Seomething to make us laugh
What it does
It tells you which celebrity you look most alike (limited by those currently in the ML database)
How I built it
Messenger api was used with a node backend was used to interact with messenger users on Facebook. Flask was used to provide a backend for the pytorch model to provide predicts on user uploaded images. Communication between Node and Flask using POST. Machine learning architecture used was a pretrained ResNext model with the final layer retrained using transfer learning.
Challenges I ran into
Challenge of collecting data for reliable machine learning outcomes Time needed for training a deep learning model (with limited GPU access). Commuincation between Node and Flask using POST .
Accomplishments that I'm proud of
Getting a messenger app up and running that interfaced with deep learning tech.
What I learned
Many things, but am glad I learned how to communicate between two servers using POST.
What's next for CelebCheck
A better pytorch model. Increase the number of Celebs in the database. Show image of the matching celeb.