odegeasslbc / oogan-pytorch

Pytorch implementation of the paper: OOGAN: Disentangling GAN with One-Hot Sampling and Orthogonal Regularization

OOGAN-pytorch

Pytorch implementation of the paper: OOGAN: Disentangling GAN with One-Hot Sampling and Orthogonal Regularization

Model overview:

OOGAN based on vanilla GAN

1. code to define the networks: oogan_models.py, oogan_modules.py
2. to train the model on your data, first prepare your images inside a root folder with subfolders containing your images, then edit the "config.py" for all hyperparameter settings, templetes are provided inside, then run

   python train.py

   the training will print log on terminal, and save the generated models and images during training.
3. to generate images from trained model, first edit the model path in "generate.py", then run:

   python generate.py

Train results

On celebA:

On 3D chair:

On MNIST:

On dSprite:

OOGAN based on styleGAN

This implementation is based on the stylegan implementation from here, please refer to that repo for updated code and usage.

1. code: ./oogan_stylegan/oo_stylegan_train.py, ./oogan_stylegan/oo_stylegan_modules.py
2. to train the model on your data

   python oo_stylegan_train.py /path/to/image_root

   the training will print log on terminal and save the generated models and images during training.

oogan-pytorch's Issues

What is the dist module?

While running the code, I got the error for "from dist import Normal" .
The error is "ModuleNotFoundError: No module named 'dist'". I couldn't find the dist module in anaconda. Please help me to figure it out the dist package issues.