notesum.ai

Published at December 5

4Real-Video: Learning Generalizable Photo-Realistic 4D Video Diffusion

cs.CV

Released Date: December 5, 2024

Authors: Chaoyang Wang¹, Peiye Zhuang¹, Tuan Duc Ngo², Willi Menapace¹, Aliaksandr Siarohin¹, Michael Vasilkovsky¹, Ivan Skorokhodov¹, Sergey Tulyakov¹, Peter Wonka³, Hsin-Ying Lee¹

Aff.: ¹Snap Inc; ²Snap Inc, Umass Amherst; ³Snap Inc, KAUST

Arxiv: http://arxiv.org/pdf/2412.04462v1

Method	FID $\downarrow$	CLIP $\uparrow$	FVD $\downarrow$		FVD-Test $\downarrow$		Visual Quality $\uparrow$		Temporal Consist. $\uparrow$		Factual Consist. $\uparrow$
			Time	View	Time	View	Time	View	Time	View	Time	View
SV4D [44]	204.81	19.46	1053.10	1245.42	814.50	323.99	2.26	2.02	2.03	1.68	2.12	1.99
MotionCtrl [42]	87.10	20.20	1556.36	1509.76	1170.04	302.18	2.36	2.30	2.38	2.25	2.38	2.33
Sequential	96.64	28.16	1662.54	1797.15	897.08	597.19	2.30	2.28	2.21	2.15	2.23	2.20
Soft w/o Obj	80.17	28.11	1392.48	1720.47	318.18	302.18	2.41	2.39	2.37	2.31	2.35	2.33
Hard Sync	79.92	28.16	972.87	1045.35	316.14	251.44	2.42	2.40	2.40	2.33	2.37	2.34
Soft Sync	78.36	28.22	906.16	1036.00	308.15	261.02	2.43	2.42	2.41	2.36	2.38	2.36