SiamDA: Dual attention Siamese network for real-time visual tracking

作者：

Highlights：

• We propose a non-local attention module to highlight the target area and suppress the background.

• We propose a channel attention module to compute channel-wise weights based on feature responses of specific targets.

• The two attention branches are jointly trained and the similarity score maps are combined.

• Our tracker achieves very remarkable performance compared to many state-of-the-art trackers and shown significant improvements towards SiamFC.

摘要

•We propose a non-local attention module to highlight the target area and suppress the background.•We propose a channel attention module to compute channel-wise weights based on feature responses of specific targets.•The two attention branches are jointly trained and the similarity score maps are combined.•Our tracker achieves very remarkable performance compared to many state-of-the-art trackers and shown significant improvements towards SiamFC.

论文关键词：Visual tracking,Siamese network,Non-local attention,Channel attention

论文评审过程：Received 11 December 2019, Revised 10 December 2020, Accepted 11 April 2021, Available online 18 April 2021, Version of Record 27 April 2021.

论文官网地址：https://doi.org/10.1016/j.image.2021.116293