登入
洞見 - 3D object tracking
PillarTrack: A Pillar-based Transformer Network for Efficient and Accurate 3D Single Object Tracking on Point Clouds
PillarTrack is a pillar-based 3D single object tracking framework that improves tracking performance while enhancing inference speed. It introduces a Pyramid-type Encoding Pillar Feature Encoder (PE-PFE) and a modality-aware Transformer-based backbone to effectively capture the geometric information in point clouds.
Efficient and Compact One-stream 3D Point Clouds Tracker: EasyTrack and EasyTrack++
EasyTrack is a novel and efficient one-stream 3D single object tracking framework that learns target-aware point cloud features through a unified feature learning and interaction module, without the need for heavy feature fusion networks. EasyTrack++ further improves the performance by introducing a center points interaction strategy to reduce the ambiguous features caused by background points.