README.md

PyProbML VAE zoo 🐘

Author: Ming Liang Ang. Summer 2021.

Compare results of different VAEs :

VAE tricks and what the different VAE try to address :

A collection of Variational AutoEncoders (VAEs) implemented in pytorch with focus on reproducibility and creating reusable blocks that can be used in any project. The aim of this project is to provide a quick and simple working example for many of the cool VAE idea in the textbook. All the models are trained on the CelebA dataset for consistency and comparison.

Requirements

• Python >= 3.7

• PyTorch >= 1.8

• Pytorch Lightning == 1.4.0

• CUDA enabled computing device

To download this subdirectory only

Copy the url of the subdirectory and past it to this webstie and then download this subdirectory as a zipfile

Instruction For Training The Model

Download celeba data

Important : Make sure to get your kaggle.json from these instructions then run

mkdir /root/.kaggle 
cp kaggle.json /root/.kaggle/kaggle.json
chmod 600 /root/.kaggle/kaggle.json
rm kaggle.json

to copy kaggle.json into a folder first. Then to download the data first donwload the following script

wget -q https://raw.githubusercontent.com/probml/pyprobml/master/scripts/download_celeba.py

and run the following script

python download_celeba.py

To Train Model

python run.py -config ./configs/vanilla_vae.yaml

Results

Model	Paper	Reconstruction	Samples
Original Images (for reconstruction)	N/A		N/A
AE (Code, Config)	N/A
VAE (Code, Config)	Link
beta-VAE (Code, Config)	Link
Hinge VAE (Code, Config)	Link
MMD VAE (Code, Config)	Link
Info VAE (Code, Config)	Link
LogCosh VAE (Code, Config)	Link
Two-stage VAE (Code, Config)	Link
Sigma VAE (Code, Config)	Link
VQ-VAE (K = 512, D = 64) (Code, Config) + PixelCNN(Code)	Link

Acknowledgement

The idea of this zoo and some of the scripts were based on Anand Krishnamoorthy Pytorch-VAE library, we also used the script from sayantanauddy to transform and download the celeba from kaggle.

---