… ability of deep features from large convolutional … whole tracking system, we propose to
learn a feature distilled network (FDN) for faster extracting representative features while distilling …

… , our channel distillation still can further reduce feature channels to achieve slightly higher …
channel distillation on tracking performance and speed when selecting different deep feature …

… Additionally, we present a specificcommon feature distillation module that transfers both
modalityspecific and shared information from a more powerful model’s backbone to the compact …

… To remedy this situation, a cross-modality distillation framework is presented to bridge the
… a compact tracker and a powerful tracker. Specifically, a specific-common feature distillation …

… tracker. To bridge the gap between deep Siamese trackers and practical use, we propose a
new feature distillation … the feature distillation into a stepwise distillation problem and perform …

… In this paper, we unify such goals by proposing a novel tracking methodology that takes …
distillation and reinforcement learning. The first allows to transfer and compress tracking …

… -target tracking method based on knowledge distillation in [57] … high tracking speed
and meanwhile achieve high tracking … of knowledge distillation, we propose to use feature-based …

… features, such as HOG and Color Names [17], [35], [52]. Other than improved feature
representations, the DCF tracking … deep features, achieving high performance on multiple tracking …

… , which directly transfers the feature extraction ability from RGB to other … features in a
balanced, symmetric manner. Furthermore, we design a complementary masked patch distillation …

… feature distillation to focus the student model on the crucial region of the teacher model’s
feature … ’s attention for the region (on the feature map) of a larger difference with respect to the …