# DFUC24 Baselines

This repository contains baseline models for the DFUC24 challenge. Two approaches were investigated:

1. Training a supervised segmentation head (using the DFUC22 dataset) on frozen image embeddings of the self-supervised pre-trained DINOv2 image backbone.

2. Using GroundedDINO + SAM in combination to achieve ulcer detection, which serves as a prompt for the SegmentAnything (SAM) model, as well as fine-tuned (using the DFUC22 dataset) versions of GroundedDINO and SAM.