notesum.ai

Published at December 4

NVComposer: Boosting Generative Novel View Synthesis with Multiple Sparse and Unposed Images

cs.CV

Released Date: December 4, 2024

Authors: Lingen Li¹, Zhaoyang Zhang², Yaowei Li³, Jiale Xu, Xiaoyu Li, Wenbo Hu, Weihao Cheng, Jinwei Gu, Tianfan Xue, Ying Shan

Aff.: ¹The Chinese University of Hong Kong; ²ARC Lab, Tencent PCG; ³Peking University

Arxiv: http://arxiv.org/pdf/2412.03517v1

[Uncaptioned image]

Model	Views	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	DISTS $\downarrow$
MotionCtrl [36]	1	13.4003	0.5539	0.4004	0.1396
CameraCtrl [9]	1	12.2995	0.4692	0.4337	0.1829
DUSt3R [34]	1	11.7650	0.4652	0.4900	0.2295
	2	14.6660	0.5158	0.4531	0.2104
	3	13.9156	0.5010	0.4699	0.2127
	4	14.8716	0.5193	0.4478	0.2072
ViewCrafter [43]	1	15.5625	0.4932	0.4122	0.2125
	2	15.6875	0.4775	0.4417	0.2212
	3	14.8593	0.4670	0.4617	0.2273
	4	15.0625	0.4712	0.4549	0.2301
NVComposer (Ours)	1	15.3101	0.6056	0.3445	0.1516
	2	16.9100	0.6445	0.2742	0.1198
	3	17.3115	0.6687	0.2558	0.1122
	4	17.9248	0.6958	0.2277	0.1023