🚀 BasicSR

GitHub | Gitee码云
English | 简体中文

BasicSR is an open source image and video super-resolution toolbox based on PyTorch (will extend to more restoration tasks in the future).
_{(ESRGAN, EDVR, DNI, SFTGAN)}

Dependencies and Installation

• Python >= 3.7 (Recommend to use Anaconda or Miniconda)
• PyTorch >= 1.3
• NVIDIA GPU + CUDA

Please run the following commands in the BasicSR root path to install BasicSR:
(Make sure that your GCC version: gcc >= 5)

pip install -r requirements.txt
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).

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.

• How to train StyleGAN2
• How to test StyleGAN2

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

Download official pre-trained models
Download reproduced models and logs

• 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

License and Acknowledgement

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

Contact

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