GANs Project: Face Generation

The project files were located in the Project Workspace of Udacity forum and included the following files:

• dlnd_face_generation_starter.ipynb
• README.md
• requirements.txt
• tests.py
• processed-celeba-small.zip

Summary

The project was completed in the file dlnd_face_generation_starter.ipynb. This project was organized as follows:

• Data Pipeline: implemented a data augmentation function and a custom dataset class to load the images and transform them.
• Model Implementation: built a custom generator and a custom discriminator to make my custom GAN
• Loss Functions and Gradient Penalty: design decision made on loss functions and then also used gradient penalty for the discriminator step.
• Training Loop: implemented the training loop and performed hyperparameter tunings to achieve the expected accuracy of the model.

Each section required me to make design decisions based on the experience I have gathered from the course.

Building a deep learning model is an iterative process, and it's especially true for GANs! Therefore, I thoroughly enjoyed the process and I learned how to implement it from the scratch, which provided me an indepth of the concepts, that I learned from the course.