README.md

StyleGAN for Landscape Photographs

⭐Generate your own landscape photos here!⭐

This repository contains codes which:

• re-implement StyleGAN in PyTorch;
• progressively train StyleGAN on landscape photos;
• classify photos in the disentangled latent space;
• display the results with Dash and deploy the model on Heroku platform.

Details

The StyleGAN models can be found in models directory. Implentmented tricks and features:

• Progressive Training
• PixelNorm Layers
• Truncation Layers
• Style Mixing
• Loss with R1 Penalty
• Gradient Clipping

The models are then trained using this notebook. After training, generated photos are classified by their time(day/night/dawn&dust) and weather(aurora or not) in the disentangled latent space using small neuron network in this notebook.

The codes for the web app can be found in the StyleGAN-Webpage directory.

References

• The codes rely heavily on NVIDIA's StyleGAN TF implementation:
  Karras, Tero, Samuli Laine, and Timo Aila. "A Style-Based Generator Architecture for Generative Adversarial Networks." arXiv preprint arXiv:1812.04948 (2018). https://arxiv.org/abs/1812.04948
  Official StyleGAN implementation: https://github.com/NVlabs/stylegan

• The implementation of Equalized Learning Rate layers refers to the following repository:
  https://github.com/huangzh13/StyleGAN.pytorch

• The web pages are inspired by the following tutorial:
  Dash tutorial (Chinese): https://blog.csdn.net/lly1122334/article/details/107056777