Double Branch Attention Block for Discriminative Representation of Siamese Trackers

Siamese trackers have achieved a good balance between accuracy and efficiency in generic object tracking. However, background distractors cause side effects to the discriminative representation of target. To suppress sensitivity distractors, we propose Double Branch Attention (DBA) block tracker equipped with DBA named DBA-Siam. First, concatenates channels multiple layers from two branches framework obtain rich feature representation. Second, channel attention is applied concatenated blocks enhance robust features selectively, thus enhancing ability distinguish target complex background. Finally, collects contextual relevance adaptively encodes it into weight detection branch for information compensation. Ablation experiments show that proposed can significantly improve tracking performance. Results on popular benchmarks DBA-Siam performs favorably against its counterparts. Compared advanced algorithm CSTNet, improves EAO by 18.9% VOT2016.