notesum.ai

Published at December 3

How to Use Diffusion Priors under Sparse Views?

cs.CV

Released Date: December 3, 2024

Authors: Qisen Wang¹, Yifan Zhao¹, Jiawei Ma¹, Jia Li¹

Aff.: ¹State Key Laboratory of Virtual Reality Technology and Systems, SCSE Beihang University

Arxiv: http://arxiv.org/pdf/2412.02225v1

Refer to caption

Methods	Setting	LLFF [22]				DTU [54]
Methods	Setting	SSIM↑	LPIPS↓	PSNR↑	AVGE↓	SSIM↑	LPIPS↓	PSNR↑	AVGE↓
SRF [55]	Trained on DTU	0.250	0.591	12.34	0.313	0.671	0.304	15.32	0.171
PixelNeRF [49]		0.272	0.682	7.93	0.461	0.695	0.270	16.82	0.147
MVSNeRF [56]		0.557	0.356	17.25	0.171	0.769	0.197	18.63	0.113
SRF ft. [55]	Fine-tuned per Scene	0.436	0.529	17.07	0.203	0.698	0.281	15.68	0.162
PixelNeRF ft. [49]		0.438	0.512	16.17	0.217	0.710	0.269	18.95	0.125
MVSNeRF ft. [56]		0.584	0.327	17.88	0.157	0.769	0.197	18.54	0.113
Mip-NeRF [57]	Based on NeRF Optimized per Scene	0.351	0.495	14.62	0.246	0.571	0.353	8.68	0.323
DietNeRF [10]		0.370	0.496	14.94	0.240	0.633	0.314	11.85	0.243
RegNeRF [7]		0.587	0.336	19.08	0.149	0.745	0.190	18.89	0.112
FreeNeRF [8]		0.612	0.308	19.63	0.134	0.787	0.182	19.92	0.098
SparseNeRF [12]		0.624	0.328	19.86	0.127	0.769	0.201	19.55	0.102
3DGS [2]	Based on 3DGS Optimized per Scene	0.456	0.385	14.97	0.208	0.795	0.178	15.06	0.136
FSGS [58]		0.682	0.248	20.43	0.108	0.825	0.145	17.69	0.101
DNGaussian [9]		0.591	0.294	19.12	0.132	0.790	0.176	18.91	0.102
DNGaussian †[9]		0.687	0.228	19.94	0.109	-	-	-	-
Ours		0.702	0.207	20.44	0.101	0.856	0.121	19.99	0.077
Ours		±0.001	±0.001	±0.08	±0.001	±0.001	±0.001	±0.10	±0.001