Name		Name	Last commit message	Last commit date
Latest commit History 22 Commits
.idea		.idea
README.md		README.md
VAE_CNN_BCEloss.py		VAE_CNN_BCEloss.py
VAE_CNN_Gaussianloss.py		VAE_CNN_Gaussianloss.py
VAE_celeba.py		VAE_celeba.py
VAE_facebook.py		VAE_facebook.py
VAE_fb_modified.py		VAE_fb_modified.py
main.py		main.py
main_new.py		main_new.py
new1.py		new1.py
new2.py		new2.py
simple_main.py		simple_main.py

Repository files navigation

Variational-Autoencoder

Contains code to learn variational autoencoder model on MNIST dataset using pytorch.

L = No. of monte carlo samples for gradient calculation

Gaussian loss is given by

$\Large \frac{1}{N}\sum_{i=1}^{N}\left[\frac{1}{L}\sum_{l=1}^{L}\left{ \frac{1}{2}\sum_{j=1}^{784}\log(\sigma_{ij}^{(l)})^2 + \frac{1}{2}\sum_{j=1}^{784}\left(\frac{x_{ij}-\mu_{ij}^{(l)}} {\sigma_{ij}^{(l)}}\right)^2 \right} \right ] - \frac{1}{N}\sum_{i=1}^{N}\left[ \sum_{j=1}^{J}\frac{1}{2}\left(1+\log(\sigma_j^{\prime(i)})^2-(\mu_j^{\prime(i)})^2 -(\sigma_j^{\prime(i)})^2\right )\right ]$

BCE loss is given by

$\Large \frac{1}{N}\sum_{i=1}^{N}\left[\frac{1}{L}\sum_{l=1}^{L}\left{x_{ij}\log p_{ij}^{(l)} + (1-x_{ij})\log(1-\log p_{ij}^{(l)}) \right} \right ] - \frac{1}{N}\sum_{i=1}^{N}\left[ \sum_{j=1}^{J}\frac{1}{2}\left(1+\log(\sigma_j^{\prime(i)})^2-(\mu_j^{\prime(i)})^2 -(\sigma_j^{\prime(i)})^2\right )\right ]$

About

Variational auto encoder in pytorch

Report repository

Releases

No releases published

Packages

No packages published

Contributors 2

Languages

Python 100.0%