This is the code for our college project (Understanding GANs)
The code for the toy examples can be found here:
Our code for training GANs on the MNIST dataset is a minor modification of pytorch-generative-model-collections by the author znxlwm
We used a DCGAN to train on the LSUN outdoor church (as it was the smallest LSUN dataset), the architecture we implemented is exactly the same as mentioned in this paper Unsupervised representation learning with deep convolutional generative adversarial networks. It also contains the WGAN implementation for the church dataset that uses the DCGAN architecture
We used the code from PyTorch-CycleGAN to train our Cycle GAN on the monet2photo dataset, our modified Cycle GAN is also an almost idententical implementation of this with minor modifications to to implement the fused auto-encoders.
We modified the Cycle GAN to fuse the encoders to achive faster convergence at the cost of quality picture generation. Check the Reports folder to see the results.