Lornatang/SRGAN-PyTorch Lornatang/SRGAN-PyTorch

  • Stars
    star
    407
  • Rank 106,183 (Top 3 %)
  • Language
    Python
  • License
    Apache License 2.0
  • Created over 4 years ago
  • Updated 5 months ago
Twitter logo Share LinkedIn logo Share Facebook logo Share
Lornatang/SRGAN-PyTorch
github

Lornatang

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

A simple and complete implementation of super-resolution paper.

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.

  • line 33: RESUMED_G_MODEL change to ./samples/SRResNet_x4-SRGAN_ImageNet/g_epoch_xxx.pth.tar.
$ 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.

  • line 38: PRETRAINED_G_MODEL change to ./results/SRResNet_x4-SRGAN_ImageNet/g_last.pth.tar.
  • line 40: RESUMED_G_MODEL change to ./samples/SRGAN_x4-SRGAN_ImageNet/g_epoch_xxx.pth.tar.
  • line 41: RESUMED_D_MODEL change to ./samples/SRGAN_x4-SRGAN_ImageNet/d_epoch_xxx.pth.tar.
$ 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.

Set5 Scale SRResNet SRGAN
PSNR 4 32.05(32.16) 29.40(30.67)
SSIM 4 0.9019(0.8938) 0.8472(0.8627)
Set14 Scale SRResNet SRGAN
PSNR 4 28.49(28.57) 26.02(27.12)
SSIM 4 0.8184(0.7815) 0.7397(0.7321)
BSD100 Scale SRResNet SRGAN
PSNR 4 27.58(27.56) 25.16(26.22)
SSIM 4 0.7620(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:

Output:

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

Abstract
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}
}

More Repositories

1

CycleGAN-PyTorch

A very simple implementation of cyclegan, which is based on pytorch.
Python
147
star
2

ESRGAN-PyTorch

A simple implementation of esrgan, which uses the pytorch framework.
Python
136
star
3

SRCNN-PyTorch

Pytorch framework can easily implement srcnn algorithm with excellent performance
Python
91
star
4

machine_learning_in_action_py3

Important book about the machine learning algorithms, and introduces the application of those who use these algorithms and tools, and how to use them in a real environment. This book and other books, behind the other books are long on machine learning theory knowledge, the book happened to be more discussion on how to use coded machine learning algorithms.
Python
81
star
5

AlexNet-PyTorch

An PyTorch implementation AlexNet.Simple, easy to use and efficient
Python
79
star
6

ESPCN-PyTorch

A PyTorch implementation of ESPCN based on CVPR 2016 paper Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network.
Python
69
star
7

RFB_ESRGAN-PyTorch

Simple realization of papers in oppo Research Institute super score competition.
Python
65
star
8

FSRCNN-PyTorch

Fast implementation of fsrcnn algorithm based on pytorch framework
Python
58
star
9

VDSR-PyTorch

PyTorch implements "Accurate Image Super-Resolution Using Very Deep Convolutional Networks"
Python
52
star
10

GoogLeNet-PyTorch

Restore the official code 100% and improve it to make it easier to use.
Python
48
star
11

WassersteinGAN_GP-PyTorch

Improved training of Wasserstein GANs
Python
46
star
12

CGAN-PyTorch

Simple implementation of conditional general adverse nets in pytorch machine learning framework
Python
44
star
13

YOLOv3-PyTorch

Pytorch implements yolov3.Good performance, easy to use, fast speed.
Python
32
star
14

Real_ESRGAN-PyTorch

PyTorch implements `Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data` paper.
Python
31
star
15

ResNet-PyTorch-backup

Fully reproduce the pairs of ImageNet and CIFAR in the paper
Python
30
star
16

ResNet-PyTorch

PyTorch implements `Deep Residual Learning for Image Recognition` paper.
Python
29
star
17

VGG-PyTorch

The implementation of VGG thesis is implemented under PyTorch framework
Python
29
star
18

UGATIT-PyTorch

Simple, fast and easy to read. Yes, we use the pytorch framework!
Python
29
star
19

YOLOv4-PyTorch

Pytorch implements yolov4.Good performance, easy to use, fast speed.
Python
27
star
20

GAN-PyTorch

A complete implementation of the Pytorch neural network framework for GAN
Python
24
star
21

PyTorch-WGANGP

PyTorch implements WGAN_GP
Python
21
star
22

pytorch-alexnet-cifar100

Cifar100 in alexnet network model under the highest accuracy
Python
18
star
23

pytorch-tutorial

fork Yunjey Choi/ [@yunjey](https://github.com/yunjey)
Python
18
star
24

WassersteinGAN_DIV-PyTorch

Wasserstein Divergence for GANs
Python
18
star
25

PyTorch

An open source deep learning platform that provides a seamless path from research prototyping to production deployment
Python
17
star
26

PyTorch-NCSN

This is the best GAN method to generate cifar-10
Python
17
star
27

DCGAN-PyTorch

PyTorch implements a Deep Convolution GAN neural network structure
Python
12
star
28

TensorFlow2-tutorials

This is a tutorial from TensorFlow's website, thanks to Google Inc
Python
11
star
29

SwinIR-PyTorch

PyTorch implements `SwinIR: Image Restoration Using Swin Transformer` paper.
Python
11
star
30

RCAN-PyTorch

PyTorch implements `Image Super-Resolution Using Very Deep Residual Channel Attention Networks` paper.
Python
11
star
31

MetaSR-PyTorch

PyTorch implements `Meta-SR: A Magnification-Arbitrary Network for Super-Resolution` paper
Python
10
star
32

pytorch-inception-v3-cifar100

Pytorch implements the inceptionv4 model for cifar100 classification
Python
10
star
33

pytorch-vgg19-cifar100

Pytorch implements the VGG19 model to classify cifar100
Python
9
star
34

TensorFlow-MNIST

MNIST, KMNIST, EMNIST and others have TensorFlow implementation of handwritten data set, based on TF2 version
Python
8
star
35

Deep-learning-paper-CN

Deep learning thesis Chinese translation, including object classification, detection.
8
star
36

PyTorch-pix2pixHD

PyTorch implements high resolution image generation
Python
8
star
37

tf-gans

Tensorflow implements the most primitive GAN
Python
8
star
38

CRNN-PyTorch

PyTorch implemnts `An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition` paper.
Python
8
star
39

MobileNetV1-PyTorch

PyTorch implements `MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications` paper.
Python
8
star
40

InceptionV4-PyTorch

PyTorch implements `Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning` paper.
Python
8
star
41

Deep-learning-with-cartoon

利用对抗性深度学习来生成动漫妹子头像(Use antagonistic deep learning to generate animated girl profile pictures)
Python
7
star
42

SRDenseNet-PyTorch

PyTorch implements `Image Super-Resolution Using Dense Skip Connections` paper.
Python
7
star
43

PyTorch-StackGAN

PyTorch implements StackGAN
Python
7
star
44

DenseNet-PyTorch

Restore the official code 100% and improve it to make it easier to research and facilitate production.
Python
7
star
45

WassersteinGAN-PyTorch

This is a new alogorithm named WGAN, an alternative to traditional GAN training!
Python
7
star
46

DayHR

Try to implement image style conversion, based on the TF2 version, in order to convert avatars to cartoon avatars.
Python
6
star
47

EDSR-PyTorch

PyTorch implements "Enhanced Deep Residual Networks for Single Image Super-Resolution"
Python
6
star
48

DRRN-PyTorch

PyTorch implements "Image Super-Resolution via Deep Recursive Residual Network"
Python
6
star
49

OpenCV4-Tutorials

This tutorial is based on OpenCV's 4.1.0-dev version.
C++
6
star
50

pytorch-project

Some pytorch projects self-fork
Python
6
star
51

RealSR-PyTorch

PyTorch implements `Toward Real-World Single Image Super-Resolution: A New Benchmark and A New Model` paper.
Python
6
star
52

TensorFlow2-GAN

Implement all generation antagonism models based on TensorFlow2
Python
5
star
53

TensorFlow-cats_vs_dogs

Version 2.0 of tensorflowa classification method for dogs and cats
Python
5
star
54

CSNLN-PyTorch

PyTorch implemnts `Image Super-Resolution with Cross-Scale Non-Local Attention and Exhaustive Self-Exemplars Mining` paper.
Python
5
star
55

ACGAN-PyTorch

PyTorch implements Auxiliary Classifier GAN
Python
5
star
56

zsb-review

C语言考试复习资料
5
star
57

MNIST-LibTorch

An example of MNIST handwritten number recognition for LibTorch written by pure CPP
C++
5
star
58

tensorflow-vgg

Implementation of VGG based on tensorflow
Python
4
star
59

InceptionV3-PyTorch

PyTorch implements `Rethinking the Inception Architecture for Computer Vision` paper.
Python
4
star
60

TensorFlow-WGAN_GP

An implements for Wasserstein GAN by TensorFlow
Python
4
star
61

PyTorch-DCGAN

PyTorch implementation DCGAN
Python
4
star
62

LapSRN-PyTorch

PyTorch implements `Deep Laplacian Pyramid Networks for Fast and Accurate Super-Resolution` paper.
Python
4
star
63

zero2Adam

Numpy implements Adam
Python
4
star
64

PyTorch-WassersteinGANs

Python
4
star
65

TensorFlow-caltech

Classification of Caltech data sets under version 2.0 of tensorflow.xml
Python
4
star
66

PyTorch-PGGAN

PyTorch implementation PGGAN
Python
4
star
67

CARN-PyTorch

PyTorch implements `Fast, Accurate, and Lightweight Super-Resolution with Cascading Residual Network` paper.
Python
4
star
68

TensorFlow-AE

TensorFlow2.0 combines Keras to achieve automatic uncoding
Python
4
star
69

Deep-learning-with-python3

Book "Deep Learning with Python" python3 code
Python
3
star
70

ContextualBilateralLoss-PyTorch

PyTorch implements `Zoom to Learn, Learn to Zoom` paper.
Python
3
star
71

Learn-OpenCV-4-By-Building-Projects-Second-Edition

Learn OpenCV 4 by Building Projects
C++
3
star
72

JAVA-Project

The Java code for the second semester of the junior year can be found here (I have not been exposed to Java for more than two years, which is required by my studies).
Java
3
star
73

LIIF-PyTorch

PyTorch implements `Learning Continuous Image Representation with Local Implicit Image Function` paper.
Python
3
star
74

zero2SGD

Implement SGD algorithm from zero
Python
3
star
75

ClassifierGAN

This paper mainly classifies the samples generated by GAN model
Python
3
star
76

DBPN-PyTorch

PyTorch implements "Deep Back-Projection Networks for Single Image Super-resolution” paper.
Python
3
star
77

zero2MBGD

Numpy implements MBGD
Python
3
star
78

SSD_PyTorch

Simple implementation of SSD
Python
3
star
79

opencv201907

C++
3
star
80

Algorithm_CPP_language_implementation

Based on the third edition
C++
3
star
81

BSRGAN-PyTorch

PyTorch implements `Designing a Practical Degradation Model for Deep Blind Image Super-Resolution` paper.
Python
3
star
82

LSRGAN-PyTorch

PyTorch implements `Optimizing Generative Adversarial Networks for Image Super Resolution via Latent Space Regularization` paper.
Python
3
star
83

PyTorch-GAN

PyTorch implementations of Generative Adversarial Networks.
Python
3
star
84

EfficientNet

A PyTorch implementation of EfficientNet
Python
3
star
85

ShuffleNetV2-PyTorch

PyTorch implements `ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design` paper.
Python
3
star
86

tf-dcgan

Implementation of TensorFlow for deep convolution adversarial neural networks
Python
3
star
87

PyTorch-VAE

PyTorch simple implements VAE
Python
3
star
88

StyleGAN-PyTorch

The PyTorch framework is a simple implementation of StyleGAN
2
star
89

Fast-Neural-Style-PyTorch

This repository contains a pytorch implementation of an algorithm for fast neural style
Python
2
star
90

FAN-PyTorch

PyTorch implements `Focusing Attention: Towards Accurate Text Recognition in Natural Images` paper.
Python
2
star
91

models

A library of pretrained model weights in managed code
2
star
92

My-girlfriend

I miss you
Python
2
star
93

RDN-PyTorch

PyTorch implements `Residual Dense Network for Image Super-Resolution` paper.
Python
2
star
94

Fuck_WEB

GoodBye ASP.NET
HTML
2
star
95

PyTorch-StyleGAN

PyTorch implements StyleGAN
2
star
96

face_recognition

Open source framework on fork, easy to implement
Python
2
star
97

Video_2_Char

CAI xukun is so handsome in basketball!
Python
2
star
98

zero2BGD

Numpy implements BGD
Python
2
star
99

PyTorch-InfoGAN

PyTorch implementation InfoGAN
Python
2
star
100

ArbSR-PyTorch

PyTorch implements `Learning A Single Network for Scale-Arbitrary Super-Resolution` paper.
Python
2
star