Download
vae is a collection of Keras experiments implementing variational autoencoders and related generative models. It includes simple VAE and conditional VAE examples along with several alternative architectures. Separate scripts explore CelebA image generation, convolutional VAEs, clustering-oriented variants, hyperspherical latent spaces, and vector-quantized autoencoders. The repository includes sample output from a CelebA training run as a visual reference. Its documented environment uses Python 2.7 with TensorFlow 1.8 or 1.13 and Keras 2.2.4. The code is organized as standalone experiments rather than a unified library API. It is best suited for studying older Keras implementations of latent-variable generative modeling techniques.
Features
- Basic variational autoencoder implementation
- Conditional VAE examples
- Convolutional VAE experiments
- CelebA image generation
- Clustering and hyperspherical latent-space variants
- Vector-quantized autoencoder implementation
Project Samples
