Three-Dimensional Unsupervised Domain Adaptation Method with Balanced Geometry Perception

Yue Cai; Lei Guo; Zhongyu Chen; Xie Han; Shichao Jiao; Huiyan Han

doi:10.3788/LOP241635

Laser & Optoelectronics Progress, Volume. 62, Issue 6, 0615001(2025)

Three-Dimensional Unsupervised Domain Adaptation Method with Balanced Geometry Perception

Yue Cai^1、*, Lei Guo^1,2,3, Zhongyu Chen¹, Xie Han^1,2,3, Shichao Jiao¹, and Huiyan Han¹

Author Affiliations

¹School of Computer Science and Technology, North University of China, Taiyuan 030051, Shanxi , China

²Shanxi Key Laboratory of Machine Vision and Virtual Reality, Taiyuan 030051, Shanxi , China

³Shanxi Province's Vision Information Processing and Intelligent Robot Engineering Research Center, Taiyuan 030051, Shanxi , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Overall network structure

Download full size

Fig. 2. Adaptive unsigned distance

Download full size

Fig. 3. Results of DBSCAN clustering, (a) Ground truth provided by the dataset; (b) DBSCAN clustering effect

Download full size

Fig. 4. Point cloud mix-up augment

Download full size

Fig. 5. Visualization for the results of part segmentation

Download full size

Fig. 6. Ablation experimentⅡ

Download full size

Table 1. The mIoU of each method on the PointSegDA dataset

View table

Table 1. The mIoU of each method on the PointSegDA dataset

Method	PointSegDA
Method	F→M	F→A	F→S	M→F	M→A	M→S	A→F	A→M	A→S	S→F	S→M	S→A	Avg
Adapt^［2］	60.1	70.5	65.3	54.0	49.1	62.8	44.2	35.4	35.1	63.8	70.1	63.8	56.5
RS^［6］	60.7	78.7	66.9	38.4	59.6	70.4	44.0	30.4	36.6	65.3	70.7	73.0	57.9
DefRec^［14］	61.8	79.7	67.4	40.1	67.1	72.6	42.5	28.9	32.2	66.2	66.4	72.2	58.1
MLSP^［9］	60.0	80.9	65.5	40.4	67.3	70.8	45.4	31.1	38.4	72.5	66.6	74.8	59.5
LST^［17］	61.8	80.3	68.5	56.6	60.8	67.8	52.3	38.6	41.1	66.6	67.4	68.0	60.8
DGCNN+Aux^［7］	63.3	76.6	69.3	62.5	65.1	72..7	51.7	47.4	46.8	73.7	63.9	72.3	63.8
PC-Adapter^［18］	61.5	78.8	69.2	62.1	68.9	72.8	49.0	39.6	43.1	75.7	66.4	74.4	63.5
Proposed method	64.9	81.1	70.0	60.9	67.4	75.7	50.6	47.2	46.9	75.9	68.5	73.6	65.2

Table 2. The classification accuracy of each method on the PointDA-10 dataset

View table

Table 2. The classification accuracy of each method on the PointDA-10 dataset

Method	PointDA-10
Method	M→S	M→S*	S→M	S→S*	S*→M	S*→S	Avg
DANN^［5］	74.8	42.1	57.5	50.9	43.7	71.6	56.8
DefRec^［14］	81.7	51.8	78.6	54.5	73.7	71.1	68.6
RS^［6］	79.9	46.7	75.2	51.4	71.8	71.2	66.0
PointDAN^［3］	83.9	44.8	63.3	45.7	43.6	56.4	56.3
PatchMixer^［16］	83.1	51.4	81.0	45.1	64.1	67.1	65.3
PC-Adapter^［18］	83.3	58.2	77.5	53.7	73.7	75.4	70.3
Proposed method	84.8	56.3	77.7	56.6	78.7	76.8	71.8

Table 3. Ablation experiment Ⅰ

View table

Table 3. Ablation experiment Ⅰ

Method	PointSegDA
Method	F→M	F→A	F→S	M→F	M→A	M→S	A→F	A→M	A→S	S→F	S→M	S→A	Avg
S	60.4	75.9	65.0	55.4	60.0	70.9	49.2	34.4	37.7	67.5	63.1	70.5	59.2
S/T+Imp	64.5	79.5	68.3	55.2	64.5	71.8	51.9	42.9	47.2	71.8	65.2	74.8	63.1
S/T+Imp+Aug	64.9	81.1	70.0	60.9	67.4	75.7	50.6	47.2	46.9	75.9	68.5	73.3	65.2

Table 4. Ablation experiment Ⅲ
View table
Table 4. Ablation experiment Ⅲ
Remove noise points mIoU /%
yes 64.4
no 65.2

Tools

Get Citation

Copy Citation Text

Yue Cai, Lei Guo, Zhongyu Chen, Xie Han, Shichao Jiao, Huiyan Han. Three-Dimensional Unsupervised Domain Adaptation Method with Balanced Geometry Perception[J]. Laser & Optoelectronics Progress, 2025, 62(6): 0615001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites