New scientific research depends on new data, especially in the field of microscopy, where taking pictures is difficult. The complexity of the process is shown by the construction of high-resolution microscopes. This project try to address these problems by creating synthetic data to improve cellular-level images. The methodology is to select eligible patches from the raw images, which are then augmented to create a comprehensive image dataset. The dataset is then processed through a Variational Autoencoder (VAE) to generate new image data.
/saved_images # Contains images of the model architecture, the model performance, and a visualization of created images
data_gen_DN # Script for generating the training data for the low quality images vector of the initial data
data_gen_SR # Script for generating the training data for the super-resolved images vector of the initial data
augmentation.py # Script with the implementation of the different translation techniques (Flipping, transposing, rotating...)
vae_training.py # Script with the architecture and VAE training