An Enhanced Visual Attention Siamese Network That Updates Template Features Online

Figure 1

ESA-Siam network framework. Based on the Siamese benchmark network framework, ESA-Siam uses ResNet50 as the backbone network to do attention screening on the last three network blocks of the template branch and the search branch. After that, cross-correlation operations are performed on the template features and their respective search features, and then the fusion features are performed to obtain the final output feature map.