Please wait a minute...
浙江大学学报(理学版)  2023, Vol. 50 Issue (6): 811-819    DOI: 10.3785/j.issn.1008-9497.2023.06.016
第15届全国几何设计与计算学术会议专题     
基于图注意力网络的三维网格分割算法
李文婷(),吴璐璐,周杰,赵勇()
中国海洋大学 数学科学学院,山东 青岛 266100
A 3D mesh segmentation algorithm based on graph attention network
Wenting LI(),Lulu WU,Jie ZHOU,Yong ZHAO()
School of Mathematical Sciences,Ocean University of China,Qingdao 266100,Shandong Province,China
 全文: PDF(1665 KB)   HTML( 2 )
摘要:

针对三维网格模型分割质量提升问题,提出了感知几何的图注意力网络。首先,定义了感知几何的图注意力系数,利用节点之间的边特征扩展由网络学习得到的注意力系数,引入与节点局部几何信息相关的注意力系数,更好地反映节点之间的相似性。然后,通过调整网络架构,将三维网格模型的几何特征与标签信息共同作为图注意力网络的输入,使标签信息参与网络训练和验证,并通过残差形式的线性连接实现网络的更稳定输出。大量实验结果表明,本文算法能够获得精确的分割边界,其在PSB数据集上的分割准确率较现有经典算法提升约2个百分点,也取得了更好的兰德指数。同时,通过消融实验验证了算法的合理性。

关键词: 网格分割图注意力系数边特征局部几何信息网络架构    
Abstract:

Improving the segmentation quality of 3D meshes is always an important problem to computer graphics. To handle this problem, this paper proposes a shape-aware graph attention network. The shape-aware graph attention coefficient is defined to better reflect the similarity between nodes, which not only expands the attention coefficient obtained by network learning with the help of edge features between nodes, but also introduces the attention coefficient related to the local shape information of nodes. On the other hand, the network architecture is adjusted by taking both shape features and labels of 3D mesh model as the input of graph attention network, which enables the participation of labels in network training and verification stages. Residual connection is further employed to make the network output more stable. A large number of experiments show that the proposed algorithm can obtain accurate segmentation boundaries. Compared with the existing classical segmentation algorithms on PSB dataset, the proposed algorithm improves 2% in accuracy, and achieves better Rand index. The reasonableness of the algorithm is proved by ablation experiment.

Key words: mesh segmentation    graph attention coefficient    edge feature    local shape information    network architecture
收稿日期: 2023-06-21 出版日期: 2023-11-30
CLC:  TP 391.41  
基金资助: 山东省自然科学基金资助项目(ZR2018MF006);浙江大学计算机辅助设计与图形系统全国重点实验室开放课题(A2228)
通讯作者: 赵勇     E-mail: lwt991212@163.com;zhaoyong@ouc.edu.cn
作者简介: 李文婷(1999—),ORCID:https://orcid.org/0009-0000-7344-9499,女,硕士研究生,主要从事计算机图形学、深度学习研究,E-mail:lwt991212@163.com.
服务  
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章  
李文婷
吴璐璐
周杰
赵勇

引用本文:

李文婷,吴璐璐,周杰,赵勇. 基于图注意力网络的三维网格分割算法[J]. 浙江大学学报(理学版), 2023, 50(6): 811-819.

Wenting LI,Lulu WU,Jie ZHOU,Yong ZHAO. A 3D mesh segmentation algorithm based on graph attention network. Journal of Zhejiang University (Science Edition), 2023, 50(6): 811-819.

链接本文:

https://www.zjujournals.com/sci/CN/10.3785/j.issn.1008-9497.2023.06.016        https://www.zjujournals.com/sci/CN/Y2023/V50/I6/811

图1  局部图结构中节点特征的聚合更新过程
图2  Armadillo模型的过分割结果
图3  局部图结构的构建
图4  图注意力网络架构
图5  三维网格分割算法对PSB数据集的分割结果
图6  不同算法的分割结果比较
图7  不同注意力系数的分割结果比较
模型类别Shapeboost10TOG1517ShapePFCN151DCNN18DL Framework19本文算法
平均值93.794.193.793.694.195.8
Human86.891.294.590.690.794.3
Cup94.099.793.894.598.199.8
Glasses96.997.696.696.398.198.6
Airplane96.196.793.095.995.296.9
Ant98.798.898.698.798.899.1
Chair98.198.798.597.797.698.8
Octopus98.298.898.398.598.798.9
Table99.499.699.599.699.099.7
Teddy98.798.297.788.398.698.8
Hand94.488.784.891.788.293.1
Plier95.296.295.595.895.396.0
Fish95.795.696.096.596.496.6
Bird89.688.388.591.088.691.5
Armadillo92.692.392.893.395.095.5
Fourleg83.387.085.087.784.687.9
Vase81.777.886.881.982.987.0
表1  不同算法的分割准确率比较 (%)
模型类别RandCuts39ShapeDiam34NormCuts39RandWalks2PMC6WcSeg40本文算法
平均值0.1650.2330.2480.2810.1550.0850.081
Human0.1650.1550.1560.2420.0590.0840.057
Cup0.0910.3540.4010.3260.0890.1250.076
Glasses0.3220.3220.3450.3070.1240.1670.123
Air-plane0.2250.1920.1900.3630.1080.0700.063
Ant0.1140.1120.1180.1620.0570.0090.115
Chair0.1880.1360.1540.1980.1160.0660.065
Octopus0.4230.3980.4050.4330.0660.0260.023
Table0.4490.2920.3200.2200.0250.0560.032
Teddy0.0730.0890.0750.1440.2940.0400.035
Hand0.1110.1850.1750.1240.2200.1030.134
Plier0.3140.3290.3870.3760.1100.0780.055
Fish0.3030.3170.3540.4840.4710.1440.136
Bird0.3030.2960.3320.4040.2120.0940.092
Arma-dillo0.0780.0970.1020.1050.1950.0760.072
Fourleg0.2820.2190.2100.3290.1750.1390.130
表2  不同算法的兰德指数比较
模型类别

网络学习得到的

注意力系数eijL

感知几何的

图注意力系数eij

平均值97.298.2
Human91.594.3
Cup98.999.8
Glasses98.398.6
Ant98.699.1
Chair97.098.8
Octopus98.598.9
Table99.199.7
Teddy97.898.8
Armadillo94.995.5
表3  不同注意力系数下分割准确率比较
1 SHLAFMAN S, TAL A, KATZ S. Metamorphosis of polyhedral surfaces using decomposition[J]. Computer Graphics Forum, 2002, 21(3): 219-228. DOI:10.1111/ 1467-8659.00581
doi: 10.1111/ 1467-8659.00581
2 LAI Y K, HU S M, MARTIN R R, et al. Fast mesh segmentation using random walks[C]// Proceedings of the 2008 ACM Symposium on Solid and Physical Modeling. New York: ACM, 2008: 183-191. DOI:10.1145/ 1364901.1364927
doi: 10.1145/ 1364901.1364927
3 CHEN X B, GOLOVINSKIY A, FUNKHOUSER T. A benchmark for 3D mesh segmentation[J]. ACM Transactions on Graphics, 2009, 28(3): 73. DOI:10. 1145/1531326.1531379
doi: 10. 1145/1531326.1531379
4 LIU H R, LIU W Y, LATECKI L J. Convex shape decomposition[C]// Proceedings of the 2010 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2010: 97-104. DOI:10.1109/CVPR.2010. 5540225
doi: 10.1109/CVPR.2010. 5540225
5 ZHENG Y Y, TAI C L. Mesh decomposition with cross-boundary brushes[J]. Computer Graphics Forum, 2010, 29(2): 527-535. DOI:10.1111/j.1467-8659.2009. 01622.x
doi: 10.1111/j.1467-8659.2009. 01622.x
6 FAN L B, LIC L, LIU K. Paint mesh cutting[J]. Computer Graphics Forum, 2011, 30(2): 603-612. DOI:10.1111/j.1467-8659.2011.01895.x
doi: 10.1111/j.1467-8659.2011.01895.x
7 ZHENG Y Y, TAI C L, AU O K. Dot scissor: A single-click interface for mesh segmentation[J]. IEEE Transactions on Visualization and Computer Graphics, 2012, 18(8): 1304-1312. DOI:10.1109/ TVCG.2011.140
doi: 10.1109/ TVCG.2011.140
8 TONG W H, YANG X K, PAN M D, et al. Spectral mesh segmentation via L0 gradient minimization[J]. IEEE Transactions on Visualization and Computer Graphics, 2020, 26(4): 1807-1820. DOI:10.1109/TVCG.2018. 2882212
doi: 10.1109/TVCG.2018. 2882212
9 LIN C, LIU L J, LI C J, et al. SEG-MAT: 3D shape segmentation using medial axis transform[J]. IEEE Transactions on Visualization and Computer Graphics, 2022, 28(6): 2430-2444. DOI:10.1109/ TVCG.2020.3032566
doi: 10.1109/ TVCG.2020.3032566
10 KALOGERAKIS E, HERTZMANN A, SINGH K. Learning 3D mesh segmentation and labeling[J]. ACM Transactions on Graphics, 2010, 29(4): No.102. DOI:10. 1145/1778765.1778839
doi: 10. 1145/1778765.1778839
11 BENHABILES H, LAVOUÉ G, VANDEBORRE J P, et al. Learning boundary edges for 3D-mesh segmentation[J]. Computer Graphics Forum, 2011, 30(8): 2170-2182. DOI:10.1111/j.1467-8659.2011. 01967.x
doi: 10.1111/j.1467-8659.2011. 01967.x
12 WANG Y H, GONG M L, WANG T H, et al. Projective analysis for 3D shape segmentation[J]. ACM Transactions on Graphics, 2013, 32(6): 192. DOI:10. 1145/2508363.2508393
doi: 10. 1145/2508363.2508393
13 XU W W, SHI Z X, XU M L, et al. Transductive 3D shape segmentation using sparse reconstruction[J]. Computer Graphics Forum, 2014, 33(5): 107-115. DOI:10.1111/cgf. 12436
doi: 10.1111/cgf. 12436
14 XIE Z G, XU K, LIU L G, et al. 3D shape segmentation and labeling via extreme learning machine[J]. Computer Graphics Forum, 2014, 33(5): 85-95. DOI:10.1111/cgf. 12434
doi: 10.1111/cgf. 12434
15 KALOGERAKIS E, AVERKIOU M, MAJI S, et al. 3D shape segmentation with projective convolutional networks[C]// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2017: 6630-6639. DOI:10. 1109/ CVPR.2017.702
doi: 10. 1109/ CVPR.2017.702
16 KUNDU A, YIN X Q, FATHI A, et al. Virtual multi-view fusion for 3D semantic segmentation[C]// Proceedings of the 16th European Conference on Computer Vision. Berlin: Springer, 2020: 518-535. DOI:10.1007/978-3-030-58586-0\_31. doi:10.1007/978-3-030-58586-0_31
doi: 10.1007/978-3-030-58586-0_31
17 GUO K, ZOU D Q, CHEN X W. 3D mesh labeling via deep convolutional neural networks[J]. ACM Transactions on Graphics, 2015, 35(1): 3. DOI:10.1145/2835487
doi: 10.1145/2835487
18 GEORGE D, XIE X H, TAM G K L. 3D mesh segmentation via multi-branch 1D convolutional neural networks[J]. Graphical Models, 2018, 96: 1-10. DOI:10.1016/j.gmod. 2018.01.001
doi: 10.1016/j.gmod. 2018.01.001
19 GEORGE D, XIE X H, LAI Y K, et al. A deep learning driven active framework for segmentation of large 3D shape collections[J]. Computer-Aided Design, 2022, 144: 103179. DOI:10.1016/j.cad.2021.103179
doi: 10.1016/j.cad.2021.103179
20 LI Y, HE X Y, JIANG Y K, et al. MeshFormer: High-resolution mesh segmentation with graph transformer[J]. Computer Graphics Forum, 2022, 41(7): 38-49. DOI:10.1111/cgf.14655
doi: 10.1111/cgf.14655
21 HANOCKA R, HERTZ A, FISH N, et al. MeshCNN: A network with edge[J]. ACM Transactions on Graphics, 2019, 38(4): 90. DOI:10.1145/3306346. 332295
doi: 10.1145/3306346. 332295
22 HUANG J W, ZHANG H T, YI L, et al. TextureNet: Consistent local parametrizations for learning from high-resolution signals on meshes[C]// Proceedings of the 2019 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2019: 4440-4449. DOI:10.1109/CVPR. 2019.00457
doi: 10.1109/CVPR. 2019.00457
23 FENG Y T, FENG Y F, YOU H X, et al. MeshNet: Mesh neural network for 3D shape representation[C]// Proceedings of the 23th AAAI Conference on Artificial Intelligence. Palo Alto: AAAI, 2019: 8279-8286. DOI:10. 1609/aaai.v33i01.33018279
doi: 10. 1609/aaai.v33i01.33018279
24 HE W C, JIANG Z, ZHANG C M, et al. CurvaNet: Geometric deep learning based on directional curvature for 3D shape analysis[C]// Proceedings of the 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York: ACM, 2020: 2214-2224. DOI:10.1145/3394486.3403272
doi: 10.1145/3394486.3403272
25 LAHAV A, TAL A. MeshWalker: Deep mesh understanding by random walks[J]. ACM Transactions on Graphics, 2020, 39(6): 263. DOI:10.1145/3414685. 3417806
doi: 10.1145/3414685. 3417806
26 HU S M, LIU Z N, GUO M H, et al. Subdivision-based mesh convolution networks[J]. ACM Transactions on Graphics, 2022, 41(3): No.25. DOI:10.1145/3506694
doi: 10.1145/3506694
27 QIAO Y L, GAO L, YANG J, et al. Learning on 3D meshes with Laplacian encoding and pooling[J]. IEEE Transactions on Visualization and Computer Graphics, 2022, 28(2): 1317-1327. DOI:10.1109/ TVCG.2020.3014449
doi: 10.1109/ TVCG.2020.3014449
28 SHARP N, ATTAIKI S, CRANE K, et al. DiffusionNet: Discretization agnostic learning on surfaces[J]. ACM Transactions on Graphics, 2022, 41(3): No. 27. DOI:10.1145/3507905
doi: 10.1145/3507905
29 PÉREZ D, SHEN Y Z, LI J. Mesh convolutional networks with face and vertex feature operators[J]. IEEE Transactions on Visualization and Computer Graphics, 2023, 29(3): 1678-1690. DOI:10.1109/ TVCG.2021.3129156
doi: 10.1109/ TVCG.2021.3129156
30 VELICKOVIC P, CUCURULL G, CASANOVA A, et al. Graph attention networks[C]// Proceedings of the 6th International Conference on Learning Representations. Vancouver: Open Review, 2018: 1-12.
31 KIM D, OH A. How to find your friendly neighborhood: Graph attention design with self-supervision[C]// Proceedings of the 9th International Conference on Learning Representations. Vancouver: Open Review, 2021:1-25.
32 WU L L, HOU Y, XU J L, et al. Robust mesh segmentation using feature-aware region fusion[J]. Sensors, 2023, 23(1): No.416. DOI:10.3390/s23010416
doi: 10.3390/s23010416
33 MEYER M, DESBRUN M, SCHRÖDER P, et al. Discrete differential-geometry operators for triangulated 2-manifolds[C]// Proceedings of the 3rd International Workshop “Visualization and Mathematics”. Berlin: Springer, 2002: 35-57. DOI:10.1007/978-3-662-05105-4\_ 2. doi:10.1007/978-3-662-05105-4_2
doi: 10.1007/978-3-662-05105-4_2
34 SHAPIRA L, SHAMIR A, COHEN-OR D. Consistent mesh partitioning and skeletonisation using the shape diameter function[J]. The Visual Computer, 2008, 24(4): 249-259. DOI:10.1007/s00371-007-0197-5
doi: 10.1007/s00371-007-0197-5
35 HILAGA M, SHINAGAWA Y, KOMURA T. Topology matching for fully automatic similarity estimation of 3D shapes[C]// Proceedings of the 28th Annual Conference on Computer Graphics and Interactive Techniques. New York: ACM, 2001: 203-212. DOI:10.1145/383259. 383282
doi: 10.1145/383259. 383282
36 BENCHEN M, GOTSMAN C. Characterizing shape using conformal factors[C]// Proceedings of the 1st Eurographics Workshop on 3D Object Retrieval. New York: ACM, 2008: 1-8. DOI:10.2312/3dor/3dor08/001-008
doi: 10.2312/3dor/3dor08/001-008
37 SUN J, OVSJANIKOV M, GUIBAS L. A concise and provably informative multi-scale signature based on heat diffusion[J]. Computer Graphics Forum, 2009, 28(5): 1383-1392. DOI:10.1111/j.1467-8659.2009. 01515.x
doi: 10.1111/j.1467-8659.2009. 01515.x
38 ZHANG G G, MA Q H, JIAO L C, et al. AttAN: Attention adversarial networks for 3D point cloud semantic segmentation[C]// Proceedings of the 29th International Joint Conference on Artificial Intelligence. San Mateo: Morgan Kaufmann, 2020: 789-796. DOI:10. 24963/ijcai.2020/110
doi: 10. 24963/ijcai.2020/110
39 GOLOVINSKIY A, FUNKHOUSER T. Randomized cuts for 3D mesh analysis[J]. ACM Transactions on Graphics, 2008, 27(5): No.145. DOI:10.1145/1409060. 1409098
doi: 10.1145/1409060. 1409098
[1] 王小超, 张雷, 余元强, 胡坤, 胡建平. 基于快速神经网络架构搜索的鲁棒图像水印网络算法[J]. 浙江大学学报(理学版), 2021, 48(3): 261-269.