LandscapeGAN

---

This repository is for the implementation of GAN (adapted DCGAN), which is able to sample Landscape Images from a random noise vector.

Z -> DCGAN -> R-ESRGAN -> OUTPUT_IMAGE

Repo borrows code for upscaling from https://github.com/xinntao/Real-ESRGAN

Documentation

The documentation of the whole procedure from start to end is documented in german language within the notebook LandsGAN.ipynb

Installation

1. Clone project locally

git clone git@github.com:SimonStaehli/LandsGAN.git

2. Download Generator model from Google Drive and add it to the folder ./models here: https://drive.google.com/drive/folders/13-GNHs1Bvjd17odvGciSg5K3eubKNEMd?usp=sharing

3. Install required packages in this order

pip install -r requirements1.txt
pip install -r requirements2.txt

Usage/Examples

Generate images:

Will generate 5 photos with size of 3x128x128 and place it into the folder image_out. Check parameter --help for more options and descriptions.

python runner.py -n 5

Upscale images:

This script will upscale images within given input folder -i by 4 times. Check parameter --help for more options and descriptions.

python upscale.py -i image_out -r True