this is the fork of bicycleGAN which is used to train and run GANs for frankenGAN. the interactive system which uses these networks is chordatlas.
requirements:
- nvidia GPU (CUDA 8+)
- pytorch 1.4
- visdom
- dominate
the entry point is test_interactive.py which listens to the ./input folders for new inputs, and writes them to ./output (these folders should exist). it will download the pre-trained model weights the first time your run it. Once it is running, set chordatlas's bikeGAN file location (in the settings menu) to the bikeGAN root directory (the one containing this file).
alternatively, use the docker container with nvidia-docker:
nvidia-docker run -v $(pwd)/input:/home/user/bikegan/input -v $(pwd)/output:/home/user/bikegan/output -it --rm twak/bikegan
if you use this project, please cite frankenGAN
@misc{frankenGAN,
Author = {Tom Kelly and Paul Guerrero and Anthony Steed and Peter Wonka and Niloy J. Mitra},
Title = {FrankenGAN: Guided Detail Synthesis for Building Mass-Models Using Style-Synchonized GANs},
Year = {2018},
Eprint = {arXiv:1806.07179},
}