The project was done from 06/2021 to 09/2021 and the research was focused around Generative Adversarial Networks (GANs).
Created a basic GAN model and trained using MNIST dataset
- Network architecture in
conditional_gan_mnist/code_files/networks.py - Network hyper-parameters in
conditional_gan_mnist/code_files/params.py - Data preprocessing functions in
conditional_gan_mnist/code_files/preprocessing.py - Training function in
conditional_gan_mnist/code_files/training.py - Function that runs the model
conditional_gan_mnist/code_files/run_model.py - Utility functions used in the model
conditional_gan_mnist/code_files/util.py
Created a dataset of particles (circles / eggs / mixture of them, of different radius) and trained with conditional GAN, investigated the effects of having intermediate labels
- Data generation done in
conditional_gan_microstructure/code_files/data_generator.py - Circle generation class in
conditional_gan_microstructure/code_files/data_class.py - Egg generation class in
conditional_gan_microstructure/code_files/egg_class.py - Mixture (of circles and eggs) generation class in
conditional_gan_microstructure/code_files/mixture_class.py - cGAN used as a reference
Created an image dataset using Google Earth satellite images, and trained using conditional GAN - more details inside the Gantastic-Earth repository