Deep Regression Tracking with Graph Attention
Chen Xie, Dawei Zhang, Zheng Zheng, Yiran He · 2022 International Conference on Image Processing, Computer Vision and Machine Learning (ICICML) · 2022
Deep regression tracking is one of the fastest tracking algorithms due to its simplicity of framework, and therefore it is suitable for practical applications. However, most regression-based trackers do not make good balance between tracking speed and accuracy. Besides, the features of two adjacent frames are simply concatenated together without feature fusion operations, which limits the tracking accuracy. To address the above issues, we present a deep regression network with graph attention for visual tracking, which makes a trade-off between tracking speed and accuracy. To be specific, we introduce the graph attention network into the deep regression tracking framework to establish part-to-part correspondence between the two adjacent frames, and propagate the target information from the previous feature to the current feature. Experiments on challenging benchmarks including GOT-10k, OTB-100 demonstrate the competitive performance of the proposed algorithm in comparison with the state-of-the-art trackers.