Keras Implementation of SRGAN

Keras and Google Cloud Ready implementation of "Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network"

Credit to https://github.com/MathiasGruber/SRGAN-Keras

1. Architecture

The generator creates a high-resolution (HR) image (4x upscaled) from a corresponding low-resolution (LR) image. The discriminator distinguishes the generated (fake) HR images from the original HR images.

1.1. Generator & Discriminator

Figure 4 from paper: Architecture of Generator and Discriminator Network with corresponding kernel size (k), number of feature maps (n) and stride (s) indicated for each convolutional layer.

1.2. Overview of input / outputs

Code Overview: Overview of the three networks; generator, discriminator, and VGG19. Generator create SR image from LR, discriminator predicts whether it's a SR or original HR, and VGG19 extracts features from generated SR and original HR images.

2. Content & Adversarial Loss

Losses Overview: The perceptual loss is a combination of content loss (based on VGG19 features) and adversarial loss. Equations are taken directly from "original paper".

3. Using this repository

3.1. Training

Local

Create 2 folders train, valid in the trainer/data directory and place your images there.

Use the bash file train.sh. To train run e.g.:

git clone https://github.com/skiler07/GAN.git
cd GAN
sh train.sh local

Remote

See train.sh file for more details and specify remote job_dir.

3.2. Testing

Check the example_usage notebook: example_usage.ipynb

georgy95 / GAN