Will Hoover

Week of 3/17 – 3/24/2024
The team is also hard at work on the different classes we have decided to take, including the LinkedIn learning class on building a GAN, a class on React.js to understand how to build the front end of the web application and classes on how to use google web services. The team has decided to go with Hugging Face as our hosting site for the web application since it is focused on deploying machine learning models like ours, so we thought the added functionality that the site offers would be beneficial for our application.

I have found the GAN class on LinkedIn to be very interesting and beneficial to the project. It has shown me, in a more in-depth way how to build a GAN, train the generator and the discriminator in a way that will produce results and different ways that I could adjust a basic GAN to get the results that I require. This is a class on the basics of a GAN so there is more to learn and more papers to read to be able to build a network that will be able to produce a human face that is recognizable and able to obtain the quality that the team is looking for. I have decided to use the Flicker Face HQ dataset to build the network, doing 3 training sessions from 128*128 up to 1024*1024 resolution. This incremental training will theoretically help with faster times and increased quality for the end model. The 128 resolution datasets will provide the network with the basics of the shapes of a human head and some features but will be much faster to train so we would be able to put out a minimum viable product (MVP) within a short amount of time, giving me the opportunity to train a more in-depth model with a more lenient time frame. The 512 and 1024 resolution training sessions will take much longer, up to a few days for the 512 and over a week for the 1024. We will need time to do this, and if we are able to get all the functions in place beforehand with the MVP then we will have that leeway to train the model up to the standard that we are looking at for our final product.

Thank you for reading, I’ll see you next week.

– Will Hoover