CRTrack: Learning Correlation-Refine network for visual object tracking

Wenkang Zhang, Fei Xie, Tianyang Xu, Jiang Zhai, Wankou Yang

Abstract: Highlights•We put forward Correlation-Refine network to address the issue of lacking semantic information.•We construct an efficient tracking framework, gradually eliminating similarity interference from semantic level.•Our tracker outperforms all previous convolutional trackers on four authoritative benchmarks.