Augmented Edge Graph Convolutional Networks for Semantic Segmentation of 3D Point Clouds

Method	OA	mIoU	ceiling	floor	wall	beam	column	window	door	table	chair	sofa	book	board	clut
PointNet^［10］	78.6	47.6	88.0	88.7	69.3	42.4	23.1	47.5	51.6	54.1	42.0	9.6	38.2	29.4	35.2
PointNet++^［11］		53.2	90.2	91.7	73.1	42.7	21.2	49.7	42.3	62.7	59.0	19.6	45.8	48.2	45.6
DeepGCNs^［22］	85.9	60.0	93.1	95.3	78.2	33.9	37.4	56.1	68.2	64.9	61.0	34.6	51.5	51.1	54.4
SPGraph^［29］	86.4	62.1	89.9	95.1	76.4	62.8	47.1	55.3	68.4	73.5	69.2	63.2	45.9	8.7	52.9
PCGCN^［21］	81.3	49. 5
AMFF-DGCNN^［30］	85.6	59.9	92.9	94.7	78.3	54.0	43.3	59.1	62.6	67.4	59.1	18.3	51.3	47.4	50.5
RG-GCN^［20］	88.1	63.7	94.0	96.2	79.1	60.4	44.3	60.1	65.9	70.8	64.9	30.8	51.9	52.6	56.4
DPFA^［31］	89.2	61.7	94.6	98.0	79.2	40.7	36.6	52.2	70.8	65.9	74.7	27.7	49.8	51.6	60.6
SC-CNN^［32］		67.4	93.6	94.1	83.1	50.5	33.0	60.3	69.6	63.0	72.3	63.8	64.3	60.6	57.1
DGCNN^［17］	84.5	55.5	93.2	95.9	72.8	54.6	32.2	56.2	50.7	62.8	63.4	22.7	38.2	32.5	46.8
AE-GCN （ours）	87.2	68.0	93.3	96.8	79.8	56.6	45.9	63.9	67.8	68.3	70.2	59.6	61.9	62.6	57.7

Table 2. Comparison of experimental results of Toronto_3D dataset

View table

Table 2. Comparison of experimental results of Toronto_3D dataset

Method	OA	mIoU	road	road marking	natural	building	utility line	pole	car	fence
PointNet++^［11］	84.9	41.8	89.3	0.0	69.0	54.1	43.7	23.3	52.0	3.0
KPFCNN^［33］	95.4	69.1	94.6	0.1	96.1	91.5	87.7	81.6	85.7	15.7
TGNet^［34］	94.1	61.3	93.5	0.0	90.8	81.6	65.3	63.0	88.7	7.9
MS-TGNet^［27］	95.7	70.5	94.4	17.2	95.7	88.8	76.0	74.0	94.2	23.6
Method of reference［35］	83.6	71.0	92.8	27.4	89.9	95.2	85.6	74.5	44.4	58.3
RG-GCN^［20］	96.5	74.5	98.2	79.4	91.8	86.1	72.4	69.9	82.1	16.0
MFA^［36］	97.0	79.9	96.8	70.0	96.1	92.3	86.3	80.4	91.5	29.4
Method of reference［37］	95.3	73.9	95.7	25.9	94.0	86.3	81.5	71.8	78.1	58.1
RandLA-Net^［12］	93.0	77.7	94.6	42.6	96.9	93.0	86.5	78.1	92.9	37.1
DGCNN^［17］	94.2	61.8	93.9	0.0	91.3	80.4	62.4	62.3	88.3	15.8
AE-GCN	97.1	80.3	96.7	65.2	96.6	93.5	85.3	77.3	92.6	35.3

Table 3. The ablation experiment of main modules
View table
Table 3. The ablation experiment of main modules
Model Edge U-Shpae structure Attention mIoU /%
A $x_{i}, x_{j} - x_{i}$ 49.1
B $x_{i}, x_{j} - x_{i}, x_{j}$ 50.6
C $x_{i}, x_{j} - x_{i}$ √ 56.3
D $x_{i}, x_{j} - x_{i}$ √ √ 61.8
E $x_{i}, x_{j} - x_{i}, x_{j}$ √ 58.5
F $x_{i}, x_{j} - x_{i}, x_{j}$ √ √ 62.6
G $x_{j} - x_{i}, x_{j}$ √ √ 55.1

Table 4. The ablation experiment of AE-Conv pooling method
View table
Table 4. The ablation experiment of AE-Conv pooling method
Model Pooling method mIoU /%
A Mean 57.0
B Max 58.2
C Sum 58.5
D Attention 62.6

Table 5. The ablation experiment of the number of neighbor points in the local area
View table
Table 5. The ablation experiment of the number of neighbor points in the local area
Model Number mIoU /%
A 8 59.4
B 12 61.3
C 16 62.6
D 20 61.4

Tools

Get Citation

Copy Citation Text

Lujian Zhang, Yuanwei Bi, Yaowen Liu, Yansen Huang. Augmented Edge Graph Convolutional Networks for Semantic Segmentation of 3D Point Clouds[J]. Laser & Optoelectronics Progress, 2024, 61(8): 0837009

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites