notesum.ai

Published at December 4

Feed-Forward Bullet-Time Reconstruction of Dynamic Scenes from Monocular Videos

cs.CV

cs.AI

cs.GR

Released Date: December 4, 2024

Authors: Hanxue Liang¹, Jiawei Ren¹, Ashkan Mirzaei¹, Antonio Torralba², Ziwei Liu³, Igor Gilitschenski⁴, Sanja Fidler⁴, Cengiz Oztireli⁵, Huan Ling⁴, Zan Gojcic¹, Jiahui Huang¹

Aff.: ¹NVIDIA; ²MIT; ³Nanyang Technological University; ⁴University of Toronto; ⁵University of Cambridge

Arxiv: http://arxiv.org/pdf/2412.03526v1

Model	Rec. Time	Render FPS	PSNR $\uparrow$	LPIPS $\downarrow$
HyperNeRF [48]	\qty64	0.40	17.60	0.367
DynNeRF [21]	\qty74	0.05	26.10	0.082
NSFF [36]	\qty223	0.16	24.33	0.199
RoDynRF[41]	\qty28	0.42	25.89	0.065
MonoNeRF[59]	\qty22	0.05	25.62	0.106
4D-GS [67]	\qty1.2	44	21.45	0.199
Casual-FVS [33]	\qty0.25	48	24.57	0.081
PGDVS [81]	\qty3^†	0.70	24.41	0.186
Depth Warp	–	–	12.63	0.564
BTimer (Ours)	\qty0.78	115	25.82	0.086