While remarkable progress has been made in robust visual tracking, accurate target state estimation still remains a highly challenging problem. In this paper, we argue that this issue is closely related to the prevalent bounding box representation, which provides only a coarse spatial extent of object. Thus an effcient visual tracking framework is proposed to accurately estimate the target state with a finer representation as a set of representative points. The point set is trained to indicate the semantically and geometrically significant positions of target region, enabling more fine-grained localization and modeling of object appearance. We further propose a multi-level aggregation strategy to obtain detailed structure information by fusing hierarchical convolution layers. Extensive experiments on several challenging benchmarks including OTB2015, VOT2018, VOT2019 and GOT-10k demonstrate that our method achieves new state-of-the-art performance while running at over 20 FPS.
@article{arxiv.2008.03467,
title = {RPT: Learning Point Set Representation for Siamese Visual Tracking},
author = {Ziang Ma and Linyuan Wang and Haitao Zhang and Wei Lu and Jun Yin},
journal= {arXiv preprint arXiv:2008.03467},
year = {2020}
}