Convolutional Variational Autoencoder on the SVHN Dataset

Data Description

• 10 classes, 73257 train images
• 3 x 32 x 32 tensors.

Install

python3 -m venv venv
source venv/bin/activate
pip install requirements.txt

Train

python train.py --gen_images_dir images --num_epochs=100 --batch_size=64

Images will be samples and generated at the end of each epoch in the --gen_images-dir directory.

Sample Generated Images (103 epochs)