1 paper
Jiaxin Cheng, Yue Wu, Yicong Zhou
Learning-based edge detection models trained with cross-entropy loss often suffer from thick edge predictions, which deviate from the crisp, single-pixel annotations typically prov…