Object 6-DoF pose estimation using auxiliary learning

Minjia CHEN; Shaoyan GAI; Feipeng DA; Jian YU

doi:10.37188/OPE.20243206.0901

Optics and Precision Engineering, Volume. 32, Issue 6, 901(2024)

Object 6-DoF pose estimation using auxiliary learning

Minjia CHEN1...2, Shaoyan GAI1,2,*, Feipeng DA1,2, and Jian YU1,23,* |Show fewer author(s)

Author Affiliations

¹School of Automation， Southeast University， Nanjing20096， China

²Key Laboratory of Measurement and Control of Complex Engineering Systems， Ministry of Education， Southeast University， Nanjing10096， China

³Key Laboratory of Space Photoelectric Detection and Perception， Ministry of Industry and Information Technology， Nanjing University of Aeronautics and Astronautics， Nanjing211106， China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(16)

Fig. 1. Network architecture of object 6-DoF pose estimation based on auxiliary learning

Download full size

View in Article

Fig. 2. Dual-branch for point cloud registration

Download full size

View in Article

Fig. 3. Local feature extractor

Download full size

View in Article

Fig. 4. Spatial information encoder

Download full size

View in Article

Fig. 5. Pipeline of fine pose estimation

Download full size

View in Article

Fig. 6. Illustration of point pair feature

Download full size

View in Article

Fig. 7. Qualitative results of pose estimation on YCB-Video Dataset

Download full size

View in Article

Fig. 8. Qualitative results of pose estimation on LineMOD Dataset

Download full size

View in Article

Table 1. Quantitative evaluation results on YCB-Video Dataset

View table

View in Article

Table 1. Quantitative evaluation results on YCB-Video Dataset

	DenseFusion^［12］		REDE^［15］		PR-GCN^［23］		BiCo-Net^［16］		DCL-Net^［24］		Ours
	AUC	<2 cm	AUC	<2 cm	AUC	<2 cm	AUC	<2 cm	AUC	<2 cm	AUC	<2 cm
002 master chef can	96.4	100.0	95.1	100.0	97.1	100.0	96.2	100.0	96.1	100.0	96.3	100.0
003 cracker box	95.5	99.5	96.3	99.7	97.6	100.0	96.6	100.0	96.4	99.4	96.3	100.0
004 sugar box	97.5	100.0	97.4	100.0	98.3	100.0	97.8	100.0	98.1	100.0	97.8	100.0
005 tomato soup can	94.6	96.9	96.9	100.0	95.3	97.6	95.7	98.1	95.8	97.7	95.7	98.2
006 mustard bottle	97.2	100.0	96.7	100.0	97.9	100.0	98.0	100.0	98.7	100.0	97.8	100.0
007 tuna fish can	96.6	100.0	96.6	100.0	97.6	100.0	96.5	100.0	97.4	100.0	96.6	100.0
008 pudding box	96.5	100.0	96.4	100.0	98.4	100.0	97.5	100.0	98.2	100.0	97.7	100.0
009 gelatin box	98.1	100.0	97.8	100.0	96.2	94.4	98.8	100.0	98.9	100.0	98.8	100.0
010 potted meat can	91.3	93.1	92.0	94.2	96.6	99.1	93.0	94.5	93.1	94.7	93.0	94.7
011 banana	96.6	100.0	97.0	99.7	98.5	100.0	97.1	100.0	98.1	100.0	97.5	100.0
019 pitcher base	97.1	100.0	97.5	100.0	98.1	100.0	97.6	100.0	98.0	99.8	97.5	100.0
021 bleach cleanser	95.8	100.0	94.2	100.0	97.9	100.0	96.6	100.0	97.0	100.0	96.2	100.0
024 bowl	88.2	98.8	96.7	99.3	90.3	96.6	96.7	100.0	97.3	100.0	96.5	100.0
025 mug	97.1	100.0	97.0	100.0	98.1	100.0	97.0	100.0	97.8	100.0	96.7	100.0
035 power drill	96.0	98.7	97.0	99.6	98.1	100.0	97.0	99.9	98.0	100.0	96.9	100.0
036 wood block	89.7	94.6	91.0	98.3	96.0	100.0	92.1	90.1	93.9	97.5	94.1	95.0
037 scissors	95.2	100.0	94.5	100.0	96.7	100.0	92.2	99.5	87.6	98.3	91.1	99.8
040 large marker	97.5	100.0	97.8	100.0	97.9	100.0	97.4	100.0	97.8	99.8	97.4	100.0
051 large clamp	72.9	79.2	77.3	80.7	87.5	93.3	94.7	98.3	95.7	98.6	94.9	98.6
052 extra large clamp	69.8	76.3	85.9	82.0	79.7	84.6	88.2	90.2	88.8	87.2	88.3	90.8
061 foam brick	92.5	100.0	94.6	100.0	97.8	100.0	97.2	100.0	97.5	100.0	96.1	100.0
MEAN	93.1	96.8	94.5	97.8	95.8	98.5	96.0	98.8	96.6	99.0	95.9	99.0

Table 2. Quantitative evaluation results on LineMOD Dataset

View table

View in Article

Table 2. Quantitative evaluation results on LineMOD Dataset

	DenseFusion^［12］	REDE^［15］	PR-GCN^［23］	BiCo-Net^［16］	DCL-Net^［24］	Ours
ape	92.3	95.6	97.6	97.3	97.4	97.6
ben.	93.2	99.4	99.2	98.8	99.4	99.1
cam	94.4	99.6	99.4	99.6	99.8	99.8
can	93.1	99.5	98.4	99.3	99.9	99.7
cat	96.5	99.5	98.7	100.0	100.0	99.9
drill.	87.0	99.3	98.8	98.9	99.9	99.2
duck	92.3	97.0	98.9	98.7	98.4	98.8
egg.	99.8	100.0	99.9	99.8	100.0	99.8
glue	100.0	99.9	100.0	99.8	99.9	99.0
hole.	86.9	98.6	99.4	99.2	100.0	99.0
iron	97.0	99.3	98.5	100.0	100.0	99.7
lamp	95.3	99.3	99.2	99.7	99.5	99.7
phone	92.8	99.3	98.4	99.2	99.7	99.1
MEAN	94.3	98.9	98.9	99.3	99.5	99.4

Table 3. Quantitative evaluation results on LM-O Dataset

View table

View in Article

Table 3. Quantitative evaluation results on LM-O Dataset

	PVN3D^［13］	PR-GCN^［23］	BiCo-Net^［16］	DCL-Net^［24］	Ours
MEAN	63.2	65.0	69.5	70.6	71.3
ape	33.9	40.2	55.6	56.7	52.6
can	88.6	76.2	83.2	80.2	87.1
cat	39.1	57.0	47.3	48.1	45.8
drill.	78.4	82.3	69.9	81.4	71.4
duck	41.9	30.0	58.3	44.6	56.3
egg.	80.9	68.2	78.1	83.6	86.1
glue	68.1	67.0	76.9	79.1	83.0
hole.	74.7	97.2	87.2	91.3	88.5

Table 4. Comparison on BOP Challenge 2022
View table
View in Article
Table 4. Comparison on BOP Challenge 2022
Input Method YCB-Video LM-O
AUC <2 cm ADD（-S）
RGB CosyPose^［28］ 89.8 - -
GDR-Net^［25］ 91.6 60.1 62.2
PFA^［26］ - 62.8 64.1
ZebraPose^［27］ 90.1 80.5 76.9
RGB-D RCVPose3D^［29］ 96.6 - 74.5
Ours 95.9 99.0 71.3

Table 5. Comparison of inference time on LineMOD dataset
View table
View in Article
Table 5. Comparison of inference time on LineMOD dataset
Seg PE Refine ALL
DenseFusion^［12］ 30 20 10 60
PR-GCN^［23］ 30 8 30 68
BiCo-Net^［16］ 30 45 - 75
Ours 30 39 - 69

Table 6. Comparison of model parameters
View table
View in Article
Table 6. Comparison of model parameters
Main model Refiner ALL
DCL-Net^［24］ 50.3 10.5 60.8
BiCo-Net^［16］ 191.1 - 191.1
Ours 41.9 - 41.9

Table 7. Experiment results of different branches
View table
View in Article
Table 7. Experiment results of different branches
Pose Reg C2M M2C AL-Net results
√ √ × × 66.3
√ × √ × 65.9
√ √ √ × 68.8
× √ √ √ 71.3

Table 8. Experiment results of auxiliary learning network
View table
View in Article
Table 8. Experiment results of auxiliary learning network
LFE SIE Coarse C2F results
√ × - √ 70.5
× √ - √ 70.4
√ √ √ - 69.8
√ √ - √ 71.3

Tools

Get Citation

Copy Citation Text

Minjia CHEN, Shaoyan GAI, Feipeng DA, Jian YU. Object 6-DoF pose estimation using auxiliary learning[J]. Optics and Precision Engineering, 2024, 32(6): 901

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites