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

Table 1

The performance on the LINEMOD dataset for objects pose estimation based on ADD (-S) scores.

Methods	Ape	Bench vise	Cam	Can	Cat	Driller	Duck	Egg box	Glue	Hole puncher	Iron	Lamp	Phone	Mean
BB8 [29]	40.40	91.80	55.70	64.10	62.60	74.70	44.30	57.80	41.20	67.20	84.70	76.50	54.00	62.70
SSD6D [1]	65.00	80.00	78.00	86.00	70.00	73.00	66.00	100	100	49.00	78.00	73.00	79.00	79.00
YOLO6D [3]	21.62	81.80	36.57	68.80	41.82	63.51	27.23	69.58	80.02	42.63	74.97	71.11	47.74	55.95
DPOD [31]	53.28	95.34	90.36	94.10	60.38	97.72	66.01	99.72	93.83	65.83	99.80	88.11	74.24	82.98
Pix2Pose [33]	58.10	91.00	60.90	84.40	65.00	73.60	43.80	96.80	79.40	74.80	83.40	82.00	45.00	72.40
CDPN [32]	64.38	97.77	91.67	95.87	83.83	96.23	66.76	99.72	99.61	85.82	97.85	97.86	90.75	89.86
PoseCNN [5]	27.80	68.90	47.50	71.40	56.70	65.40	42.80	98.30	95.60	50.90	65.60	70.30	54.60	62.70
PVNet [8]	43.62	99.90	86.86	95.47	79.34	96.43	52.58	99.15	95.66	81.92	98.88	99.33	92.41	86.27
DPVL [11]	69.05	100	94.12	98.52	83.13	99.01	63.47	100	97.97	88.20	99.90	99.81	96.35	91.50
PDAL + AFAM [12]	69.43	100	92.45	99.21	87.72	99.01	67.79	100	98.94	86.01	99.38	99.81	95.10	91.91
L+ [7]	65.34	100	92.65	97.84	90.22	97.72	62.54	99.72	95.56	88.97	99.30	99.53	95.87	91.18
Ours	76.27	100	96.80	99.38	87.85	99.40	71.42	100	99.68	94.72	100	99.92	98.64	94.16

Some objects like glue and egg box are symmetric objects. The bold values given in Table 1 indicate the high value among the compared methods for pose estimation on the LINEMOD dataset with respect to ADD (-S) metric.