Bidirectional Generative Adversarial Network implementation for Pytorch. This implementation uses the MNIST dataset.
The current implentation proposes a "meta-class" BiGAN, containing three networks (Generator, Encoder and Discriminator). The NetManager class can be used to train the BiGAN (its networks), and to produce logs (tensorboard) / plot results. T-SNE can be used to visualize large latent spaces in 2D.
Using main.py:
# default launch
python3 main.py
# training
python3 main.py --batch-size 64 --epochs 40 --seed 1 --log-interval 10
# loading existing weights
python3 main.py --weights weightsNote: you should create the following folders at the root of the repo: logs and results.