An application of knowledge distillation in GAN SuperResolution

This code uses an implimentation of the SRGAN network, modified from the code of this repository SRGAN.

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

Downloaded pretrained VGG layer here.

Download both HR and a set of LR images of this datset: DIV2K - bicubic downscaling x4 competition

This code uses the "unknown downscaling x4".

Set your image folder and other options in config.py.

• Train a network without knowledge distillation

python main.py

• Start evaluation.

python main.py --mode=evaluate 

• Train a network with knowledge distillation.

python train_distill.py

• [1] Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network

• BenWare144

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