Skip to content
/ FUIR Public

A Flexible and Unified Image Restoration Framework (PyTorch), including state-of-the-art image restoration model. Such as NAFNet, Restormer, MPRNet, MIMO-UNet, SCUNet, SwinIR, HINet, etc. ⭐⭐⭐⭐⭐⭐

License

Notifications You must be signed in to change notification settings

murufeng/FUIR

Repository files navigation

PWC PWC PWC PWC PWC PWC PWC PWC PWC PWC PWC PWC PWC PWC PWC PWC PWC

A Flexible and Unified Image Restoration Framework.

GitHub stars GitHub forks download visitors

当前,对于Low-level-Vision中的图像恢复任务(Image Super-resolution, Image Denoising, Image Deblurring等) 已经出现了很多优秀实用的工具包。 但是,对于去年出现的一系列Transformer网络架构还没有一个统一的项目将它集成起来,本项目主要在具体图像恢复任务的数据处理和网络训练配置上将当前主流的CNN网络和基于Transformer的网络架构集成了起来。本项目将作为图像恢复任务的一个灵活统一的工具。利用本项目可以快速实现图像去噪,图像/视频去模糊,图像去雨等一系列经典任务的训练。 本项目提供在GOPRO、SIDD、REDS、Rain13K数据集上的数据预处理以及网络架构训练教程,后续我们将持续更新。希望本项目既能让图像处理初学者快速入门,又能服务科研和工业社区。

Installation

See INSTALL.md for the installation of dependencies required to run FUIR.

News

  • 2022-5-17 Update the State-of-the-Art models, such as NAFNet, Restormer, MPRNet, HINet, MIMO-UNet

Model Zoo

NAFNet

网络模块结构如下所示:

模块设计如下:

两大经典图像恢复实验结果如下:

  1. Image Denoising (在SIDD数据集实现SOTA):

  1. Image Deblurring (在GoPro数据集实现SOTA):

模型训练配置

模型训练代码

Restormer

网络模块结构如下所示:

三大图像恢复实验结果对比图:

  1. Image Deraining(目前保持SOTA):

  1. Image Deblurring:

  1. Image Denoising:

模型训练配置

模型训练代码

HINet

网络模块结构如下所示:

三大图像恢复实验结果:

  1. Image Denoising:

  1. Image Deblurring:

  1. Image Deraining:

模型训练配置

模型训练代码

MPRNet

网络模块结构如下所示:

三大图像恢复实验结果:

  1. Image Deraining:
  2. Image Deblurring:
  3. Image Denoising:

模型训练配置

模型训练代码

MIMONet

网络模块结构如下所示:

图像去模糊实验结果与可视化图:

MIRNet

网络模块结构如下所示:

SCUNet

网络模块结构如下所示:

部分实验结果如下:

Experiments

Image Restoration Tasks

Task Dataset Train/Test Instructions
Image Deblurring GoPro link
Image Denoising SIDD link
Image Deblurring with JPEG artifacts REDS link
Image Derain Rain13K link
Stereo Image Super-Resolution Flickr1024+Middlebury link

Citations

If you find this project is useful, please give me a star and fork!

Contact

If you have any questions, please contact me. [email protected]


Acknowledgment

This implementation based on BasicSR and NAFNet.

statistics

visitors

About

A Flexible and Unified Image Restoration Framework (PyTorch), including state-of-the-art image restoration model. Such as NAFNet, Restormer, MPRNet, MIMO-UNet, SCUNet, SwinIR, HINet, etc. ⭐⭐⭐⭐⭐⭐

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages