Synthetic-Point-Clouds-for-semantic-segmentation

The results for the paper "Impact of color and mixing proportion of synthetic point clouds on semantic segmentation"

Introduction

In the paper, we propose a method for generating synthetic point clouds (SPC) based on BIM models, and explore the impact of the mixing proportions (synthetic & real point clouds) on the training performance of semantic segmentation models. The impact of the mixing proportions between synthetic and real point clouds on the training of PointNet, PointNet++, and DGCNN models has been presented in the paper. At the same time, we are actively exploring the potential of SPC on more advanced semantic segmentation models or network architectures.
The latest results will be continuously updated in this repository.

Results

Statement

Experiments Info

SPC proportion: the proportion of SPC to the total number of point clouds in the training set (proportion of scene quantity).
mixing training: experiments of mixed training of synthetic and real point clouds with certain SPC proportions.
benchmark: remove SPC from mixing training as benchmark experiments

Evaluation Metrics

mIoU: mean Intersection over Union.
OA: Overall Accuracy.
OA₇: overall accuracy not considering "clutter" class.

legend

PointNet (CVPR 2017) [paper]

PointNet++ (NIPS 2017) [paper]

DGCNN (ACM TOG 2019) [arXiv]

Point Transformer (ICCV 2021) [arXiv]

Citation

@article{zsj26,
	author = {Shaojie Zhou and Jia-Rui Lin and Peng Pan and Yuandong Pan and Ioannis Brilakis},
	title = {Impact of color and mixing proportion of synthetic point clouds on semantic segmentation},
	journal = {Automation in Construction},
	volume = {171},
	pages = {105963},
	year = {2025},
	doi = {https://doi.org/10.1016/j.autcon.2025.105963},
}