Super Resolution Examples

We run this script under TensorFlow 1.4 and the TensorLayer 1.8.0+.

SRGAN Architecture

TensorFlow Implementation of "Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network"

Results

Prepare Data and Pre-trained VGG

• 1. You need to download the pretrained VGG19 model in here as tutorial_vgg19.py show.
• 2. You need to have the high resolution images for training.
  • In this experiment, I used images from DIV2K - 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 use Yahoo MirFlickr25k, just simply download it using train_hr_imgs = tl.files.load_flickr25k_dataset(tag=None) in main.py.
  • If you want to use your own images, you can set the path to your image folder via config.TRAIN.hr_img_path in config.py.

Run

• Set your image folder in config.py, if you download DIV2K - 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/"

• Start training.

python main.py

• Start evaluation. (pretrained model for DIV2K)

python main.py --mode=evaluate 

Reference

• [1] Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
• [2] Is the deconvolution layer the same as a convolutional layer ?

Author

• zsdonghao

License

• For academic and non-commercial use only.
• For commercial use, please contact [email protected].