DGPoint: A Dynamic Graph Convolution Network for Point Cloud Semantic Segmentation

Fig. 1. Network architecture: four local feature encoding blocks (dotted box above, with detail structure of a single local feature encoding block shown in lower left dotted box) are assembled recursively to encode local features, then the four outputs (lower right solid box) are concatenated to be input to decoding blocks

Download full size

Fig. 2. Example of edge convolution operation

Download full size

Fig. 3. S3DIS data set

Download full size

Fig. 4. Accuracy (dotted line) and loss function (solid line) curves

Download full size

Fig. 5. Segmentation accuracy of S3DIS data set for different k

Download full size

Fig. 6. Sample semantic segmentation results of S3DIS data set. (a) Ground truth; (b) segmentation result of PointNet++; (c) segmentation result of DGPoint

Download full size

Table 1. Comparison of semantic segmentation accuracy on entire S3DIS data set
View table
Table 1. Comparison of semantic segmentation accuracy on entire S3DIS data set
Method mIoU /% oAcc /%
DGPoint 68.3 86.2
PointNet^［8］ 47.6 78.6
PointNet++^［9］ 50.8 82.2
DGCNN^［28］ 56.1 84.1
ShellNet^［26］ 66.8 --
KVGCN^［33］ 60.9 87.4

Table 2. IoU of some semantic class in S3DIS data set

View table

Table 2. IoU of some semantic class in S3DIS data set

Method	Celling	Floor	Wall	Door	Table	Chair	Bookcase	Clutter
DGPoint	97.8	98.6	78.7	62.6	76.9	73.3	30.3	55.9
PointNet^［8］	88.8	97.3	69.8	10.8	58.9	52.6	40.3	33.2
PointNet++^［9］	93.1	97.6	77.2	58.1	62.3	65.7	45.2	50.4
SegCloud^［20］	90.1	96.1	69.9	23.1	70.4	78.6	58.4	41.6
SPG^［29］	89.9	95.1	72.0	60.0	69.2	73.5	3.2	52.9
KVGCN^［33］	94.5	94.1	79.5	63.2	64.3	67.5	23.6	53.2

Tools

Get Citation

Copy Citation Text

Youqun Liu, Jianfeng Ao, Zhongtai Pan. DGPoint: A Dynamic Graph Convolution Network for Point Cloud Semantic Segmentation[J]. Laser & Optoelectronics Progress, 2022, 59(16): 1610014

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites