3D object detection based on fusion of point cloud and image by mutual attention

Table 1. Detection parameters related to car objects
View table
View in Article
Table 1. Detection parameters related to car objects
Parameters Car
Range $[- 3,1] [- 40,40] [0,70.4]$
Voxel size $(v_{D}, v_{H}, v_{W})$ $[0.4,0.2,0.2]$
Anchor size $(l^{a}, w^{a}, h^{a})$ $[3.9 m, 1.6 m, 1.56 m]$
Centered at $z^{a}$ $- 1.0$ m
Two rotations 0 and 90 degrees

Table 2. Comparison of 3D Car detection performance on KITTI： average accuracy （AP）%

View table

View in Article

Table 2. Comparison of 3D Car detection performance on KITTI： average accuracy （AP）%

Method	Modality	Car （AP-0.7）
Method	Modality	Easy	Moderate	Hard
MV3D^［4］	Lidar+Img	71.29	62.68	56.56
PointFusion^［13］	Lidar+Img	77.92	63.00	53.27
VoxelNet^［10］	Lidar	81.97	65.46	62.85
F-PointNet^［16］	Lidar+Img	83.76	70.92	63.65
AVOD-FPN^［22］	Lidar+Img	84.41	74.44	68.65
ContFuse^［17］	Lidar+Img	86.32	73.25	67.81
PointRCNN^［11］	Lidar	86.75	76.05	74.30
PointPillars^［12］	Lidar	87.22	76.95	73.52
Painted VoxelNet^［18］	Lidar+Img	87.15	76.66	74.75
MANet（ours）	Lidar+Img	88.76	77.63	76.14

Tools

Get Citation

Copy Citation Text

Jun-ying CHEN, Tong-yao BAI, Liang ZHAO. 3D object detection based on fusion of point cloud and image by mutual attention[J]. Optics and Precision Engineering, 2021, 29(9): 2247

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites