notesum.ai

Published at December 10

ResGS: Residual Densification of 3D Gaussian for Efficient Detail Recovery

cs.CV

Released Date: December 10, 2024

Authors: Yanzhe Lyu¹, Kai Cheng¹, Xin Kang¹, Xuejin Chen¹

Aff.: ¹University of Science and Technology of China

Arxiv: http://arxiv.org/pdf/2412.07494v1

Refer to caption

Dataset

Mip-NeRF360

Tanks&Temples

Deep Blending

Method

Metrics

PSNR

\uparrow

SSIM

\uparrow

LPIPS

\downarrow

Mem.

PSNR

\uparrow

SSIM

\uparrow

LPIPS

\downarrow

Mem.

PSNR

\uparrow

SSIM

\uparrow

LPIPS

\downarrow

Mem.

Plenoxels

23.08

0.626

0.463

2.1GB

21.08

0.719

0.379

2.3GB

23.06

0.795

0.510

2.7GB

Instant-NGP

25.59

0.699

0.331

48MB

21.92

0.745

0.305

48MB

24.96

0.817

0.390

48MB

Mip-NeRF360

27.69

0.792

0.237

8.6MB

22.22

0.759

0.257

8.6MB

29.40

0.901

0.245

8.6MB

Scaffold-GS^∗

27.69

0.812

0.225

176MB

23.96

0.853

0.177

87MB

30.21

0.906

0.254

66MB

Octree-GS^∗

27.52

0.812

0.218

124MB

24.52

0.866

0.153

84MB

30.41

0.913

0.238

93MB

3D-GS

27.21

0.815

0.214

734MB

23.14

0.841

0.183

411MB

29.41

0.903

0.243

676MB

AbsGS

27.49

0.820

0.191

728MB

23.73

0.853

0.162

304MB

29.67

0.902

0.236

444MB

Pixel-GS^∗

27.52

0.822

0.191

1.32GB

23.78

0.853

0.151

1.06GB

28.91

0.892

0.250

1.09GB

FreGS

27.85

0.826

0.209

-

23.96

0.849

0.178

-

29.93

0.904

0.240

-

Mini-Splatting-D

27.51

0.831

0.176

1.11GB

23.23

0.853

0.140

1.01GB

29.88

0.906

0.211

1.09GB

Ours-Small (AbsGS)

27.94

0.830

0.191

342MB

24.33

0.862

0.150

187MB

30.01

0.906

0.234

289MB

Ours (3D-GS)

28.00

0.831

0.187

600MB

24.26

0.865

0.141

354MB

29.93

0.903

0.232

596MB

Ours (AbsGS)

28.00

0.833

0.174

698MB

24.38

0.867

0.132

351MB

29.91

0.902

0.227

586MB