Generative Models

This repository contains the (unofficial) implementation of generative models. Results from these models can be found in ./output.

Usage

train.py:
  --[no]animation: If yes, animation of latent space interpolation will be created
    (default: 'false')
  --batch_size: Batch size.
    (default: '64')
    (an integer)
  --dataset: Dataset to train.
    (default: 'mnist')
  --epoch: Number of epoch.
    (default: '20')
    (an integer)
  --latent_dims: Number of latent dimensions.
    (default: '2')
    (an integer)
  --learning_rate: Initial learning rate.
    (default: '0.001')
    (a number)
  --model: Model to train.
    (default: 'VAE')
  --model_params: parameters for some models
    (default: 'key:value|key:value')

Autoencoders

• Vanilla Autoencoder (VanillaAE)
• Variational Autoencoder (VAE):
  • Its convolutional version (ConvVAE) is also implemented.
• BetaVAE

Implementaion Ideas

• [1406.5298] Semi-Supervised Learning with Deep Generative Models
  • GitHub - wohlert/semi-supervised-pytorch: Implementations of different VAE-based semi-supervised and generative models in PyTorch
• [1506.02216] A Recurrent Latent Variable Model for Sequential Data
• [1502.04623] DRAW: A Recurrent Neural Network For Image Generation
• [1410.6460] Markov Chain Monte Carlo and Variational Inference: Bridging the Gap

Acknowledgements

• This repository is inspired by @wiseodd's generative-models repository.
• Tutorial on Variational Autoencoders
• Variational Autoencoder and Extensions - VideoLectures.NET