- Notifications
You must be signed in to change notification settings - Fork814
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
tensorlayer/SRGAN
Folders and files
| Name | Name | Last commit message | Last commit date | |
|---|---|---|---|---|
Repository files navigation
Implementation of"Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network"
For earlier version, please checksrgan release andtensorlayer.
For more computer vision applications, checkTLXCV
- You need to download the pretrained VGG19 model weights inhere.
- You need to have the high resolution images for training.
- In this experiment, I used images fromDIV2K - bicubic downscaling x4 competition, so the hyper-paremeters in
config.py(like number of epochs) are seleted basic on that dataset, if you change a larger dataset you can reduce the number of epochs. - If you dont want to use DIV2K dataset, you can also useYahoo MirFlickr25k, just simply download it using
train_hr_imgs = tl.files.load_flickr25k_dataset(tag=None)inmain.py. - If you want to use your own images, you can set the path to your image folder via
config.TRAIN.hr_img_pathinconfig.py.
🔥🔥🔥🔥🔥🔥 You need installTensorLayerX at first!
🔥🔥🔥🔥🔥🔥 Please install TensorLayerX via source
pip install git+https://github.com/tensorlayer/tensorlayerx.git
- Set your image folder in
config.py, if you downloadDIV2K - bicubic downscaling x4 competition dataset, you don't need to change it. - Other links for DIV2K, in case you can't find it :test_LR_bicubic_X4,train_HR,train_LR_bicubic_X4,valid_HR,valid_LR_bicubic_X4.
config.TRAIN.img_path="your_image_folder/"
Your directory structure should look like this:
srgan/ └── config.py └── srgan.py └── train.py └── vgg.py └── model └── vgg19.npy └── DIV2K └── DIV2K_train_HR ├── DIV2K_train_LR_bicubic ├── DIV2K_valid_HR └── DIV2K_valid_LR_bicubic- Start training.
python train.py
🔥Modify a line of code intrain.py, easily switch to any framework!
importosos.environ['TL_BACKEND']='tensorflow'# os.environ['TL_BACKEND'] = 'mindspore'# os.environ['TL_BACKEND'] = 'paddle'# os.environ['TL_BACKEND'] = 'pytorch'
🚧 We will support PyTorch as Backend soon.
🔥 We have trained SRGAN on DIV2K dataset.🔥 Download model weights as follows.
| SRGAN_g | SRGAN_d | |
|---|---|---|
| TensorFlow | Baidu,Googledrive | Baidu,Googledrive |
| PaddlePaddle | Baidu,Googledrive | Baidu,Googledrive |
| MindSpore | 🚧Coming soon! | 🚧Coming soon! |
| PyTorch | 🚧Coming soon! | 🚧Coming soon! |
Download weights file and put weights under the folder srgan/models/.
Your directory structure should look like this:
srgan/ └── config.py └── srgan.py └── train.py └── vgg.py └── model └── vgg19.npy └── DIV2K ├── DIV2K_train_HR ├── DIV2K_train_LR_bicubic ├── DIV2K_valid_HR └── DIV2K_valid_LR_bicubic └── models ├── g.npz # You should rename the weigths file. └── d.npz # If you set os.environ['TL_BACKEND'] = 'tensorflow',you should rename srgan-g-tensorflow.npz to g.npz .- Start evaluation.
python train.py --mode=eval
Results will be saved under the folder srgan/samples/.
- [1]Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
- [2]Is the deconvolution layer the same as a convolutional layer ?
If you find this project useful, we would be grateful if you cite the TensorLayer paper：
@article{tensorlayer2017,author = {Dong, Hao and Supratak, Akara and Mai, Luo and Liu, Fangde and Oehmichen, Axel and Yu, Simiao and Guo, Yike},journal = {ACM Multimedia},title = {{TensorLayer: A Versatile Library for Efficient Deep Learning Development}},url = {http://tensorlayer.org},year = {2017}}@inproceedings{tensorlayer2021, title={TensorLayer 3.0: A Deep Learning Library Compatible With Multiple Backends}, author={Lai, Cheng and Han, Jiarong and Dong, Hao}, booktitle={2021 IEEE International Conference on Multimedia \& Expo Workshops (ICMEW)}, pages={1--3}, year={2021}, organization={IEEE}}- For academic and non-commercial use only.
- For commercial use, please contacttensorlayer@gmail.com.
About
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network