基于多尺度特征和注意力机制的深度学习点云压缩

doi:10.7523/j.ucas.2023.077

中国科学院大学学报 ›› 2024, Vol. 41 ›› Issue (5): 687-694.DOI: 10.7523/j.ucas.2023.077

• 电子信息与计算机科学 • 上一篇

基于多尺度特征和注意力机制的深度学习点云压缩

黄玉林^1,2, 梁磊³, 李卫军^1,2, 习晓环³

1. 中国能源建设集团广东省电力设计研究院有限公司, 广州 510663;
2. 广东科诺勘测工程有限公司, 广州 51066;
3. 中国科学院空天信息创新研究院数字地球重点实验室, 北京 100094

收稿日期:2023-03-06 修回日期:2023-09-13 发布日期:2023-10-25
通讯作者: 梁磊,E-mail:202231051040@mail.bnu.edu.cn
基金资助:
国家自然科学基金(41871264)和广西自然科学基金创新项目(2019GXNSFGA245001)资助

Point cloud compression of deep learning based on multi-scale feature and attention mechanism

HUANG Yulin^1,2, LIANG Lei³, LI Weijun^1,2, XI Xiaohuan³

1. Guangdong Electric Power Design Institute of China Energy Engineering Group Co. Ltd, Guangzhou 510663, China;
2. Guangdong Kenuo Surveying Engineering Co. Ltd, Guangzhou 51066;
3. Key Laboratory of Digital Earth Science, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China

Received:2023-03-06 Revised:2023-09-13 Published:2023-10-25

摘要/Abstract

摘要： 三维点云广泛应用于无人驾驶、实景三维等领域，然而复杂场景的海量点云对存储、处理和传输等带来极大挑战。提出一种基于多尺度特征和注意力机制的深度变分自编码点云几何信息压缩算法MSA-GPCC，通过加入多尺度模型提取特征、变分自编码器构建熵模型，继而结合尺度注意力模块和多尺度特征，实现基于熵编码的点云几何信息高码率、低失真压缩。在MPEG数据集上进行的实验表明，相比G-PCC算法和基于深度学习的D-PCC算法，MSA-GPCC算法在点间等比特率下平均质量增益分别提升7.72和4.91 dB，点到面等比特率下平均质量增益分别提升5.56和3.09 dB。

关键词: 点云压缩, 深度学习, 注意力机制, 变分自编码器, 多尺度特征

Abstract: 3D point clouds have extensive applications in the auto-drive, 3D real scene, and other fields. But complex scene requires massive point clouds to represent which brings great challenges to storage space, data processing and transmission bandwidth. A multi-scale attention point cloud geometry compression (MSA-GPCC) is proposed to compress point cloud data based on multi-scale features, attention mechanism, and variational auto-encoders (VAE). Experiments and analysis are carried out based on MPEG data sets. The results show that MSA-GPCC performs better than those of the traditional G-PCC and deep-learning-based D-PCC algorithms, D₁ BD-PSNR is improved by 7.72 and 4.91 dB respectively, and D₂ BD-PSNR is improved by 5.56 and 3.09 dB respectively.

Key words: point cloud compression, deep learning, attention mechanism, variational auto-encoders, multi-scale feature

中图分类号:

TP391.41

黄玉林, 梁磊, 李卫军, 习晓环. 基于多尺度特征和注意力机制的深度学习点云压缩[J]. 中国科学院大学学报, 2024, 41(5): 687-694.

HUANG Yulin, LIANG Lei, LI Weijun, XI Xiaohuan. Point cloud compression of deep learning based on multi-scale feature and attention mechanism[J]. Journal of University of Chinese Academy of Sciences, 2024, 41(5): 687-694.

参考文献

[1] 杨必胜,梁福逊,黄荣刚.三维激光扫描点云数据处理研究进展、挑战与趋势[J].测绘学报, 2017, 46(10):1509-1516. DOI:10.11947/j.AGCS.2017.20170351.
[2] Schnabel R, Klein R. Octree-based point-cloud compression[C]//IEEE VGTC Conference on Point-Based Graphics. July 29, 2006, Goslar, Germany. IEEE VGTC, 2006:111-120. DOI:10.2312/SPBG/SPBG06/111-120.
[3] Kammerl J, Blodow N, Rusu R B, et al. Real-time compression of point cloud streams[C]//2012 IEEE International Conference on Robotics and Automation. May 14-18, 2012, Saint Paul, MN, USA. IEEE, 2012:778-785. DOI:10.1109/ICRA.2012.6224647.
[4] 律帅,达飞鹏,黄源.基于数据类型转换的点云快速有损压缩算法[J].图学学报, 2016, 37(2):199-205. DOI:10.11996/JG.j.2095-302X.2016020199.
[5] 冯燕,何明一,魏江.基于神经网络的多光谱遥感图像无损压缩[J].遥感技术与应用, 2004, 19(1):42-46. DOI:10.3969/j.issn.1004-0323.2004.01.010.
[6] Yan W, Shao Y, Liu S, et al. "Deep autoencoder-based lossy geometry compression for point clouds" [EB/OL]. 2019:arXiv:1905.03691.(2019-04-18)[2023-03-06]. https://arxiv.org/abs/1905.03691.
[7] Huang L L, Wang S L, Wong K, et al. OctSqueeze:octree-structured entropy model for LiDAR compression[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). June 13-19, 2020, Seattle, WA, USA. IEEE, 2020:1310-1320. DOI:10.1109/CVPR42600.2020.00139.
[8] Biswas S, Liu J, Wong K, et al. "MuSCLE:multi sweep compression of LiDAR using deep entropy models" [EB/OL]. 2020:arXiv:2011.07590.(2020-11-15)[2023-03-06]. https://arxiv.org/abs/2011.07590.
[9] Quach M, Valenzise G, Dufaux F. Learning convolutional transforms for lossy point cloud geometry compression[C]//2019 IEEE International Conference on Image Processing (ICIP). September 22-25, 2019, Taipei, China. IEEE, 2019:4320-4324. DOI:10.1109/ICIP.2019.8803413.
[10] Guarda A F R, Rodrigues N M M, Pereira F. Point cloud coding:adopting a deep learning-based approach[C]//2019 Picture Coding Symposium (PCS). November 12-15, 2019, Ningbo, China. IEEE, 2020:1-5. DOI:10.1109/PCS48520.2019.8954537.
[11] 徐嘉诚,方志军,黄勃,等.端到端优化的3D点云几何信息有损压缩模型[J].武汉大学学报(理学版), 2022, 68(3):297-303. DOI:10.14188/j.1671-8836.2021.0083.
[12] Wang J Q, Ding D D, Li Z, et al. Multiscale point cloud geometry compression[C]//2021 Data Compression Conference (DCC). March 23-26, 2021, Snowbird, UT, USA. IEEE, 2021:73-82. DOI:10.1109/DCC50243.2021.00015.
[13] Tu C X, Takeuchi E, Carballo A, et al. Point cloud compression for 3D LiDAR sensor using recurrent neural network with residual blocks[C]//2019 International Conference on Robotics and Automation (ICRA). May 20-24, 2019, Montreal, QC, Canada. IEEE, 2019:3274-3280. DOI:10.1109/ICRA.2019.8794264.
[14] Mentzer F, Agustsson E, Tschannen M, et al. Practical full resolution learned lossless image compression[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). June 15-20, 2019, Long Beach, CA, USA. IEEE, 2020:10621-10630. DOI:10.1109/CVPR.2019.01088.
[15] Wu Z R, Song S R, Khosla A, et al. 3D ShapeNets:a deep representation for volumetric shapes[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). June 7-12, 2015, Boston, MA USA. IEEE, 2015:1912-1920. DOI:10.1109/CVPR.2015.7298801.
[16] Loop C, Cai Q, Escolano S O, et al. Jpeg pleno database:Microsoft voxelized upper bodies-a voxelized point cloud dataset[J]. ISO/IEC JTC1/SC29 Joint WG11/WG1(MPEG/JPEG) input document m38673/M72012, 2021.
[17] Krivokuca M, Chou P A, Savill P. 8i voxelized surface light field (8iVSLF) dataset[J]. ISO/IEC JTC1/SC29/WG11 MPEG, input document m42914, 2018.
[18] Graziosi D, Nakagami O, Kuma S, et al. An overview of ongoing point cloud compression standardization activities:video-based (V-PCC) and geometry-based (G-PCC)[J]. APSIPA Transactions on Signal and Information Processing, 2020, 9(1):1-7. DOI:10.1017/atsip.2020.12.
[19] Schwarz S, Martin-Cocher G, Flynn D, et al. Common test conditions for point cloud compression[J]. Document ISO/IEC JTC1/SC29/WG11 w17766, Ljubljana, Slovenia, 2018.

基于多尺度特征和注意力机制的深度学习点云压缩

Point cloud compression of deep learning based on multi-scale feature and attention mechanism

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

本文评价

访问统计

联系我们

[1]	陈经纬, 李宇, 陈俊, 张洪群. 基于MFF-Deeplabv3+网络的高分辨率遥感影像建筑物提取方法[J]. 中国科学院大学学报, 2024, 41(5): 654-664.
[2]	林昱彤, 王红, 柴团耀. 基于卷积神经网络多尺度特征的大豆基因组表型预测[J]. 中国科学院大学学报, 2024, 41(4): 468-476.
[3]	王兆瑞, 岩延, 张宝贤. 基于时空依赖关系多智能体强化学习的多路口交通信号协同控制方法[J]. 中国科学院大学学报, 2024, 41(3): 398-410.
[4]	李雪源, 韩丛英. Actor-critic框架下的二次指派问题求解方法[J]. 中国科学院大学学报, 2024, 41(2): 275-284.
[5]	肖俊, 石光田. 三维点云去噪技术[J]. 中国科学院大学学报, 2023, 40(5): 577-595.
[6]	姚沐风, 昝露洋, 李柏鹏, 李庆亭, 陈正超. 基于CAR-Siamese网络的高分辨率遥感图像建筑物变化检测[J]. 中国科学院大学学报, 2023, 40(3): 380-387.
[7]	盛晓光, 王颖, 张迎伟, 项若曦, 付红萍. WILS：面向学业预警的非均衡增量式学习方法[J]. 中国科学院大学学报, 2023, 40(3): 422-432.
[8]	李翔, 王艳, 李宝清. 基于FVC-CNN模型的野外车辆声信号分类[J]. 中国科学院大学学报, 2023, 40(2): 208-216.
[9]	霍鑫怡, 李焱磊, 陈龙永, 张福博, 孙巍. 基于卷积注意力和胶囊网络的 SAR少样本目标识别方法[J]. 中国科学院大学学报, 2022, 39(6): 783-792.
[10]	陈若男, 彭玲, 刘玉菲, 卫志超, 吕蓓茹, 陈德跃. 引入空间距离信息的城郊山区道路提取与应用[J]. 中国科学院大学学报, 2022, 39(5): 658-667.
[11]	古煜民, 阎福礼. 基于不同骨架UNet++网络的建筑物提取[J]. 中国科学院大学学报, 2022, 39(4): 512-523.
[12]	林大权, 范睿, 张良峰. 基于高维特征的图像对抗攻击算法[J]. 中国科学院大学学报, 2022, 39(3): 421-431.
[13]	赵敏钧, 赵亚伟, 赵雅捷, 罗刚. 一种新的基于深度学习的重叠关系联合抽取模型（英文）[J]. 中国科学院大学学报, 2022, 39(2): 240-251.
[14]	周龙, 王伟强, 吕科. FACR：一种快速且准确的车辆识别器[J]. 中国科学院大学学报, 2021, 38(1): 130-136.
[15]	李森, 彭玲, 胡媛, 池天河. 基于FD-RCF的高分辨率遥感影像耕地边缘检测[J]. 中国科学院大学学报, 2020, 37(4): 483-489.