Siamese Attention Networks with Adaptive Templates for Visual Tracking

<table class="table-group" id="tab5"><tr><td><table class="table"><tr><td class="thead-hr" colspan="4"><hr/></td></tr><tr class="thead"><td class="align_left">Templates fusion strategies</td><td class="align_center">Accuracy↑</td><td class="align_center">Robustness↓</td><td class="align_center">EAO↑</td></tr><tr><td class="thead-hr" colspan="4"><hr/></td></tr><tr><td class="align_left">- (Baseline)</td><td class="align_center">0.600</td><td class="align_center">0.234</td><td class="align_center">0.414</td></tr><tr><td class="align_left">Fixed weighted sum</td><td class="align_center">0.610</td><td class="align_center">0.223</td><td class="align_center">0.436</td></tr><tr><td class="align_left">Learnable weighted sum (spatial attention module)</td><td class="align_center">0.618</td><td class="align_center">0.202</td><td class="align_center">0.445</td></tr><tr class="table-tr"><td colspan="4"><hr class="tbody-hr"/></td></tr></table></td></tr></table>

<div>Comparison results of different templates fusion strategies on VOT2018. SiamRPN++ is baseline. “-” denotes without templates fusion.</div>

Mobile Information Systems

tab5

Table 5

Table 5: Siamese Attention Networks with Adaptive Templates for Visual Tracking 