SA-DPNet: Structure-aware dual pyramid network for salient object detection

作者：

Highlights：

• We propose a structure-aware spatial non-local block to harvest structure context into the spatial pyramid module.

• The gradient-based edge loss is exploited to enhance the edge structure context and the patch-based global structure context.

• We evaluate the proposed method on nine datasets and our network sets a new state-of-the-art performance.

摘要

•We propose a structure-aware spatial non-local block to harvest structure context into the spatial pyramid module.•The gradient-based edge loss is exploited to enhance the edge structure context and the patch-based global structure context.•We evaluate the proposed method on nine datasets and our network sets a new state-of-the-art performance.

论文关键词：Saliency detection,Structure coherence,Deep neural network

论文评审过程：Received 1 April 2020, Revised 24 February 2022, Accepted 5 March 2022, Available online 6 March 2022, Version of Record 20 March 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108624