Learning Crisp Boundaries Using Deep Refinement Network and Adaptive Weighting Loss

Significant progress has been made in boundary detection with the help of convolutional neural networks. Recent models not only focus on real object but also “crisp” boundaries (precisely localized along object's contour). There are two methods to evaluate crisp performance. One uses more strict tolerance measure distance between ground truth and detected contour. The other focuses evaluating contour map without any postprocessing. In this study, we analyze both conclude that aspects evaluation. Accordingly, propose a novel network named deep refinement (DRNet) stacks multiple modules achieve richer feature representation loss function, which combines cross-entropy dice through effective adaptive fusion. Experimental results demonstrated state-of-the-art performance for several available datasets.