Submitted by Gabriele Trivigno 5 INSID3: Training-Free In-Context Segmentation with DINOv3 Politecnico di Torino 228 2