AUTHOR=Chen Yongliang , Lin Chuan , Qiao Yakun 

TITLE=DPED: Bio-inspired dual-pathway network for edge detection

JOURNAL=Frontiers in Bioengineering and Biotechnology

VOLUME=Volume 10 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/bioengineering-and-biotechnology/articles/10.3389/fbioe.2022.1008140

DOI=10.3389/fbioe.2022.1008140

ISSN=2296-4185

ABSTRACT=As the basis of high-level visual tasks, edge detection is significant. Most of the encoder-decoder edge detection methods used convolutional neural networks, such as VGG16 or Resnet, as the encoding network. And the studies on designing the decoding network have achieved good results. Swin Transformer (Swin) has recently attracted much attention in various visual tasks as a possible alternative to convolutional neural networks. Physiology studies have shown that two visual pathways converge in the visual cortex in the biological vision system, and the complex information transmission and communication exists widely. Inspired by the research on Swin and the biological vision pathway, we have designed a two-pathway encoding network. The first-pathway network is the fine-tuned Swin, and the second-pathway network mainly comprises deep separable convolution. To simulate the attention transmission and feature fusion between the first-pathway network and the second-pathway network, we design the second-pathway attention module and the pathways fusion module. On BSDS500 datasets, our proposed method outperforms the CNN-based SOTA method BDCN. Moreover, our proposed method and the Transformer-based SOTA method EDTER have their own advantages in terms of performance. In terms of FLOPs and FPS, our method has more benefits than EDTER.