Pytorch ALAE

A simple Pytorch implementation of the paper "Adverserial Latent AutoEncoder". The paper: https://arxiv.org/abs/2004.04467
Official implementation https://github.com/podgorskiy/ALAE

Experiments

1: train MLP-ALAE on mnist

The first experiment on the paper was training an MLP version of ALAE on Mnist The training result of 100 epochs: (First row: Real images; second row: reconstructed images; third row randomly generated digits)

reprodue:

python3 train_MlpALAE.py

2: train Style-GAN on face images.

This experiment serves as a baseline to styleAlae Style gan on LFW (64x64 resize) with no tricks (truncation, style-mixing):

reprodue:

python3 train_StyleGan.py --dataset_name LFW

3: train Style-ALE on face images

Results of Style ALAE on FFHQ (64x64 resize): (First row: Real images; second row: reconstructed images; third row randomly generated digits)

reprodue:

python3 train_StyleALAE.py --dataset_name FFHQ

requirements

• imageiofile
• kaggle
• numpy
• opencv-python
• torch
• torchvision
• tqdm

Relevant papers:

• PGGANs: https://arxiv.org/pdf/1710.10196.pdf
• StyleGAN: https://arxiv.org/abs/1812.04948

Credits:

I Consulted and used parts of the codes in

• https://github.com/podgorskiy/ALAE
• https://github.com/rosinality/style-based-gan-pytorch.git
• https://github.com/SiskonEmilia/StyleGAN-PyTorch.git