Efficient semantic segmentation and compression of 3D point cloud data are essential for enabling scalable, real-world applications involving large-scale LiDAR environments. This work presents a lightweight hybrid framework that integrates point transformer v3 (PTv3) with squeeze-and-excitation (SE) attention blocks to enhance feature learning in sparse and imbalanced point clouds. To address class imbalance, we incorporate the synthetic minority over-sampling technique (SMOTE) during preprocessing. Additionally, a truncated pyramid-based compression scheme is employed to reduce data size while preserving geometric structure. The proposed approach achieves strong segmentation performance on SemanticKITTI and ShapeNet, reaching mean IoU scores of 87.5% and 89.4%, respectively. These results demonstrate the effectiveness of combining hierarchical attention, channel-wise modulation, and geometric compression for accurate and compact 3D scene understanding.

Keywords - Point cloud segmentation, LiDAR, PTv3, Squeeze-Excitation, semantic segmentation

[1] Y. Shen et al., “Techniques for extracting powerlines and pylons from LiDAR point clouds,” International Journal of Applied Earth Observation and Geoinformation, vol. 132, p. 104056, 2024.

[2] M. E. Atik and Z. Duran, “An efficient ensemble deep learning approach for semantic point cloud segmentation,” Sensors, vol. 22, no. 16, p. 6210, 2022.

[3] D. P. Singh and M. Yadav, “Deep learning-based semantic segmentation of three-dimensional point cloud: A comprehensive review,” International Journal of Remote Sensing, vol. 45, no. 2, pp. 532–586, 2024.

[4] S. Vora, A. H. Lang, B. Helou, and O. Beijbom, “PointPainting: Sequential fusion for 3D object detection,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2020, pp. 4604–4612.

[5] C. Qi et al., “PointNet++: Deep hierarchical feature learning on point sets in a metric space,” in Proc. NeurIPS, 2017.

[6] Y. Wang et al., “Dynamic Graph CNN for Learning on Point Clouds,” in ACM Trans. Graphics (SIGGRAPH Asia), 2019.

[7] H. Zhao et al., “Point Transformer,” in Proc. ICCV, 2021.

[8] Zhen, Y. et al. “Point-BERT: Pre-Training 3D Point Cloud Transformers with Masked Point Modeling,” CVPR, 2022.

[9] Thomas, H. et al. “KPConv: Flexible and Deformable Convolution for Point Clouds,” ICCV, 2019.

[10] L. Comesańa-Cebral et al., “Transport infrastructure management based on LiDAR synthetic data,” Infrastructures, vol. 9, no. 1, pp. 12–25, 2024.

[11] J. Hu, L. Shen, and G. Sun, “Squeeze-and-Excitation Networks,” in Proc. CVPR, 2018.

[12] N. V. Chawla et al., “SMOTE: Synthetic Minority Over-sampling Technique,” J. Artif. Intell. Res., vol. 16, pp. 321–357, 2002.

[13] R. Maskeliūnas, S. Maqsood, M. Vaškevičius, and J. Gelšvartas, “Fusing LiDAR and photogrammetry for accurate 3D data: A hybrid approach,” Remote Sensing, vol. 17, no. 3, p. 443, 2025.

[14] Behley et al., “SemanticKITTI: A Dataset for Semantic Scene Understanding of LiDAR Sequences,” Proc. ICCV, 2019.

[15] A. X. Chang et al., “ShapeNet: An Information-Rich 3D Model Repository,” arXiv preprint, arXiv:1512.03012, 2015.

[16] A. Dosovitskiy et al., “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint arXiv:2010.11929, 2020.

[17] H. I. Lin and M. C. Nguyen, “Boosting minority class prediction on imbalanced point cloud data,” Applied Sciences, vol. 10, no. 3, p. 973, 2020.

[18] R. Roriz, H. Silva, F. Dias, and T. Gomes, “A survey on data compression techniques for automotive LiDAR point clouds,” Sensors, vol. 24, no. 10, p. 3185, 2024.

[19] X. -F. Han, Y. -F. Jin, H. -X. Cheng and G. -Q. Xiao, "Dual Transformer for Point Cloud Analysis," in IEEE Transactions on Multimedia, vol. 25, pp. 5638-5648, 2023, doi: 10.1109/TMM.2022.3198318.