Description

FaceCraft was my final project for my bachelor's degree at Full Sail University. This five-month-long capstone project was a team of three that conducted research on generating realistic faces from generative adversarial networks on consumer-market GPUs. Our research found that a StyleGAN2 architecture is the most suitable model design for consumer performance and GPU memory usage. Read about my work in this project here: Capstone Blogs

My Work in the Project

• Machine Learning Engineering: Developed the discriminator component of a GAN model from the StyleGAN2 research paper.
• Environment Development: Addressed compatibility issues with CUDA toolkit and Docker to ensure smooth development and deployment processes.
• Front-End Development: Sole developer of the project's front-end using HTML, CSS, and Gradio; focusing on user interface design and responsiveness.
• Project Management and Design Documentation: Created design documents and developed project management strategies using Jira for team collaboration and efficiency.
• HuggingFace Hosting: Research cost effective hosting, landing on HuggingFace. Addressed challenges with integrating a custom model on the platform using Gradio.
• Exploratory Data Analysis (EDA): Conducted exploratory data analysis on image accuracy and generated captions using statistical techniques and visualizations.
• Image Manipulation: Implemented background removal and watermarking features, enhancing the visual output of generated images.
• Documentation: Created detailed instructions for using Docker to train and generate images, and established a new GitHub repository for the project's final product.