pytorch 0.4 version
BasicSR wants to provide some basic deep-learning based models for super-resolution, including:
- PSNR-oriented SR models (e.g., SRCNN, VDSR, SRResNet and etc)
- want to compare more structures for SR. e.g. ResNet Block, ResNeXt Block, Dense Block, Residual Dense Block, Poly Block, Dual Path Block, Squeeze-and-Excitation Block and etc.
- want to provide some useful tricks for training SR networks.
- GAN-based models for more visual-pleasing performance, especially textures.
The codes will be explained in each folder with README.md and the exploration will be put in Wiki.
Testing and Training can be found in codes/README.md.
🌞
-
It now supports a framework to train and test PSNR-oriented SR models. And we will gradually train and compare other models and try other techniques.
-
For SRGAN, we reproduce the results using DIV2K dataset (w/o BatchNorm in the generator).
- For SFTGAN, we provide the training and testing codes.
Welcome to report bugs 😜 and welcome to contribute to this repo 😜 . I am not expert at coding, but I will try to keep the codes tidy.
- Linux
- Python3
- Pytorch 0.4
- NVIDIA GPU + CUDA
There are some classical SR datasets, for example:
- training datasets: BSD200, T91, General100;
- testing datasets: Set5, Set14, urban100, BSD100, manga109, historical
You can download these classical SR datasets from Google Drive or Baidu Drive.
Currently, there is a new DIVerse 2K resolution high quality images for SR called DIV2K, which can be downloaded from DIV2K offical page, or from Baidu Drive.
Pretrained models can be downloaded from Google Drive. You can put them in experiments/pretrained_models folder.
More details about the pretrained models, please see experiments/pretrained_models.
- Code architecture is inspired from pytorch-cyclegan.
- Thanks to Wai Ho Kwok, who develops the initial version.