EDSNN: Edge Detection with Spiking Neuron Network

Yimeng Fan, Changsong Liu, Mingyang Li, Wei Zhang

Keywords: Spiking Neural Networks, Edge Detection, Spiking Multi-Scale Block, Membrane Average Decoding

TL;DR: We propose the first SNN-based method named EDSNN (Edge Detection with Spiking Neural Network) for edge detection.

Abstract: Edge detection has made great progress under the development of Artificial Neural Networks (ANNs), particularly Convolutional Neural Networks (CNNs) and Transformers, some of them even have achieved a beyond human-level performance. However, these methods come with complex designs and high energy consumption. Spiking Neural Networks (SNNs), with their low energy consumption and biological interpretability, offer a promising solution to address these issues. In this work, we propose the first SNN-based method named EDSNN (Edge Detection with Spiking Neural Network) for edge detection. We construct a novel Spiking Multi-Scale Block (SMSB) to effectively utilize multi-scale information, thereby helping the network generate precise and clean edge maps. In addition, to more accurately decode spike trains, we present a Membrane Average Decoding (MAD) method in the prediction block. Our method has the advantages of remarkable efficiency and high performance across multiple datasets. It surpasses the human-level performance on BSDS500 (ODS=0.804 vs. ODS=0.803) while consuming only 14.64 mJ, remains competitive performance among top-performing ANN-based approaches on NYUDv2 (ODS=0.750), and achieves state-of-the-art performance on BIPED (ODS=0.891). Our codes are publicly available in supplementary materials.

Primary Area: applications to neuroscience & cognitive science

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 4599