SGAT: Shuffle and graph attention based Siamese networks for visual tracking

Area of Science:

• Computer Vision
• Artificial Intelligence

Background:

• Siamese-based trackers excel by learning target similarity.
• Current methods often underutilize spatial-channel information and employ linear matching.
• Existing approaches neglect structured and part-level target details.

Purpose of the Study:

• To propose a novel visual tracking algorithm for improved feature extraction.
• To enhance target representation by exploiting spatial-channel correlations.
• To introduce a graph attention matching mechanism for robust tracking.

Main Methods:

• Utilizing convolutional neural networks and shuffle attention for feature extraction.
• Implementing graph attention matching for similarity computation.
• Exploiting spatial-channel correlations to highlight target regions.

Main Results:

• The proposed algorithm effectively highlights target regions using spatial-channel correlations.
• Graph matching significantly reduces the impact of appearance variations like occlusions.
• Achieved excellent tracking performance on multiple challenging benchmarks.

Conclusions:

• The novel tracking algorithm demonstrates superior performance compared to state-of-the-art methods.
• The integration of shuffle attention and graph attention matching enhances tracking robustness.
• The approach offers a promising direction for advanced visual object tracking.