We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CV
< prev | next >
new | recent | 2006

Change to browse by:

cs

References & Citations

DBLP - CS Bibliography

listing | bibtex

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo

Computer Science > Computer Vision and Pattern Recognition

Title: 1st Place Solutions for Waymo Open Dataset Challenges -- 2D and 3D Tracking

Authors: Yu Wang, Sijia Chen, Li Huang, Runzhou Ge, Yihan Hu, Zhuangzhuang Ding, Jie Liao
(Submitted on 28 Jun 2020)
Abstract: This technical report presents the online and real-time 2D and 3D multi-object tracking (MOT) algorithms that reached the 1st places on both Waymo Open Dataset 2D tracking and 3D tracking challenges. An efficient and pragmatic online tracking-by-detection framework named HorizonMOT is proposed for camera-based 2D tracking in the image space and LiDAR-based 3D tracking in the 3D world space. Within the tracking-by-detection paradigm, our trackers leverage our high-performing detectors used in the 2D/3D detection challenges and achieved 45.13% 2D MOTA/L2 and 63.45% 3D MOTA/L2 in the 2D/3D tracking challenges.
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Cite as: arXiv:2006.15506 [cs.CV]
  (or arXiv:2006.15506v1 [cs.CV] for this version)

Submission history

From: Runzhou Ge [view email]
[v1] Sun, 28 Jun 2020 04:49:59 GMT (8053kb)
Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)

Link back to: arXiv, form interface, contact.