Skip to content

Open Source Image and Video Restoration Toolbox for Super-resolution, Denoise, Deblurring, etc. Currently, it includes EDSR, RCAN, SRResNet, SRGAN, ESRGAN, EDVR, etc. Also support StyleGAN2, DFDNet.

Notifications You must be signed in to change notification settings

gtnbssn/BasicSR

 
 

Repository files navigation

🚀 BasicSR

English | 简体中文GitHub | Gitee码云

google colab logo Google Colab: GitHub Link | Google Drive Link
Ⓜ️ Model Zoo ⏬ Google Drive: Pretrained Models | Reproduced Experiments ⏬ 百度网盘: 预训练模型 | 复现实验
📁 DatasetsGoogle Drive百度网盘 (提取码:basr)
📈 Training curves in wandb
💻 Commands for training and testing
HOWTOs


BasicSR (Basic Super Restoration) is an open source image and video restoration toolbox based on PyTorch, such as super-resolution, denoise, deblurring, JPEG artifacts removal, etc.
(ESRGAN, EDVR, DNI, SFTGAN) (HandyView, HandyFigure, HandyCrawler, HandyWriting)

✨ New Features

  • Nov 29, 2020. Add ESRGAN and DFDNet colab demo.
  • Sep 8, 2020. Add blind face restoration inference codes: DFDNet.
  • Aug 27, 2020. Add StyleGAN2 training and testing codes: StyleGAN2.
More
  • Sep 8, 2020. Add blind face restoration inference codes: DFDNet.
    ECCV20: Blind Face Restoration via Deep Multi-scale Component Dictionaries
    Xiaoming Li, Chaofeng Chen, Shangchen Zhou, Xianhui Lin, Wangmeng Zuo and Lei Zhang
  • Aug 27, 2020. Add StyleGAN2 training and testing codes.
    CVPR20: Analyzing and Improving the Image Quality of StyleGAN
    Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen and Timo Aila
  • Aug 19, 2020. A brand-new BasicSR v1.0.0 online.

⚡ HOWTOs

We provides simple pipelines to train/test/inference models for quick start. These pipelines/commands cannot cover all the cases and more details are in the following sections.

GAN
StyleGAN2 Train Inference
Face Restoration
DFDNet - Inference
Super Resolution
ESRGAN TODO TODO SRGAN TODO TODO
EDSR TODO TODO SRResNet TODO TODO
RCAN TODO TODO
EDVR TODO TODO DUF - TODO
BasicVSR TODO TODO TOF - TODO
Deblurring
DeblurGANv2 - TODO
Denoise
RIDNet - TODO CBDNet - TODO

🔧 Dependencies and Installation

Pip install

pip install basicsr
  • pip installation does not compile cuda extensions.
  • If you want to use cuda extensions, set environment variable BASICSR_JIT=True. Note that every time you run the model, it will compile the extensions just time.
    • Example: StyleGAN2 inference colab.

Git clone and compile

  1. Clone repo

    git clone https://github.com/xinntao/BasicSR.git
  2. Install dependent packages

    cd BasicSR
    pip install -r requirements.txt
  3. Install BasicSR

    Please run the following commands in the BasicSR root path to install BasicSR:
    (Make sure that your GCC version: gcc >= 5)
    If you do need the cuda extensions:
    dcn for EDVR
    upfirdn2d and fused_act for StyleGAN2
    please add --cuda_ext when installing.
    If you use the EDVR and StyleGAN2 model, the above cuda extensions are necessary.

    python setup.py develop --cuda_ext

    Otherwise, install without compiling cuda extensions

    python setup.py develop

    You may also want to specify the CUDA paths:

    CUDA_HOME=/usr/local/cuda \
    CUDNN_INCLUDE_DIR=/usr/local/cuda \
    CUDNN_LIB_DIR=/usr/local/cuda \
    python setup.py develop

Note that BasicSR is only tested in Ubuntu, and may be not suitable for Windows. You may try Windows WSL with CUDA supports :-) (It is now only available for insider build with Fast ring).

⏳ TODO List

Please see project boards.

🐢 Dataset Preparation

  • Please refer to DatasetPreparation.md for more details.
  • The descriptions of currently supported datasets (torch.utils.data.Dataset classes) are in Datasets.md.

💻 Train and Test

  • Training and testing commands: Please see TrainTest.md for the basic usage.
  • Options/Configs: Please refer to Config.md.
  • Logging: Please refer to Logging.md.

🏰 Model Zoo and Baselines

  • The descriptions of currently supported models are in Models.md.
  • Pre-trained models and log examples are available in ModelZoo.md.
  • We also provide training curves in wandb:

📝 Codebase Designs and Conventions

Please see DesignConvention.md for the designs and conventions of the BasicSR codebase.
The figure below shows the overall framework. More descriptions for each component:
Datasets.md | Models.md | Config.md | Logging.md

overall_structure

📜 License and Acknowledgement

This project is released under the Apache 2.0 license.
More details about license and acknowledgement are in LICENSE.

🌏 Citations

If BasicSR helps your research or work, please consider citing BasicSR.
The following is a BibTeX reference. The BibTeX entry requires the url LaTeX package.

@misc{wang2020basicsr,
  author =       {Xintao Wang and Ke Yu and Kelvin C.K. Chan and
                  Chao Dong and Chen Change Loy},
  title =        {BasicSR},
  howpublished = {\url{https://github.com/xinntao/BasicSR}},
  year =         {2020}
}

Xintao Wang, Ke Yu, Kelvin C.K. Chan, Chao Dong and Chen Change Loy. BasicSR. https://github.com/xinntao/BasicSR, 2020.

📧 Contact

If you have any question, please email [email protected].

About

Open Source Image and Video Restoration Toolbox for Super-resolution, Denoise, Deblurring, etc. Currently, it includes EDSR, RCAN, SRResNet, SRGAN, ESRGAN, EDVR, etc. Also support StyleGAN2, DFDNet.

Resources

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Python 83.1%
  • Cuda 9.5%
  • C++ 6.4%
  • Other 1.0%