Designed, trained, and tested a U-Net convolutional neural network using PyTorch. My algorithm segments microcopy cell images to predict the location of a cell's nucleus. During training, each input image has a corresponding mask which labels a each pixel with a binary encoding as either a nucleus pixel or not. During testing, this mask was predicted on new data.
For a full description of my design architecture and results, please refer to my final report (U-Net Report.pdf).
The original 2015 paper which introduced the U-Net architecture can be found linked within my project report (U-Net Report.pdf).