Boundary-Aware Segmentation
Overview
Boundary-aware segmentation aims to delineate objects and structures accurately, especially where boundaries are weak, thin, ambiguous, or clinically important. Common approaches add edge or distance-field prediction, boundary-weighted and topology-aware losses, frequency or geometric features, and targeted refinement guided by uncertainty; these methods are used in medical, remote-sensing, natural-scene, and 3D segmentation. Work also adapts foundation models with prompts, boundary-aware decoders, or distilled edge cues, and evaluates contours with measures such as Boundary F-score and Hausdorff distance alongside region-overlap metrics. A central challenge is improving boundary reliability without sacrificing region accuracy, generalization, or efficiency, particularly when annotations are uncertain or target domains differ.
Latest papers
106 papers6 papers in the last four weeks, down 57% on the four weeks before. 0.0% of all new papers.