U-Net NK Cell Segmentation
DTU course project on segmenting Natural Killer cells in fluorescence microscopy images. Implemented an OpenCV baseline (62.49% IoU) and two U-Net models in PyTorch — achieving 80.35% IoU with a supervised ResNet50 encoder and 78.85% with a self-supervised DINO-pretrained ResNet18, using nested cross-validation on 41 annotated images.
View Project →