Space-Correlated Transformer: Jointly Explore the Matching and Motion Clues in 3D Single Object Tracking

Fei Xie, Zhongdao Wang, Chao Ma

Keywords: 3D Single Object Tracking; Point Cloud; Transformer; Space-Correlation

TL;DR: We present a novel and conceptually simple tracking framework dubbed SCtrack, which jointly explores the matching and motion clues in 3D single object tracking for point clouds.

Abstract: 3D Single Object Tracking (3D SOT) in LiDAR point clouds plays a crucial role in autonomous driving. Current approaches mostly follow two paradigms, i.e., Siamese matching-based and motion-centric. However, LiDAR point clouds lack enough appearance information, while the motion-centric trackers suffer from complex model structures. To address these issues, we present a novel and conceptually simple tracking framework dubbed SCtrack, which jointly explores the matching and motion clues in point clouds. Specifically, SCtrack embeds point clouds into spatially structured features and conducts space correlation along the aligned spatial region. The target relative motion is directly inferred from the correlated features. In contrast to prevalent PointNet-based features, our spatially structured representation inherently models motion clues among the consecutive frames of point clouds, thereby being complementary to appearance matching. To better utilize the aligned structured features, we employ a strategy of varied-size space regions that adapt to different target shapes and locations during space correlation. Without bells and whistles, SCtrack achieves leading performance, with 89.1%, 71.5%, and 62.7% precision on KITTI, NuScenes, and Waymo Open Dataset, and runs at a considerably high speed of 60 Fps on a single RTX3090 GPU. Extensive studies validate the effectiveness of our SCtrack framework. The code will be released.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 292