A Robust Convolutional Neural Network for 6D Object Pose Estimation from RGB Image with Distance Regularization Voting Loss

Table 4

The performance on the occlusion LINEMOD dataset for objects pose estimation based on 2D projection errors.

Methods	Ape	Can	Cat	Driller	Duck	Egg box	Glue	Hole puncher	Mean
PVNet [8]	69.14	86.09	65.12	73.06	61.44	8.43	55.37	69.84	61.06
L+ [7]	67.61	86.75	62.85	79.91	64.07	3.75	60.47	80.25	63.21
Ours	68.88	87.02	64.35	82.75	66.00	7.27	60.89	83.06	65.02

These values indicate the highest value of performance on the occlusion LINEMOD dataset for objects pose estimation based on 2D projection errors.