Table 3 Generalization to different sparsity levels on the NYUv2 dataset. All methods have a single model trained with 500 points. Bold type indicates the best performance and underline indicates the second best performance.

From: RGB-conditioned frequency domain refinement for sparse-to-dense depth completion

Method

500-Points

200-Points

100-Points

50-Points

RMSE

REL

RMSE

REL

RMSE

REL

RMSE

REL

SpAgNet47

0.114

0.015

0.155

0.024

0.209

0.038

0.272

0.058

NLSPN48

0.092

0.012

0.136

0.019

0.245

0.037

0.431

0.081

CFormer49

0.091

0.012

0.141

0.021

0.429

0.092

0.707

0.181

OGNI-DC50

0.087

0.011

0.124

0.018

0.157

0.025

0.207

0.038

Ours

0.085

0.011

0.120

0.018

0.142

0.024

0.233

0.047

Back to article page