Conceptual
Login

Context Clustering Vision Mamba for Medical Image Segmentation

A U-shaped visual state space (Mamba) segmentation architecture whose CCS6 layer combines VMamba's four-direction cross-scan with a context clustering layer that treats image patches as sets of points and dynamically clusters them within local windows. Students learn how learnable local clustering restores the short-range spatial dependencies that flattening-based Mamba scans lose, while keeping linear-complexity global feature interactions for tasks like nuclei, skin lesion, and multi-organ segmentation.