notesum.ai

Published at October 24

Dynamic 3D Gaussian Tracking for Graph-Based Neural Dynamics Modeling

cs.HC

cs.AI

cs.CV

cs.LG

Released Date: October 24, 2024

Authors: Mingtong Zhang¹, Kaifeng Zhang², Yunzhu Li²

Aff.: ¹University of Illinois Urbana-Champaign; ²Columbia University

Arxiv: https://arxiv.org/abs/2410.18912v1

Refer to caption

Method	Metric	Rope	Cloth	Toy Animals	Metric	Rope	Cloth	Toy Animals
CoTracker [12]	3D MTE [mm] $\downarrow$	46.96	55.84	51.84	2D MTE [mm] $\downarrow$	42.73	49.82	46.30
PIPS++ [61]		49.37	58.10	58.45		46.17	50.28	49.96
SpatialTracker [15]		38.72	53.85	42.82		32.29	46.26	37.49
Dyn3DGS [8]		62.41	69.20	66.19		57.39	61.28	60.17
Ours		6.90	13.14	12.83		4.92	11.72	10.94
CoTracker [12]	3D $\delta_{avg}$ $\uparrow$	79.28	75.79	75.46	2D $\delta_{avg}$ $\uparrow$	83.71	79.28	78.44
PIPS++ [61]		69.83	66.92	68.95		76.30	72.48	76.96
SpatialTracker [15]		86.56	83.85	79.42		92.36	90.52	89.62
Dyn3DGS [8]		60.32	56.28	61.97		67.20	62.28	67.96
Ours		89.26	89.13	82.71		93.27	92.18	94.19
CoTracker [12]	3D Survival $\uparrow$	94.41	95.19	92.26	2D Survival $\uparrow$	97.20	100	96.06
PIPS++ [61]		92.61	91.39	85.73		96.74	94.28	92.82
SpatialTracker [15]		100	98.46	96.26		100	100	100
Dyn3DGS [8]		84.29	79.04	74.82		87.83	82.14	79.32
Ours		100	100	98.83		100	100	100