/VAE-Tensorflow

(beta-)VAE Tensorflow

Primary LanguagePythonMIT LicenseMIT

(beta-)VAE

Tensorflow implementation of VAE and beta-VAE

Exemplar results

  • Celeba

    • ConvNet (z_dim: 100, beta: 0.05)

      Generation Reconstruction
  • Mnist

    • ConvNet (z_dim: 10, beta: 0.1)

      Generation Reconstruction
    • MLP (z_dim: 10, beta: 0.1)

      Generation Reconstruction

Usage

  • Prerequisites

    • Tensorflow 1.8
    • Python 2.7 or 3.6
  • Examples of training

    CUDA_VISIBLE_DEVICES=0 python train.py --z_dim 10 --beta 0.1 --dataset mnist --model mlp_mnist --experiment_name z10_beta0.1_mnist_mlp
    CUDA_VISIBLE_DEVICES=0 python train.py --z_dim 10 --beta 0.1 --dataset mnist --model conv_mnist --experiment_name z10_beta0.1_mnist_conv
    CUDA_VISIBLE_DEVICES=0 python train.py --z_dim 32 --beta 0.1 --dataset celeba --model conv_64 --experiment_name z32_beta0.1_celeba_conv

Datasets

  1. Celeba should be prepared by yourself in ./data/celeba/img_align_celeba/.jpg*
  2. Mnist will be automatically downloaded