This project has been developed for the Vision and Perception course at Sapienza, and the aim was to build a GAN (Generative Adversarial Network) able to generate from scratch faces of anime, with the ability to overcome the non convergence problems that raise at resolution higher than 64×64 pixels, and adopting strategies that allowed the generation of 128×128 pixels images, with a good level of detail and realism.
The implementation details and choices can be found in powerpoint presentation file
The code is available in the python notebook above.