In this project, we would train a StyleGAN2 model for medical datasets. This could be beneficial for synthetic data augmentation, or potentially encoding into and studying the latent space could be useful for other medical applications.
Personally, I am more interested in histopathological datasets:
However, I don’t mind looking into other medical datasets and training a StyleGAN2 for them:
This repository has a JAX model implementation of StyleGAN2 but unfortunately not a training script.
At the end of the project we would hopefully have:
- A JAX-based pipeline for training StyleGAN2 models on arbitrary datasets and for exploring the latent space of these models.
- A trained StyleGAN2 model for the selected datasets (with an online demo)
Let me know if you find this project interesting and if you have any related ideas…