An End-to-End Robotic Visual Localization Algorithm Based on Deep Learning

Table 3

Comparison of localization errors on the Microsoft 7-Scenes dataset.

Scene	PoseNet	Bayesian PoseNet	VidLoc	Ours (tra/m,ori/°)
Chess	0.32 m, 8.12°	0.37 m, 7.24°	0.18 m, N/A	0.14 m, 7.56°
Fire	0.47 m, 14.4°	0.43 m, 13.7°	0.26 m, N/A	0.22 m, 10.84°
Heads	0.29 m, 12.0°	0.31 m, 12.0°	0.14 m, N/A	0.18 m, 11.63°
Office	0.48 m, 7.68°	0.48 m, 8.04°	0.26 m, N/A	0.23 m, 6.23°
Pumpkin	0.47 m, 8.42°	0.61 m, 7.08°	0.36 m, N/A	0.25 m, 5.84°
Red Kitchen	0.59 m, 8.64°	0.58 m, 7.54°	0.31 m, N/A	0.38 m, 8.42°
Stairs	0.47 m, 13.8°	0.48 m, 13.1°	0.26 m, N/A	0.16 m, 11.23°
Average all	0.44 m, 10.4°	0.47 m, 9.81°	0.25 m, N/A	0.22 m, 8.77°