AFTER: Attention-Based Fusion Router for RGBT Tracking.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Multi-modal feature fusion is crucial for Red-Green-Blue-Depth (RGBT) tracking.
• Existing RGBT trackers often use fixed fusion structures, limiting adaptability to dynamic scenarios.

Purpose of the Study:

• To develop a novel attention-based fusion router (AFTER) for robust RGBT tracking.
• To enable dynamic adaptation of fusion structures to challenging scenarios.

Main Methods:

• Designed a fusion structure space using a hierarchical attention network.
• Each unit in the network represents a fusion operation, and their combinations form fusion structures.
• Developed a dynamic routing algorithm with routers to predict combination weights for efficient fusion structure optimization.

Main Results:

• AFTER demonstrates superior performance compared to state-of-the-art RGBT trackers.
• Extensive experiments on five mainstream RGBT tracking datasets validate the effectiveness of the proposed method.

Conclusions:

• The proposed AFTER method offers a dynamic and adaptive approach to RGBT tracking.
• AFTER's ability to optimize fusion structures enhances robustness in challenging, dynamic environments.