SC-VAE

This repository contains official implementation for the paper titled "SC-VAE: Sparse Coding-based Variational Autoencoder with Learned ISTA".

Installing Dependencies

To install dependencies, create a conda or virtual environment with Python 3 and then run pip install -r requirements.txt.

Running the SC-VAE

To run the SC-VAE simply run python main-stage1.py. You could change the config files in line 279 to train SC-VAE model with different downsampling blocks.

parser.add_argument('--model-config', type=str, default='./configs/ffhq/stage1/ffhq256-scvae16x16.yaml')

Evaluating the Model

Image Reconstruction

Image Generation

Image Pathches Clustering

Unsupervised Image Segmentation

Citation

@article{xiao2023sc,
     title={SC-VAE: Sparse Coding-based Variational Autoencoder with Learned ISTA},
     author={Xiao, Pan and Qiu, Peijie and Ha, Sung Min and Bani, Abdalla and Zhou, Shuang and Sotiras, Aristeidis},
     journal={Available at SSRN 4794775},
     year={2023}
}

To-Do List

• Installing Dependencies
• Training the Model
• Upload Pre-trained SC-VAEs
• Evaluating the Model: reconstuction; image generation; image patches clustering; unsupervised image segmentaton.