ATCC: Accurate tracking by criss-cross location attention

In recent years, discriminative correlation filters (DCF) and Siamese networks based trackers have significantly advanced the performance in tracking. However, the problem of accurate target state estimation is not fully solved yet. Therefore, in this paper, we propose a Criss-Cross Location Attention (CCLA) module, which pays more concerns to global and local contextual information and is used for the adaptation of IoU-Net based trackers. Besides, our CCLA module has capability of high computational efficiency with a slight increase of network parameters. Then, we present our tracker called ATCC, a Siamese architecture with CCLA. Finally, we evaluate our tracker on OTB100, VOT-2018, LaSOT, and TrackingNet benchmark datasets. Experimental results show that our tracker performs favorably against other state-of-the-art trackers, while operating at 30 FPS on single GPU. We will release the code and models at https://github.com/yongwuSHU/atcc. (c) 2021 Elsevier B.V. All rights reserved.

Automated summary

In this paper, a Criss-Cross Location Attention (CCLA) module is proposed to address the issue of accurate target state estimation in tracking. The module focuses on global and local contextual information and is used for the adaptation of IoU-Net based trackers. The ATCC tracker, which combines a Siamese architecture with CCLA, shows favorable performance against state-of-the-art trackers on benchmark datasets while operating at 30 FPS on a single GPU.