notesum.ai

Published at December 5

PriorMotion: Generative Class-Agnostic Motion Prediction with Raster-Vector Motion Field Priors

cs.CV

cs.PF

cs.RO

Released Date: December 5, 2024

Authors: Kangan Qian¹, Xinyu Jiao, Yining Shi, Yunlong Wang, Ziang Luo, Zheng Fu, Kun Jiang, Diange Yang

Aff.: ¹Tsinghua University, China

Arxiv: http://arxiv.org/pdf/2412.04020v1

Refer to caption

Method	Backbone	Static		Speed $\leq$ 5m/s		Speed $>$ 5m/s		Motion.S
Method	Backbone	Mean $\downarrow$	Median $\downarrow$	Mean $\downarrow$	Median $\downarrow$	Mean $\downarrow$	Median $\downarrow$	Variance $\downarrow$
StaticModel	Rules	0	0	0.6111	0.0971	8.6517	8.1412	0
FlowNet3D[24]	PointNet	0.0410	0	0.8183	0.1782	8.5261	8.0230	0
HPLFlowNet[11]	BCL	0.0041	0.0002	0.4458	0.0960	4.3206	2.4881	0
PointRCNN[56]	PointNet	0.0204	0	0.5514	0.1627	3.9888	1.6252	0
LSTM-EM[29]	LSTM	0.0358	0	0.3551	0.1044	1.5885	1.0003	0
Pillar.M(L&I)[25]	Pillar.E	0.0245	0	0.2286	0.0930	0.7784	0.4685	-
MotionNet[49]	STPN	0.0262	0	0.2467	0.0961	0.9878	0.6994	0.0267
MotionNet[49]†	STPN	0.0201	0	0.2292	0.0952	0.9454	0.6180	0.0251
MotionNet[43]‡	STPN	0.0271	0	0.2267	0.0945	0.8427	0.5173	-
PriorMotion(Ours)	STPN	0.0285	0	0.2364( $\downarrow$ 4.18%)	0.0893	0.8064( $\downarrow$ 18.36%)	0.5869	0.0074
ST-Trm[47]	STT	0.0214	0	0.2426	0.0957	1.0504	0.7247	0.0231
PriorMotion(Ours)	STT	0.0267	0	0.2341( $\downarrow$ 3.51%)	0.0949	0.9136( $\downarrow$ 13.02%)	0.5738	0.0094
STI[44]	STI	0.0244	0	0.2375	0.0950	0.9078	0.6262	0.0221
BE-STI[44]†	STI	0.0220	0	0.2115	0.0929	0.7511	0.5413	0.0108
PriorMotion(Ours)	STI	0.0273	0	0.2207( $\downarrow$ 7.07%)	0.0931	0.7287( $\downarrow$ 19.73%)	0.5329	0.0075