A Twofold Siamese Network for Real-Time Object Tracking

RGB-T object tracking: Benchmark and baseline

“…2) End-to-End Deep Learning based Trackers: Benefiting from deep learning techniques, many end-to-end deep trackers have been proposed and achieve excellent performance [37][38][39][40][41]. Hyeonseob et al [37] introduce the idea of multi-domain and train a domain-specific layers online for each video to adapt the target representations.…”

“…To handle this problem, Guo et al [40] propose a dynamic Siamese network with a fast general transformation learning model. He et al [41] design a twofold Siamese network that consists of a semantic branch and an appearance branch to improve the discrimination power of SiameFC in tracking. Besides, the correlation filter is also embedded into deep neural network [39], named CFNet, to achieve end-to-end tracking.…”