STEdge: Self-training Edge Detection with Multi-layer Teaching and Regularization

  • 2022-01-13 18:26:36
  • Yunfan Ye, Renjiao Yi, Zhiping Cai, Kai Xu
  • 1

Abstract

Learning-based edge detection has hereunto been strongly supervised withpixel-wise annotations which are tedious to obtain manually. We study theproblem of self-training edge detection, leveraging the untapped wealth oflarge-scale unlabeled image datasets. We design a self-supervised frameworkwith multi-layer regularization and self-teaching. In particular, we impose aconsistency regularization which enforces the outputs from each of the multiplelayers to be consistent for the input image and its perturbed counterpart. Weadopt L0-smoothing as the 'perturbation' to encourage edge prediction lying onsalient boundaries following the cluster assumption in self-supervisedlearning. Meanwhile, the network is trained with multi-layer supervision bypseudo labels which are initialized with Canny edges and then iterativelyrefined by the network as the training proceeds. The regularization andself-teaching together attain a good balance of precision and recall, leadingto a significant performance boost over supervised methods, with lightweightrefinement on the target dataset. Furthermore, our method demonstrates strongcross-dataset generality. For example, it attains 4.8% improvement for ODS and5.8% for OIS when tested on the unseen BIPED dataset, compared to thestate-of-the-art methods.

 

Quick Read (beta)

loading the full paper ...