The task for this project is to implement a Variational Autoencoder (VAE) model for the purpose of generating medical images.
GPU: Recommended.
Your task (0.3): Train a VAE model to learn the distribution of images in MedMNIST. For +0.3 bonus, it is okay to focus only on one type of modality.
Possible extension (0.7): For the full bonus (+0.7), implement and evaluate a VAE variant (e.g. Beta VAE, Discrete label VAE, VQ VAE, CVAE) to disentangle the latent space, or to learn the class-conditional distribution for at least three different image classes. Implement a simple user interface (PyQt, Flask+HTML, TkInter,...) where the user can select an image class (PathMNIST, ChestMNIST,...) to sample from.
To run the project start a flask server (flask run) in the App folder. Then go to localhost:5000/home to visit the Flask frontend.
For just playing with the different models go into the Code folder where the models and some jupyter notebooks are available.
For other samples go into the Samples folder. Here you'll also finde some tensorflow implementations.
Our poster is in the folder Poster and our reports can be found in the Reports folder.