4x upscaling factors. To achieve this, we propose a perceptual loss function which consists of an adversarial loss and a content loss. The adversarial loss pushes our solution to the natural image manifold using a discriminator network that is trained to differentiate between the super-resolved images and original photo-realistic images. In addition, we use a content loss motivated by perceptual similarity instead of similarity in pixel space. Our deep residual network is able to recover photo-realistic textures from heavily downsampled images on public benchmarks. An extensive mean-opinion-score (MOS) test shows hugely significant gains in perceptual quality using SRGAN. The MOS scores obtained with SRGAN are closer to those of the original high-resolution images than to those obtained with any state-of-the-art method.

[Paper]

@InProceedings{srgan,
    author = {Christian Ledig, Lucas Theis, Ferenc Huszar, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, Wenzhe Shi},
    title = {Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network},
    booktitle = {arXiv},
    year = {2016}
}