Natural Language Guided Attention Mixer for Object Tracking
Qingyang Lu, Guanglin Yuan, Congli Li, Hong Zhu, Xiaoyan Qin · 2023
The current research on object tracking using visual and language information shows promise. A significant amount of research has been done to achieve this goal by constructing models that uniformly characterize multimodal information. Different from previous work, this paper introduces a language-guided attention mixer. The mixer adaptively guides visual tracking attention using natural language and continuously corrects visual bias through linguistic information. This approach not only improves target feature recognition during the tracking process but also reduces training costs and introduces smaller parameters, resulting in significant improvements. The tracker proposed in this paper achieves state-of-the-art performance on four benchmarks with language annotation, all while running in real-time.