notesum.ai

Published at December 4

TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception

cs.CV

Released Date: December 4, 2024

Authors: Runjian Chen¹, Hyoungseob Park², Bo Zhang³, Wenqi Shao³, Ping Luo¹, Alex Wong²

Aff.: ¹The University of Hong Kong; ²Yale University; ³Shanghai AI Laboratory

Arxiv: http://arxiv.org/pdf/2412.03054v1

Refer to caption

Init.	F.T.	mAP	Vehicle			Pedestrian			Cyclist
Init.	F.T.	mAP	0-30m	30-50m	50m-	0-30m	30-50m	50m-	0-30m	30-50m	50m-
Rand*	5%	20.48	58.03	25.22	12.98	11.62	9.75	6.97	21.55	6.83	3.11
TREND*		29.95 ^+9.47	63.54	33.05	18.94	18.47	14.51	9.44	41.82	20.55	8.53
Rand		46.07	76.71	51.15	31.84	37.53	20.12	9.84	62.00	42.61	24.18
TREND		47.84 ^+1.77	79.14	55.68	36.34	35.23	18.00	11.18	64.99	45.80	28.15
Rand*	20%	54.65	80.53	57.79	39.62	48.30	36.30	19.65	68.10	51.02	32.90
TREND*		55.97 ^+1.42	83.34	63.10	44.99	47.56	34.27	18.95	68.60	53.02	34.64
Rand		57.68	82.70	63.37	46.34	52.61	36.48	19.03	71.03	55.34	36.34
TREND		58.93 ^+1.25	84.08	65.80	50.51	50.31	33.37	19.42	72.54	56.31	39.26
Rand*	100%	64.00	86.21	70.20	58.20	57.80	41.18	23.55	75.95	61.45	45.80
TREND*		64.79 ^+0.79	87.90	72.38	60.29	57.21	40.60	24.97	77.26	61.92	46.14
Rand		65.03	88.18	74.23	61.75	57.32	38.90	21.96	78.07	64.32	48.16
TREND		65.66 ^+0.63	88.81	74.63	61.98	57.94	39.85	20.95	79.98	64.18	47.62