Siamese Attention Networks with Adaptive Templates for Visual Tracking

Figure 2

Illustration of main framework of SiamAttnAT. The network consists of backbone (ResNet50), memory pool, template generation module, template selection module, and Siamese Region Proposal (SiamRPN) modules. The network generates the classification and regression boxes by fusing the outputs from multiple SiamRPN blocks with weighted sum. Best viewed in colour.