Bi-Directional Pyramid Network for Edge Detection

Multi-scale representation plays a critical role in the field of edge detection. However, most existing research focuses on one two aspects: fast training and accurate testing. In this paper, we propose novel multi-scale method to resolve balance between them. Specifically, according multi-stream structures image pyramid principle, construct down-sampling network lightweight up-sampling enrich from encoder decoder, respectively. Next, these networks backbone constitute our overall architecture, bi-directional (BDP-Net). Extensive experiments show that compared with state-of-the-art model, could improve speed by about time while retaining similar test accuracy. Especially, under single-scale test, approach also reaches human perception (F1 score 0.803) BSDS500 database.