Beyond 3D Siamese Tracking: A Motion-Centric Paradigm for 3D Single Object Tracking in Point Clouds

3D single object tracking (3D SOT) in LiDAR point clouds plays a crucial rolein autonomous driving. Current approaches all follow the Siamese paradigm basedon appearance matching. However, LiDAR point clouds are usually textureless andincomplete, which hinders effective appearance matching. Besides, previousmethods greatly overlook the critical motion clues among targets. In this work,beyond 3D Siamese tracking, we introduce a motion-centric paradigm to handle 3DSOT from a new perspective. Following this paradigm, we propose a matching-freetwo-stage tracker M^2-Track. At the 1^st-stage, M^2-Track localizes the targetwithin successive frames via motion transformation. Then it refines the targetbox through motion-assisted shape completion at the 2^nd-stage. Extensiveexperiments confirm that M^2-Track significantly outperforms previousstate-of-the-arts on three large-scale datasets while running at 57FPS (~8%,~17%, and ~22%) precision gains on KITTI, NuScenes, and Waymo Open Datasetrespectively). Further analysis verifies each component's effectiveness andshows the motion-centric paradigm's promising potential when combined withappearance matching.