Multi-Modal 3D Object Detection Algorithm Based on Kolmogorov-Arnold Network

Fig. 6. Visualization of 3D object detection results by different methods on KITTI test set. (a) SECOND^[15]; (b) Part-A2^[20]; (c) PVRCNN^[24]; (d) proposed method

Download full size

Fig. 7. Visualization of 3D object detection results in BEV space of colored point clouds

Download full size

Table 1. 3D object detection results for the car category based on the AP metric on the KITTI test set

View table

Table 1. 3D object detection results for the car category based on the AP metric on the KITTI test set

Method	Speed /Hz	mAP of BEV /%	AP_BEV（R_IoU=0.7） /%			mAP of 3D /%	AP_3D（R_IoU=0.7） /%
Method	Speed /Hz	mAP of BEV /%	Easy	Moderate	Hard	mAP of 3D /%	Easy	Moderate	Hard
MV3D^［10］	2.8	80.10	86.35	77.50	76.45	63.86	71.65	63.02	56.90
VoxelNet^［14］	4.4	84.48	89.71	84.93	78.82	71.32	82.15	65.75	63.10
F-PointNet^［26］	5.9	83.01	88.40	84.15	76.55	73.05	84.01	71.25	63.91
SECOND^［15］	20.0	85.68	90.10	87.20	79.75	77.97	87.65	76.82	69.45
P-RCNN^［19］	10.0	85.33	90.03	87.52	78.43	80.58	88.32	77.92	75.51
MVXNet^［17］	13.1	84.62	89.65	85.02	79.20	76.06	85.51	74.23	68.43
Part-A2^［20］	12.5	88.04	90.34	87.75	86.02	81.25	88.65	78.43	76.68
IA-SSD^［38］	83.0	88.53	90.34	88.73	86.52	80.56	88.76	79.52	73.40
SeSame^［39］	17.9	87.16	90.73	87.53	83.22	77.88	85.26	76.86	71.55
Proposed	14.3	89.46	94.47	88.12	85.78	81.72	89.85	78.93	76.37

Table 2. Detection results for the car category based on the AP metric on the nuScenes data set
View table
Table 2. Detection results for the car category based on the AP metric on the nuScenes data set
Method AP@0.5 AP@1.0 AP@2.0 AP@4.0
SECOND^［15］ 62.47 73.87 76.83 78.92
PointPillars^［40］ 66.65 78.58 83.41 85.52
Proposed 73.25 83.32 85.34 88.59

Table 3. Ablation experiment results for different modules on the KITTI validation set

View table

Table 3. Ablation experiment results for different modules on the KITTI validation set

Module	Car（R_IoU=0.7）			Cyclist（R_IoU=0.5）			Pedestrian（R_IoU=0.5）
Module	Easy	Moderate	Hard	Easy	Moderate	Hard	Easy	Moderate	Hard
None	87.26	77.18	73.56	72.14	54.18	50.35	58.19	53.43	50.20
M1	87.82	77.81	73.77	75.75	56.86	53.12	62.38	56.29	52.93
M1+M2	89.74	78.79	76.32	79.97	58.34	55.92	64.52	59.14	55.06

Table 4. Comparison results for original point clouds and colored point clouds on the KITTI validation set
View table
Table 4. Comparison results for original point clouds and colored point clouds on the KITTI validation set
Point cloud mAP of BEV AP_BEV（R_IoU=0.7） mAP of 3D AP_3D（R_IoU=0.7）
Easy Moderate Hard Easy Moderate Hard
OPC 87.52 91.97 85.59 85.01 79.23 87.76 76.31 73.62
CPC 89.45 94.51 88.15 85.81 81.62 89.74 78.79 76.32

Table 5. Comparison of 2D projection results for car category on the KITTI test set based on the AP metric with other advanced methods

View table

Table 5. Comparison of 2D projection results for car category on the KITTI test set based on the AP metric with other advanced methods

Method	mAP of 2D	AP_2D（R_IoU=0.7）
Method	mAP of 2D	Easy	Moderate	Hard
SECOND^［15］	89.82	90.84	89.81	88.80
P-RCNN^［19］	91.36	95.62	89.36	89.10
MVXNet^［17］	93.23	96.33	92.95	90.40
Part-A2^［20］	89.72	90.69	89.39	89.07
Proposed	94.28	98.11	92.69	92.03

Tools

Get Citation

Copy Citation Text

Yanwu Ling, Junmin Rao, Yan Li, Fanming Li. Multi-Modal 3D Object Detection Algorithm Based on Kolmogorov-Arnold Network[J]. Laser & Optoelectronics Progress, 2025, 62(16): 1615007

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Machine Vision

Received: Jan. 20, 2025

Accepted: Mar. 14, 2025

Published Online: Aug. 11, 2025

The Author Email: Fanming Li (lfmjws@163.com)

DOI:10.3788/LOP250553

CSTR:32186.14.LOP250553

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. 3D object detection results for the car category based on the AP metric on the KITTI test set

Table 1. 3D object detection results for the car category based on the AP metric on the KITTI test set

Table 2. Detection results for the car category based on the AP metric on the nuScenes data set

Table 2. Detection results for the car category based on the AP metric on the nuScenes data set

Table 3. Ablation experiment results for different modules on the KITTI validation set

Table 3. Ablation experiment results for different modules on the KITTI validation set

Table 4. Comparison results for original point clouds and colored point clouds on the KITTI validation set

Table 4. Comparison results for original point clouds and colored point clouds on the KITTI validation set

Table 5. Comparison of 2D projection results for car category on the KITTI test set based on the AP metric with other advanced methods

Table 5. Comparison of 2D projection results for car category on the KITTI test set based on the AP metric with other advanced methods