This repository contains the implementation of score-based generative models with SDEs in PyTorch.

We implement a score-based generative model with SDEs to generate novel digital images of the retina for detecting retinopathy. The models and op folders are incorporated from the GitHub repo of Yang Song link to repo.

We observe that the sampler generates high quality samples comparable to the retinal dataset images. The sampler takes 2000 steps to generate these images.

Original Retinal Images

Generated Retinal Images

The model was trained on the Digital Retinal Images for Vessel Extraction dataset link.

To set the parameters, edit the config.py file in the configs folder.
The jupyter notebook implementation(main.ipynb) can be used to either train the model further or generate more samplers.

Score-Based Generative Modeling through Stochastic Differential Equations by Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole

@inproceedings{
  song2021scorebased,
  title={Score-Based Generative Modeling through Stochastic Differential Equations},
  author={Yang Song and Jascha Sohl-Dickstein and Diederik P Kingma and Abhishek Kumar and Stefano Ermon and Ben Poole},
  booktitle={International Conference on Learning Representations},
  year={2021},
  url={https://openreview.net/forum?id=PxTIG12RRHS}
}