The project aims to automate the manual worm counting process which is tedious and time consuming for scientists. Each experiment can take 2 days for scientists to count egg/dead/alive worms.

  • Develop machine vision protocol for scientists to collect worm images during their experiments
  • Collect data annotation from the scientists and through Amazon MTurk
  • Develop and evaluate deep learning models (Faster R-CNN, U-Net) to detect different worm types (egg, dead, alive) using Python and Tensorflow
  • Deploy the best model on production, reducing the worm counting time from 2 days (manual counting) to 1 hour (automated counting with computer vision).