EyeSpy AI: Retinal Video Sequences

  • For our entry in UC Berkeley’s 5th Annual Datathon, I, along with my three teammates Jonathan Ferrari, Dhruv Pendharkar, and Miles Wang, crafted a Variational Autoencoder capable of producing videos from retinal scans and their numerical frameshift sequences. We earned an Honorable Mention in the competition.
  • The link to our presentation can be found here!

Description

  • The GitHub Repo for this project can be found here!
  • Coded a Variational Auto-Encoder (VAE) neural network to convert retinal movement numerical data into generative AI video simulations of retinal movement afflicted by one of three different diseases
  • Implemented a data preprocessing pipeline that converts .avi files into sets of 300 individual frames
  • Ran training data through a Long-Short Term Memory (LSTM) network to encode data into latent space, then decoded data using a Gated Recurrent Unit (GRU), producing video simulations in 512x512 resolution

Tools

  • Python
  • NumPy
  • PyTorch
  • Cv2

Skills

  • Computer Vision
  • Machine Learning
  • Deep Learning