Infrared and Laser Engineering, Volume. 51, Issue 8, 20210702(2022)
Object point cloud classification and segmentation based on semantic information compensating global features
Figures & Tables(23)
Fig. 17. Segmentation viewable under different point set densities
Table 1. Experimental platform configuration
View tableView in ArticleTable 1. Experimental platform configuration
Environment configuration Model parameters Name Configuration Name Value CPU Intel i7-10700 F Batch size 32 GPU RTX3090 Number point 1024 RAM 32 G Max epoch 250 Operation system Ubuntu18.04 Optimizer Adam Language Python 3.7 Learning rate 0.001 Learning framework TensorFlow GPU 1.15.0 Momentum 0.9
Table 2. Experimental comparison of various classification algorithms of ModelNet40[13]
View tableView in ArticleTable 2. Experimental comparison of various classification algorithms of ModelNet40[13]
Method Representation Input Eval accuracy Avg class acc PointNet[ 5 ]Points 1024×3 89.2% 86.2% PointNet++[ 6 ]Points(+normal) 1024×(3+3) 90.7% 87.8% DGCNN[ 8 ]Points 1024×3 92.2% 88.9% PointCNN[ 7 ]Points 1024×3 91.7% 88.5% Pointwise[ 15 ]Points 1024×3 91.6% 89.1% SRN-PointNet[ 16 ]Points 1024×3 91.5% 88.6% GGM[ 17 ]Points 1024×3 92.5% 89.0% MSDGCNN[ 18 ]Points 1024×3 91.8% 88.3% EllNet[ 19 ]Points 1024×2 92.6% 89.0% Ours Points 1024×3 92.7% 89.3%
Table 3. Comparison of algorithm test time
View tableView in ArticleTable 3. Comparison of algorithm test time
Method Size/MB Time/s Accuracy PointNet[ 5 ]40 78.9 89.2% PointNet++[ 6 ]12 163.2 90.7% DGCNN[ 8 ]21 89.7 92.2% PointCNN[ 7 ]94 117.0 91.7% Ours 21 86.4 92.7%
Table 4. Local segmentation test results of each algorithm on the ShapeNet Part[14] data set
View tableView in ArticleTable 4. Local segmentation test results of each algorithm on the ShapeNet Part[14] data set
Method mIoU Shapes IoU Plane Bag Cap Car Chair Earcup Guitar Knife Lamp Laptop Motor Mug Pistol Rocket Skate Table PointNet[ 5 ]83.7% 83.4% 78.7% 82.5% 74.9% 89.6% 73.0% 91.5% 85.9% 80.8% 95.3% 65.2% 93.0% 81.2% 57.9% 72.8% 80.6% PointNet[ 6 ]85.1% 82.4% 79.0% 87.7% 77.3% 90.8% 71.8% 91.0% 85.9% 83.7% 95.3% 71.6% 94.1% 81.3% 58.7% 76.4% 82.6% DGCNN[ 8 ]85.2% 84.0% 83.4% 86.7% 77.8% 90.6% 74.7% 91.2% 87.5% 82.8% 95.7% 66.3% 94.9% 81.1% 63.5% 74.5% 82.6% PointCNN[ 7 ]86.1% 84.1% 86.4% 86.0% 80.8% 90.6% 79.7% 92.3% 88.4% 85.3% 96.1% 77.2% 95.3% 84.8% 64.2% 80.0% 83.0% Pointwise[ 15 ]85.1% 82.9% 80.7% 87.8% 76.6% 90.8% 79.2% 91.0% 86.6% 83.3% 95.3% 71.9% 94.4% 80.9% 62.0% 75.1% 82.5% SRNPNet[ 16 ]85.3% 82.4% 79.8% 88.1% 77.9% 90.7% 69.6% 90.9% 86.3% 84.0% 95.4% 72.2% 94.9% 81.3% 62.1% 75.9% 83.2% GMM[ 17 ]85.2% 83.9% 82.8% 88.0% 79.8% 90.7% 76.8% 91.3% 87.6% 82.6% 95.5% 66.6% 94.8% 81.8% 62.6% 73.8% 82.6% MSDGCN[ 18 ]85.4% 83.7% 84.7% 87.5% 77.0% 90.8% 68.2% 91.5% 86.5% 96.0% 95.5% 72.0% 95.1% 83.4% 61.9% 77.4% 82.9% Ell-Net[ 19 ]85.0% 82.8% 81.5% 87.6% 76.8% 90.6% 78.8% 90.8% 86.8% 86.9% 95.1% 71.8% 94.2% 80.8% 61.8% 75.0% 82.2% Ours 85.5% 84.3% 85.8% 88.1% 80.0% 90.8% 79.5% 91.6% 88.2% 91.6% 95.8% 76.7% 96.1% 82.6% 65.6% 81.2% 82.8%
Table 5. Analysis of the best
\begin{document}$ K $\end{document} View tableView in ArticleTable 5. Analysis of the best
\begin{document}$ K $\end{document} KNN point number Avg class acc Eval accuracy 16 89.5% 91.4% 20 89.8% 92.7% 25 89.1% 91.3% 30 88.6% 91.1% 32 88.5% 90.8%
Table 6. Analysis of module impact
View tableView in ArticleTable 6. Analysis of module impact
Dilated-Edge conv KNN-VLAD Avg class acc √ × 92.0% × √ 90.8% × √ 92.7%
Tools
Get Citation
Copy Citation Text
Sen Lin, Zhenyu Zhao, Xiaokui Ren, Zhiyong Tao. Object point cloud classification and segmentation based on semantic information compensating global features[J]. Infrared and Laser Engineering, 2022, 51(8): 20210702
Paper Information
Category: Image processing
Received: Jan. 20, 2022
Accepted: --
Published Online: Jan. 9, 2023
The Author Email: Zhao Zhenyu (610685324@qq.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20