This is a modified version of Yijunmaverick's implementation of CartoonGAN [Chen et al., CVPR18]. This repo was created due to lack of activity in original repo.

This model is normally used for poster POCs and Blender3D->2D background tests and pre-processing, more like a Photoshop filter for visualising characters with the background before post-processing with light rays, additional shadows, among others.

You can check out the demo here:

Contribute to the demo repository here: https://github.com/venture-anime/anime-background-gan-hf-space

• Unix-based OS
• python 3.9.9 (other 3.x versions may work as well)
  • NOTE: This also works with M1 Macbooks
• wget (brew install wget) (if you don't have it)
• (optional, will default to CPU if no GPU/CUDA detected) NVIDIA GPU

1. pip install -r requirements.txt
2. Download the pretrained models with sh pretrained_model/download_pth.sh

python main.py --input_dir test_img --style Hosoda

Create a pull request with your changes and assign it to Sho. Make sure to tag your commits using the numpy commit guidelines for easier review.