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

Table 3

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

Methods	Ape	Bench vise	Cam	Can	Cat	Driller	Duck	Egg box	Glue	Hole puncher	Iron	Lamp	Phone	Mean
BB8 [29]	96.60	99.10	86.00	91.20	98.80	80.90	92.20	91.00	92.30	95.30	84.80	75.80	85.30	89.30
YOLO6D [3]	92.10	95.06	93.24	97.44	97.41	79.41	94.65	90.33	96.53	92.86	82.94	76.87	86.07	90.37
CDPN [32]	96.86	98.35	98.73	99.41	99.8	95.34	98.59	98.97	99.23	99.71	97.24	95.49	97.64	98.10
PoseCNN [5]	83.00	50.00	71.90	69.80	92.00	43.60	91.80	91.10	88.00	82.10	41.80	48.40	58.80	70.20
PVNet [8]	99.23	99.81	99.21	99.90	99.30	96.92	98.02	99.34	98.45	100	99.18	98.27	99.42	99.00
DPVL [11]	99.04	99.71	99.41	100	99.70	98.12	99.06	99.43	99.51	100	99.69	99.14	99.42	99.40
L+ [7]	99.05	99.71	99.61	99.71	99.81	98.62	98.97	99.44	99.23	99.91	99.80	98.28	99.52	99.00
CSA6D [40]	98.60	95.80	98.80	97.40	99.50	95.10	98.40	99.90	99.90	98.20	97.80	95.50	97.60	98.10
Ours	99.42	99.83	99.55	100	99.86	98.83	99.59	99.84	99.86	100	99.91	99.53	99.65	99.68

The bold values given in this Table show the highest value(s) of performance on the LINEMOD dataset for objects pose estimation based on 2D projection errors.