kozistr / Awesome-Super-Resolution

lots of single image super resolution model implementations in tensorflow

Awesome Super Resolution

Lots of SISR (Single Image Super Resolution) implementations in tensorflow maybe w/ pre-trained model!

maybe later, this repo could be supported via pip package.

Currently, Work-In-Progress

Usage

PIP

[not available yet :(]

Download

$ git clone https://github.com/kozistr/Awesome-Super-Resolution
$ cd ./Awesome-Super-Resolution

Dependency Install

$ pip3 install -r ./requirements.txt

Train / Eval / Inference

$ python3 train.py [w/ some parameters]
$ python3 eval.py [w/ some parameters]
$ python3 inference.py [w/ some parameters]

DataSets

• DIV2K
• Flicker2K
• Set*

Repo Tree

│
├── assets (dir, images used in readme.md)
├── models
│    ├── vgg16.py (VGG19 model loader)
│    ├── vgg19.py (VGG16 model loader)
│    ├── xxx (dir, model name)
│    │     ├── logs      (tensorboard logs)
│    │     ├── config.py (configurations)
│    │     ├── model.py  (model script)
│    │     └── readme.md (results & explains)
│    └── ... (dir, model name)
│          └── ...
├── train.py       (trainer)
├── eval.py        (evaluator)
├── inference.py   (inferencer)
├── ops.py         (useful tf operators)
├── utils.py       (useful image utilities)
├── metrics.py     (metrics for evaluating SR Model)
├── dataloader.py  (dataset loader / feeder)
└── readme.py      (readme)

Papers & Codes

Name	Summary	Paper	Code
2015
SRCNN	Image Super-Resolution Using Deep Convolutional Networks	[arXiv]	[code]
2016
SRGAN	Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network	[arXiv]	[code]
FSRGAN	Accelerating the Super-Resolution Convolutional Neural Network	[arXiv]	[code]
EnhanceNet	Single Image Super-Resolution Through Automated Texture Synthesis	[arXiv]	[code]
2017
LapSRN	Deep Laplacian Pyramid Networks for Fast and Accurate Super-Resolution	[arXiv]	[code]
EDSR	Enhanced Deep Residual Networks for Single Image Super-Resolution	[arXiv]	[code]
2018
RCAN	Image Super-Resolution Using Very Deep Residual Channel Attention Networks	[arXiv]	[code]
ESRGAN	Enhanced SRGAN. Champion PIRM Challenge on Perceptual Super-Resolution	[arXiv]	[code]
FEQE	Fast and Efficient Image Quality Enhancement via Desubpixel Convolutional Neural Networks	[ECCV]	[code]
IDN	Fast and Accurate Single Image Super-Resolution via Information Distillation Network	[ECCV]	[code]
2019
NNTSR	Image Super-Resolution by Neural Texture Transfer	[arXiv]	[code]

Pre-Trained Models

It's on the plan, but, because of the lack of hardware resources, it can be.

To-Be-Done

0. TBD

ETC

Any suggestions and PRs and issues are WELCOME :)

Author

HyeongChan Kim / @kozistr