PyTorch VAE

A Collection of Variational AutoEncoders (VAEs) implemented in PyTorch with focus on reproducibility. The aim of this project is to provide a quick and simple working example for many of the cool VAE models out there. All the models are trained on the CelebA dataset for consistency and comparison. The architecture of all the models are kept as similar as possible with the same layers, except for cases where the original paper necessitates a radically different architecture.

Requirements

• Python >= 3.5
• PyTorch >= 1.3
• Pytorch Lightning >= 0.5.3 (GitHub Repo)

Installation

$ git clone https://github.com/AntixK/PyTorch-VAE
$ cd PyTorch-VAE
$ pip install -r requirements.txt

Usage

$ cd PyTorch-VAE
$ python run.py -c configs/<config-file-name.yaml>

Config file template

model_params:
  name: "<name of VAE model>"
  in_channels: 3
  latent_dim: 

exp_params:
  data_path: "<path to the celebA dataset>"
  img_size: 64    # Models are designed to work for this size
  batch_size: 64  # Better to have a square number
  LR: 0.005

trainer_params:
  gpus: 1         
  max_nb_epochs: 50

logging_params:
  save_dir: "logs/"
  name: "<experiment name>"
  manual_seed: 

---

Results

Model	Paper	Reconstruction	Samples
VAE	Link
WAE - MMD (RBF Kernel)	Link
WAE - MMD (IMQ Kernel)	Link
Beta-VAE	Link
IWAE (5 Samples)	Link
DFCVAE	Link
MSSIM VAE	Link

TODO

• VanillaVAE
• Conditional VAE
• Gamma VAE
• Beta VAE
• Beta TC-VAE
• DFC VAE
• MSSIM VAE
• InfoVAE (MMD-VAE)
• WAE-MMD
• AAE
• TwoStageVAE
• VAE-GAN
• Vamp VAE
• HVAE (VAE with Vamp Prior)
• IWAE
• VLAE
• FactorVAE
• PixelVAE
• VQVAE
• StyleVAE

Contributing

If you have trained a better model using these implementations by finetuning the hyper-params in the config file, I would be happy to include your result (along with your config file) in this repo, citing your name 😊.