LandscapeGAN

This repository contains the code for generating landscape imagery.

The code makes use of two types of neural networks for generating the landscape images:

1. A GAN, which generates 192x108 images, training on the images provided in the sampleimages_lowres directory
2. An SRCNN, which takes 768x432 generated images (after being upscaled from 192x108), and generates higher resolution versions of these images

Description of Scripts

At the time of writing, 6 main python scripts are used for generating the landscape images:

1. train_imagegan.py
   • used for training the GAN, and saving the trained GAN model.
2. train_srcnn.py
   • used for training the SRCNN, and saving the trained SRCNN model.
3. image_downscale.py
   • used for downscaling the images from the sampleimages_highres directory into suitable resolutions for the sampleimages_midres (768x432) and sampleimages_lowres (192x108) directories, which is where the resulting images are stored.
   • this is done when if want to add images for the neural networks to train on (must be 1920x1080 and must be placed into sampleimages_highres).
4. image_upscale.py
   • used for upscaling the 192x108 images generated by the GAN in the generatedimages directory, which are upscaled to 768x432 and stored in the generatedimages_upscaled directory.
5. generate_images.py
   • used for generating a new set of landscape images, using the saved GAN model. The resulting images are stored in the generatedimages directory.
6. srcnn_images.py
   • used for improving the resolution of the generated landscape images, using the saved SRCNN model. The resulting images are stored in the generatedimages_srcnn directory.