Fig 1.
The deep convolution and conventional convolution kernel are included in DO-Conv. ∘ means the depthwise convolution operator and * means convolution operator.
Fig 2.
It consists of the feature extraction subnetwork and cross-correlation operation. The feature extraction subnetwork contains conventional convolution layers and DO-Conv.
Table 1.
Ablation study on different convolution layers.
DOSiam achieves the best tracking performance when the DO-Conv is placed in the second layer.
Fig 3.
Comparison with state-of-the-art tracking algorithms.
The tracker DOSiam gets best tracking results in challenging environments of fast motion, scale variation, motion blur, in-plane and out-plane rotations.
Fig 4.
Precision and success plots on OTB2015.
Fig 5.
Precision and success rate on OTB2015.
Fig 6.
Precision and success plots of OPE.
Our tracker has good tracking performance in challenge environments fast motion, in-plane rotation, motion blur and the success rate and precision rate in these challenge environments are OTB2015 benchmark tested.
Fig 7.
A comparison of the DOSiam with state-of-the-art trackers in terms of success rate and precision on OTB2015 with different attributes.
Fig 8.
A comparison of the DOSiam with state-of-the-art trackers of precision and success rate in terms of fast motion, in-plane rotation, motion blur and scale variation.
Fig 9.
A comparison of the DOSiam with state-of-the-art trackers in terms of success rate and precision on VOT2016 and VOT2018 with different attributes, respectively.
Table 2.
Comparison with state-of-the-art trackers on VOT2016 in terms of EAO, A and R.
Table 3.
Comparison with state-of-the-art trackers on VOT2018 in terms of EAO, A and R.
Table 4.
Comparison with SiamFC on VOT2019-RGBT(TIR) in terms of EAO, A, R and FPS.
Table 5.
Comparison with SiamFC in terms of accuracy in some attributes on VOT2019-RGBT(TIR).
Fig 10.
A comparison of the DOSiam with state-of-the-art trackers in GOT-10k.
Table 6.
Comparison with state-of-the-art trackers on GOT-10k in terms of AO and SR0.5.