baudm / vaegan-celebs-keras

Autoencoding beyond pixels using a learned similarity metric by Larsen et al. (https://arxiv.org/abs/1512.09300)

vaegan-celebs-keras

EE 298 Group 6:

Darwin Bautista | Paolo Valdez

Hello!

This is our implementation of "Autoencoding beyond pixels using a learned similarity metric" by Larsen et al. (https://arxiv.org/abs/1512.09300) Many of the implementation details (which were unclear in the paper) were directly lifted from the authors' official code

Prequisites

Tensorflow (>=1.4)

Keras (>= 2.1.4)

OpenCV (>= 3.4.0)

Numpy

Dataset

Available @ http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html

Presenstation Slides

Our Presentation Material can be found at: https://docs.google.com/presentation/d/1PhjrLkPf-UstSI_oZXod0qb4k_HsPbtNlm9ekJE4gvY/edit?usp=sharing

Pre-trained weights:

https://drive.google.com/open?id=1ELiB3GNeT_I9RyTqpeoNNQ2CoOZ-v81X

Data sample:

Variational Auto Encoder (VAE) Architecture

Encoder Training Model

Decoder Training Model

General Adverserial Network (GAN) Architecture

Discriminator Training Model

Generated Image Results: