SC-VAE: Sparse coding-based variational autoencoder with learned ISTA

Introduction

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.

Training 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')

Share the link to model's weight

Model weights

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={Pattern Recognition},
     year={2025}
}

To-Do List

• Installing dependencies
• Training the Model
• Uploading pre-trained models to google driver