Progress in Novel View Synthesis Using Neural Radiance Fields

Gaoxiang He; Bin Zhu; Bo Xie; Yi Chen

doi:10.3788/LOP231578

Laser & Optoelectronics Progress, Volume. 61, Issue 12, 1200005(2024)

Progress in Novel View Synthesis Using Neural Radiance Fields

Gaoxiang He^1,2, Bin Zhu^1,2、*, Bo Xie^1,2, and Yi Chen^1,2

¹State Key Laboratory of Pulsed Power Laser Technology, College of Electronic Engineering, National University of Defense Technology, Hefei 230037, Anhui , China

²Infrared and Low Temperature Plasma Key Laboratory of Anhui Province, Hefei 230037, Anhui , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Schematic of novel view synthesis^[7]

Download full size

Fig. 2. Overview of novel view synthesis based on NeRF

Download full size

Fig. 3. Pipeline of NeRF^[15]. (a) Generating sampling points from rays; (b) calculating the color and density of sampling points; (c) calculating ray color through volume rendering; (d) optimizing radiation field

Download full size

Fig. 4. Qualitative comparison of partial models on 360v2 dataset^[44]

Download full size

Fig. 5. Comparison of Ha-NeRF and NeRF-W on Phototourism dataset^[48]

Download full size

Fig. 6. Train pipeline of the MVSNeRF^[55]

Download full size

Fig. 7. Comparison of FreeNeRF and NeRF on multiple datasets^[65]

Download full size

Fig. 8. Algorithm flowchart of Plenoxels model^[39]

Download full size

Fig. 9. Effect showcase of Instant NGP on Lego scene^[46]

Download full size

Table 1. Quantitative comparison of related models on 360v2 dataset

View table

Table 1. Quantitative comparison of related models on 360v2 dataset

Model	PSNR	SSIM	LPIPS
NeRF^［15］	24.85	0.659	0.426
Mip-NeRF^［25］	24.04	0.616	0.441
Plenoxels^［39］	24.67	0.649	0.417
DVGOv2^［38］	25.42	0.695	0.429
NeRF++^［16］	26.21	0.729	0.348
Mip-NeRF360^［34］	27.69	0.792	0.237
Instant NGP^［46］	25.68	0.705	0.302
NeRFAcc^［45］	27.58		0.292
Zip-NeRF^［44］	28.54	0.828	0.189

Table 2. Quantitative comparison of sparse view models on DTU dataset

View table

Table 2. Quantitative comparison of sparse view models on DTU dataset

Method	PSNR	SSIM	LPIPS
PixelNeRF^［18］	18.95	0.710	0.269
MVSNeRF^［55］	18.54	0.769	0.197
DS-NeRF^［58］	16.90	0.570	0.450
SparseNeRF^［59］	19.47	0.829	0.183
DietNeRF^［61］	11.85	0.633	0.314
RegNeRF^［62］	18.89	0.745	0.190
InfoNeRF^［63］	11.23	0.445	0.543
FreeNeRF^［65］	19.92	0.787

Table 3. Quantitative comparison of partial models on the NeRF-Synthetic dataset

View table

Table 3. Quantitative comparison of partial models on the NeRF-Synthetic dataset

Model	PSNR	SSIM	LPIPS	Speed /（frame·s^-1）	Training time
NeRF^［15］	31.01	0.947	0.081	0.023	56 h
NSVF^［42］	31.75	0.953	0.047	0.815	100 h
PlenOctrees^［68］	31.71	0.958	0.053	167.68	58 h
DVGO^［69］	31.95	0.957	0.053		14.2 min
DVGOv2^［38］	32.76	0.962	0.046		6 min
Plenoxels^［39］	31.71	0.958	0.049		11 min
ReLU fields^［70］	30.04		0.05		10 min
TensoRF^［72］	33.14	0.963	0.047		17.6 min
PlenVDB^［41］	31.90			20.75	12.4 min
EfficientNeRF^［71］	31.68	0.954	0.028	238.46	6 h
Instant NGP^［46］	32.11	0.961	0.053		5 min
NeRFAcc^［45］	33.11		0.053		4.5 min

Tools

Get Citation

Copy Citation Text

Gaoxiang He, Bin Zhu, Bo Xie, Yi Chen. Progress in Novel View Synthesis Using Neural Radiance Fields[J]. Laser & Optoelectronics Progress, 2024, 61(12): 1200005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites