Learning Efficient Transformer Representation for Siamese Tracker to UAV

Hao Wang, Xiaolou Sun, Wankou Yang, Qiang Wang

Abstract: In the last few years, there has been growing recognition of the vital links between visual tracking and unmanned aerial vehicle (UAV). Questions have been raised about the feasibility of CNN-based and transformer-based trackers on UAVs. However, deep neural network modules or self-attention modules can be adversely affected when employing on UAVs for their complex architecture. In this paper, we propose an efficient transformer tracker (ET2) with a lightweight network. The study set out to examine the usability of a transformer-based tracker. Our tracker performs at frame rates far surpassing real-time, achieving the balance of precision and running speed.