notesum.ai

Published at November 13

3D Multi-Object Tracking with Semi-Supervised GRU-Kalman Filter

cs.RO

cs.AI

Released Date: November 13, 2024

Authors: Xiaoxiang Wang, Jiaxin Liu, Miaojie Feng, Zhaoxing Zhang, Xin Yang

Arxiv: http://arxiv.org/abs/2411.08433v1

Refer to caption

Method	Detector	Input Data	AMOTA $\uparrow$	AMOTP $\downarrow$	IDS $\downarrow$
CBMOT[26]	CenterPoint[2] & CenterTrack[27]	2D + 3D	68.1	52.8	761
EagerMOT[10]	CenterPoint & Cascade R-CNN[28]	2D + 3D	67.7	55.0	1156
SimpleTrack[6]	CenterPoint	3D	66.8	55.0	575
OGR3MOT[9]	CenterPoint	3D	65.6	62.0	288
CenterPoint	CenterPoint	3D	65.0	53.5	684
Poly-MOT[7]	CenterPoint	3D	70.0	50.9	331
Poly-MOT¹	LargeKernel3D[29]	2D + 3D	75.4(74.2)	42.2(46.5)	292(276)
Ours	CenterPoint	3D	70.0	50.4	353
Ours	LargeKernel3D	2D + 3D	74.4	42.9	352