Enhanced Extractor-Selector Framework and Symmetrization Weighted Binary Cross-Entropy for Edge Detections

Hao Shu·January 23, 2025

Summary

The Enhanced Extractor-Selector Framework with SWBCE improves edge detection by addressing traditional E-S method limitations. It outputs richer features, emphasizing edge recall and non-edge suppression, outperforming baselines on BIPED2 with significant accuracy boosts. This work surpasses previous methods, maintaining competitive performance across diverse datasets without needing specific hyperparameter tuning. Extensive experiments validate the superiority of the proposed methods, referencing key studies on edge detection and image segmentation.

Key findings

3

Advanced features