Fig 1.
Architecture of the proposed Shuffle and Graph Attention Tracker(SGAT).
Fig 2.
The details of the channle and spatial blocks.
Table 1.
Ablation experiments on OTB-100 benchmark.
GM denotes graph matching, Xcorr denotes cross correlation, respectively, and SA means shuffle attention unit.
Fig 3.
Comparison with state-of-the-art trackers on OTB-100 in terms of success plots.
Fig 4.
Success rate vs. tracking speed on OTB-100.
Here, the x-axis represents the 10th power of the tracking speed and the y-axis represents the success rate.
Table 2.
Comparison with state-of-the-art trackers on GOT-10k benchmark.
AO, success rates(SR)0.5 and success rates(SR)0.75 represent the average overlap and the success rate at the threshold of 0.5 and 0.75.
Fig 5.
Comparison with other state-of-the-art trackers on the GOT-10k benchmark, the proposed tracker SGAT achieve the best tracking performance.
A in SiamFC++_A indicates that the backbone network AlexNet.
Fig 6.
Comparision with state-of-the-art trackers on LaSOT in terms of the precision rate, success rate and normalized precision plots.
Fig 7.
Comparision with different trackers on each attribute of LaSOT.
SiamFC_pp and SiamRPN_pp represent trackers SiamFC++ and SiamRPN++, respectively.
Table 3.
A performance comparison with other competitive methods on the test split of LaSOT, where Suc., Pre. and Norm.Pre. represent the success rate, precision, normalized precision, respectively.
Table 4.
Comparison with other competitive methods on the test split of UAV123 in terms of success rate and precision rates.
Fig 8.
Qualitative results on four challenging sequences with other state-of-the-art trackers.
Fig 9.
Two cases of failure, in which the red box mean ground truth and the green box mean SGAT tracker.