“…In [2], Bertinetto et al proposed SiamFC, a pioneering work that combines naive feature correlation with a fullyconvolutional Siamese network for object tracking. Subsequently, some improvements [85,68,75,82,76] are made to Siamese trackers, such as combining with a region proposal network [17,39,80,65] or an anchor-free FCOS detector [11], using a deeper architecture [38] or two-branch structure [23], exploiting attention [67,84] or self-attention [7], applying triplet loss [14]. However, these methods are specially designed for 2D object tracking, so they cannot be directly applied to 3D point clouds.…”