基于深度学习的三维形状补全研究综述

doi:10.11996/JG.j.2095-302X.2023020201

图学学报 ›› 2023, Vol. 44 ›› Issue (2): 201-215.DOI: 10.11996/JG.j.2095-302X.2023020201

基于深度学习的三维形状补全研究综述

杨柳¹^,²(), 吴晓群¹^,²()

1.北京工商大学计算机学院，北京 100048
2.食品安全大数据技术北京市重点实验室，北京 100048

收稿日期:2022-05-21 接受日期:2022-09-26 出版日期:2023-04-30 发布日期:2023-05-01
通讯作者: 吴晓群(1984-)，女，副教授，博士。主要研究方向为计算机图形学、数字几何处理、图像处理。E-mail：wuxiaoqun@btbu.edu.cn
作者简介:杨柳(1998-)，女，硕士研究生。主要研究方向为计算机图形学、数字几何处理、图像处理。E-mail：yliu112825@163.com
基金资助:
国家自然科学基金面上项目(62272014);“十三五”时期北京市属高校高水平教师队伍建设支持计划项目(CIT&TCD201904036)

3D shape completion via deep learning: a method survey

YANG Liu¹^,²(), WU Xiao-qun¹^,²()

1. School of Computer Science and Engineering, Beijing Technology and Business University, Beijing 100048, China
2. Beijing Key Laboratory of Big Data Technology for Food Safety, Beijing 100048, China

Received:2022-05-21 Accepted:2022-09-26 Online:2023-04-30 Published:2023-05-01
Contact: WU Xiao-qun (1984-), associate professor, Ph.D. Her main research interests cover computer graphics, digital geometry processing, image processing. E-mail：wuxiaoqun@btbu.edu.cn
About author:YANG Liu (1998-), master student. Her main research interests cover computer graphics, digital geometry processing, image processing. E-mail：yliu112825@163.com
Supported by:
National Natural Science Foundation of China General Program(62272014);Support Project of High-Level Teachers in Beijing Municipal Universities in the Period of 13th Five-Year Plan(CIT&TCD201904036)

摘要/Abstract

摘要：

三维形状补全是计算机图形学与计算机视觉的基础任务之一，具有广泛的应用背景。其目的旨在从部分缺失的形状数据中推断出完整的形状。针对现有基于深度学习的三维模型补全算法进行概述，根据描述符的形式不同，主要将其分为基于二维形状描述符的补全方法和基于三维形状描述符的补全方法两类。前者即将三维模型投影到二维空间中进行特征提取进而获得完整模型，包括基于二维图像和基于深度图的三维模型补全方法；后者即直接利用三维表示进行模型补全，按照对三维模型的表示方式不同，可进一步分为基于体素、基于点云和基于隐式的方法。同时，汇总了现有基于深度学习的三维模型补全算法所涉及的数据集与评价标准，并对该算法目前存在的问题进行分析和讨论，展望未来研究的新方向。

关键词: 三维形状补全, 深度学习, 卷积神经网络, 计算机视觉, 多层感知机

Abstract:

The task of 3D shape completion, a fundamental aspect of computer graphics and computer vision, has been widely employed in many fields. 3D shape completion aims to infer complete shapes from partially missing shape data. This paper reviewed the current 3D model completion algorithms based on deep learning, and analyzed their advantages and disadvantages. According to the different forms of descriptors, the 3D model completion algorithms could be broadly classified into two categories: the completion methods based on 2D shape descriptors and the completion methods based on 3D shape descriptors. The former involved the projecting of the 3D model into the 2D space for feature extraction to obtain a complete model, including 3D model completion methods based on 2D images and depth maps. The latter involved the direct use of 3D representation for model completion, and according to different representations of 3D models, could be further divided into voxel-based, point cloud-based and implicit-based methods. Meanwhile, this survey provided an overview of the commonly used datasets, metrics, and state-of-the-art performance in the field. It also analyzed and discussed the problems facing current 3D model completion algorithms based on deep learning, and suggested potential avenues for future research.

Key words: 3D shape completion, deep learning, convolutional neural network, computer vision, multi-layer perceptron

中图分类号:

TP391.41

杨柳, 吴晓群. 基于深度学习的三维形状补全研究综述[J]. 图学学报, 2023, 44(2): 201-215.

YANG Liu, WU Xiao-qun. 3D shape completion via deep learning: a method survey[J]. Journal of Graphics, 2023, 44(2): 201-215.

图/表 21

参考文献 65

[1]	ENGEL J, SCHÖPS T, CREMERS D. LSD-SLAM: large-scale direct monocular SLAM[M]//Computer Vision - ECCV 2014. Cham: Springer International Publishing, 2014: 834-849.
[2]	HOU J, DAI A, NIEßNER M. 3D-SIS: 3D semantic instance segmentation of RGB-D scans[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2019: 4416-4425.
[3]	BOUD A C, HANIFF D J, BABER C, et al. Virtual reality and augmented reality as a training tool for assembly tasks[C]// 1999 IEEE International Conference on Information Visualization. New York: IEEE Press, 1999: 32-36.
[4]	CHANG A X, FUNKHOUSER T, GUIBAS L, et al. ShapeNet: an information-rich 3D model repository[EB/OL]. [2021-12-09]. https://arxiv.org/abs/1512.03012.
[5]	YU X M, RAO Y M, WANG Z Y, et al. PoinTr: diverse point cloud completion with geometry-aware transformers[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2021: 12478-12487.
[6]	YUAN W T, KHOT T, HELD D, et al. PCN: point completion network[C]// 2018 International Conference on 3D Vision. New York: IEEE Press, 2018: 728-737.
[7]	TCHAPMI L P, KOSARAJU V, REZATOFIGHI H, et al. TopNet: structural point cloud decoder[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2019: 383-392.
[8]	HUANG Z T, YU Y K, XU J W, et al. PF-net: point fractal network for 3D point cloud completion[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 7659-7667.
[9]	ZHANG X C, FENG Y T, LI S Q, et al. View-guided point cloud completion[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 15885-15894.
[10]	PAN L, CHEN X Y, CAI Z A, et al. Variational relational point completion network[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 8520-8529.
[11]	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. New York: IEEE Press, 2015: 1912-1920.
[12]	GEIGER A, LENZ P, STILLER C, et al. Vision meets robotics: the KITTI dataset[J]. The International Journal of Robotics Research, 2013, 32(11): 1231-1237. DOI URL
[13]	GU J Y, MA W C, MANIVASAGAM S, et al. Weakly- supervised 3D shape completion in the wild[M]// Computer Vision - ECCV 2020. Cham: Springer International Publishing, 2020: 283-299.
[14]	DAI A, CHANG A X, SAVVA M, et al. ScanNet: richly- annotated 3D reconstructions of indoor scenes[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2017: 2432-2443.
[15]	ZHANG S L, LI S, HAO A M, et al. Point clou semantic scene completion from RGB-D images[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2021, 35(4): 3385-3393. DOI URL
[16]	HU T, HAN Z Z, SHRIVASTAVA A, et al. Render4Completion: synthesizing multi-view depth maps for 3D shape completion[C]// 2019 IEEE/CVF International Conference on Computer Vision Workshop. New York: IEEE Press, 2019: 4114-4122.
[17]	HU T, HAN Z Z, ZWICKER M. 3D shape completion with multi-view consistent inference[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 10997-11004. DOI URL
[18]	CHARLES R Q, HAO S, MO K C, et al. PointNet: deep learning on point sets for 3D classification and segmentation[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2017: 77-85.
[19]	QI C R, YI L, SU H, et al. PointNet++: deep hierarchical feature learning on point sets in a metric space[C]// The 31st International Conference on Neural Information Processing Systems. New York: ACM, 2017: 5105-5114.
[20]	DAI A, QI C R, NIEßNER M. Shape completion using 3D-encoder-predictor CNNs and shape synthesis[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2017: 6545-6554.
[21]	HAN X G, LI Z, HUANG H B, et al. High-resolution shape completion using deep neural networks for global structure and local geometry inference[C]// 2017 IEEE International Conference on Computer Vision. New York: IEEE Press, 2017: 85-93.
[22]	GKIOXARI G, JOHNSON J, MALIK J. Mesh R-CNN[C]// 2019 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2019: 9784-9794.
[23]	WANG P S, LIU Y, TONG X. Deep octree-based CNNs with output-guided skip connections for 3D shape and scene completion[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE Press, 2020: 1074-1081.
[24]	XIE H Z, YAO H X, ZHOU S C, et al. GRNet: gridding residual network for dense point cloud completion[M]// Computer Vision - ECCV 2020. Cham: Springer International Publishing, 2020: 365-381.
[25]	GUO Q, WANG Z J, JUEFEI-XU F, et al. CarveNet: carving point-block for complex 3D shape completion[EB/OL]. [2022-04-21]. https://arxiv.org/abs/2107.13452.
[26]	WANG X G, ANG M H, LEE G H. Voxel-based network for shape completion by leveraging edge generation[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2021: 13169-13178.
[27]	WEN X, XIANG P, HAN Z Z, et al. PMP-net: point cloud completion by learning multi-step point moving paths[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 7439-7448.
[28]	WEN X, LI T Y, HAN Z Z, et al. Point cloud completion by skip-attention network with hierarchical folding[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 1936-1945.
[29]	WANG X G, ANG M H, LEE G H. Cascaded refinement network for point cloud completion[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 787-796.
[30]	WANG X G, ANG M H, LEE G H. Point cloud completion by learning shape priors[C]// 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems. New York: IEEE Press, 2020: 10719-10726.
[31]	WANG Y D, TAN D J, NAVAB N, et al. SoftPoolNet: shape descriptor for point cloud completion and classification[M]// Computer Vision - ECCV 2020. Cham: Springer International Publishing, 2020: 70-85.
[32]	ZONG D M, SUN S L, ZHAO J. ASHF-net: adaptive sampling and hierarchical folding network for robust point cloud completion[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2021, 35(4): 3625-3632. DOI URL
[33]	GROUEIX T, FISHER M, KIM V G, et al. A papier-Mache approach to learning 3D surface generation[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 216-224.
[34]	LIU M H, SHENG L, YANG S, et al. Morphing and sampling network for dense point cloud completion[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 11596-11603. DOI URL
[35]	SON H, KIM Y M. SAUM: symmetry-aware upsampling module for consistent point cloud completion[C]// Computer Vision - ACCV 2020: 15th Asian Conference on Computer Vision, Revised Selected Papers, Part I. New York: ACM, 2020: 158-174.
[36]	ZHANG W X, YAN Q G, XIAO C X. Detail preserved point cloud completion via separated feature aggregation[M]// Computer Vision - ECCV 2020. Cham: Springer International Publishing, 2020: 512-528.
[37]	NIE Y Y, LIN Y Q, HAN X G, et al. Skeleton-bridged point completion: from global inference to local adjustment[EB/OL]. [2022-04-21]. https://arxiv.org/abs/2010.07428.
[38]	WANG Y, SUN Y B, LIU Z W, et al. Dynamic graph CNN for learning on point clouds[J]. ACM Transactions on Graphics, 2019, 38(5): 146.
[39]	PAN L. ECG: edge-aware point cloud completion with graph convolution[J]. IEEE Robotics and Automation Letters, 2020, 5(3): 4392-4398. DOI URL
[40]	WANG K Q, CHEN K, JIA K. Deep cascade generation on point sets[C]// The 28th International Joint Conference on Artificial Intelligence. New York: ACM, 2019: 3726-3732.
[41]	SHI J Q, XU L Y, HENG L, et al. Graph-guided deformation for point cloud completion[J]. IEEE Robotics and Automation Letters, 2021, 6(4): 7081-7088. DOI URL
[42]	ZHU L P, WANG B Y, TIAN G Y, et al. Towards point cloud completion: point rank sampling and cross-cascade graph CNN[J]. Neurocomputing, 2021, 461: 1-16. DOI URL
[43]	ALLIEGRO A, VALSESIA D, FRACASTORO G, et al. Denoise and contrast for category agnostic shape completion[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 4627-4636.
[44]	GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial networks[J]. Communications of the ACM, 2020, 63(11): 139-144. DOI URL
[45]	SARMAD M, LEE H J, KIM Y M. RL-GAN-net: a reinforcement learning agent controlled GAN network for real-time point cloud shape completion[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2019: 5891-5900.
[46]	XIE C L, WANG C X, ZHANG B, et al. Style-based point generator with adversarial rendering for point cloud completion[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 4617-4626.
[47]	CHEN X L, CHEN B Q, MITRA N J. Unpaired point cloud completion on real scans using adversarial training[EB/OL]. [2022-04-21]. https://arxiv.org/abs/1904.00069.
[48]	ZHANG J Z, CHEN X Y, CAI Z A, et al. Unsupervised 3D shape completion through GAN inversion[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 1768-1777.
[49]	WEN X, HAN Z Z, CAO Y P, et al. Cycle4Completion: unpaired point cloud completion using cycle transformation with missing region coding[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 13075-13084.
[50]	CAI Y J, LIN K Y, ZHANG C, et al. Learning a structured latent space for unsupervised point cloud completion[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 5533-5543.
[51]	XIANG P, WEN X, LIU Y S, et al. SnowflakeNet: point cloud completion by snowflake point deconvolution with skip-transformer[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2021: 5479-5489.
[52]	YU L Q, LI X Z, FU C W, et al. PU-net: point cloud upsampling network[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 2790-2799.
[53]	YU L Q, LI X Z, FU C W, et al. EC-net: an edge-aware point set consolidation network[M]//Computer Vision - ECCV 2018. Cham: Springer International Publishing, 2018: 398-414.
[54]	QIAN Y, HOU J H, KWONG S, et al. PUGeo-net: a geometry-centric network for 3D point cloud upsampling[M]// Computer Vision - ECCV 2020. Cham: Springer International Publishing, 2020: 752-769.
[55]	WANG Y F, WU S H, HUANG H, et al. Patch-based progressive 3D point set upsampling[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2019: 5951-5960.
[56]	QIAN G C, ABUALSHOUR A, LI G H, et al. PU-GCN: point cloud upsampling using graph convolutional networks[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 11678-11687.
[57]	LUO L Q, TANG L L, ZHOU W Y, et al. PU-EVA: an edge-vector based approximation solution for flexible-scale point cloud upsampling[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2021: 16188-16197.
[58]	LI R H, LI X Z, FU C W, et al. PU-GAN: a point cloud upsampling adversarial network[C]// 2019 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2019: 7202-7211.
[59]	PARK J J, FLORENCE P, STRAUB J, et al. DeepSDF: learning continuous signed distance functions for shape representation[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2019: 165-174.
[60]	CHIBANE J, ALLDIECK T, PONS-MOLL G. Implicit functions in feature space for 3D shape reconstruction and completion[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 6968-6979.
[61]	PING Y H, WEI G D, YANG L, et al. Self-attention implicit function networks for 3D dental data completion[J]. Computer Aided Geometric Design, 2021, 90: 1-12.
[62]	YAN W, ZHANG R N, WANG J, et al. Vaccine-style-net: point cloud completion in implicit continuous function space[C]// The 28th ACM International Conference on Multimedia. New York: ACM, 2020: 2067-2075.
[63]	YAN X G, LIN L Q, MITRA N J, et al. ShapeFormer: transformer-based shape completion via sparse representation[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 6229-6239.
[64]	TRETSCHK E, TEWARI A, GOLYANIK V, et al. PatchNets: patch-based generalizable deep implicit 3D shape representations[C]// Computer Vision - ECCV 2020: 16th European Conference. New York: ACM, 2020: 293-309.
[65]	CHABRA R, LENSSEN J E, ILG E, et al. Deep local shapes: learning local SDF priors for detailed 3D reconstruction[M]// Computer Vision - ECCV 2020. Cham: Springer International Publishing, 2020: 608-625.

名称	出处	相机数	类别数	模型总数
PCN dataset^[6]	PCN	8	8	30 974
Completion3D^[7]	TOP-Net	1	8	29 774
PF-Net dataset^[8]	PF-Net	-	13	14 473
ShapeNet-ViPC^[9]	ViPC	24	13	38 328
MVP dataset^[10]	VRC-Net	26	16	140 000

名称	出处	相机数	类别数	模型总数
PCN dataset^[6]	PCN	8	8	30 974
Completion3D^[7]	TOP-Net	1	8	29 774
PF-Net dataset^[8]	PF-Net	-	13	14 473
ShapeNet-ViPC^[9]	ViPC	24	13	38 328
MVP dataset^[10]	VRC-Net	26	16	140 000

方法	年份	思想	是否依赖GroundTruth完整模型	缺点	数据集
MVCN^[16]	2019	多视图补全	是	需要完整模型训练	ShapeNet KITTI
MVCI^[17]	2020	一致性推理	是	补全精度不佳	ShapeNet KITTI
Weakly-supervised^[13]	2020	弱监督补全	否	遗漏细节	ShapeNet
ViPC^[18]	2021	视图获得信息	是	-	ShapeNet-VIPC

方法	年份	思想	是否依赖GroundTruth完整模型	缺点	数据集
MVCN^[16]	2019	多视图补全	是	需要完整模型训练	ShapeNet KITTI
MVCI^[17]	2020	一致性推理	是	补全精度不佳	ShapeNet KITTI
Weakly-supervised^[13]	2020	弱监督补全	否	遗漏细节	ShapeNet
ViPC^[18]	2021	视图获得信息	是	-	ShapeNet-VIPC

方法	年份	思想	优点	缺点	数据集
3D-EPN^[20]	2017	体素补全	捕获形状信息	低分辨率输出	ShapeNet
3D-FCN^[21]	2017	局部几何细化	提炼局部几何	- 数据指数级增长 -	ShapeNet
Mesh R-CNN^[22]	2019	利用网格顶点边缘卷积进行细化	提高边缘感知		ShapeNet
O-CNNs^[23]	2020	体素转化为八叉树	提高补全鲁棒性		ShapeNet
GR-Net^[24]	2020	单元格加权卷积	空间感知特征	传感器影响	PCN datasets TopNet datasets KITTI
CarveNet^[25]	2021	逐点卷积	减少冗余点	时间空间复杂度高	ShapeNet KITTI
VE-PCN^[26]	2021	体素结合边缘信息	重建高分辨率	-	PCN datasets TopNet datasets

基于深度学习的三维形状补全研究综述

3D shape completion via deep learning: a method survey

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 21

参考文献 65

相关文章 15

编辑推荐

Metrics

本文评价

方法	年份	思想	优点	缺点	数据集
PCN^[6]	2018	MLP+折叠解码	捕获空间几何信息	低分辨率输出	ShapeNet PCN datasets
Top-Net^[7]	2019	树形结构解码	点云的拓扑或结构一致性问题	数据指数级增长	ShapeNet TopNet datasets
PMP-Net^[27]	2021	建立点对应关系	-	-	-
SA-Net^[28]	2020	跳跃注意连接解码	-	时空复杂度	PCN datasets KITTI
CR-Net^[29]	2020	级联细化	-	输出固定数量点云	PCN datasets TopNet datasets
LSP^[30]	2020	形状先验	保留局部几何细节	需要输入完整点云，利用完整点云监督训练	PCN datasets TopNet datasets
SoftPoolNet ^[31]	2020	基于软池思想	-	-	ShapeNet KITTI
ASHF-Net^[32]	2021	自适应采样和分层折叠			ShapeNet KITTI
VRCNet^[10]	2021	概率建模			MVP dataset
AtlasNet^[33]	2018	参数化映射曲面	生成任意分辨率的形状时避免内存占用问题	-	ShapeNets
MSN^[34]	2020	估计参数曲面	生成均匀分布点云并保持局部细节	-	ShapeNets
SAUM^[35]	2020	双支对称感知	提高分辨率	-	PCN datasets TopNet datasets KITTI

方法	年份	思想	优点	数据集
PF-Net^[8]	2020	分形几何	优点保留局部特征	ShapeNet PF-Net datasets
NSFA^[36]	2020	分离特征聚合	潜在形状预测	ShapeNet KITTI
SK-PCN^[37]	2020	骨架点位移	保持形状拓扑和局部细节	ShapeNet

方法	年份	思想	优点	缺点	数据集
ECG^[39]	2020	点边缘感知	边缘恢复	错误判断光滑边缘	ShapeNetPart TopNet datasets
DCG^[40]	2019	点特征边缘感知	细节恢复	忽略不连续边缘	ShapeNet Core
GGD^[41]	2021	粗补全后边缘卷积	结果平滑	冗余特征	PCN datasets KITTI Pandar40
PRSCN^[42]	2021	分层组合特征	减少冗余信息	-	ShapeNet
DeCo^[43]	2021	局部和全局信息组合输入编码	局部和全局相结合	-	ShapeNet

方法	年份	思想	优点	缺点	数据集
RL-GAN-Net^[45]	2019	强化学习	实时补全	精度不足	ShapeNet
SpareNet^[46]	2021	可微点渲染	保留局部几何细节	-	ShapeNet，KITTI
UPCC-AD^[47]	2019	无监督学习	未配对点补全	丢失细节	ScanNet，Matterport3D，KITTI
ShapeInversion^[48]	2021	反转变换	保持一致性	耗时	ShapeNet，KITTI，ScanNet，Matterport3D
Cycle4Completion^[49]	2021	循环变换	-	耗时	ShapeNet，KITTI
SLS^[50]	2022	统一潜在空间	增强结构性	-	ShapeNet，KITTI

方法	Avg	Airplane	Cabinet	Car	Chair	Lamp	Sofa	Table	Boat
3D-EPN^[20]	20.14	13.16	21.80	20.30	18.81	25.76	21.89	21.76	17.64
PCN^[6]	18.22	9.79	22.70	12.43	25.14	22.72	20.26	20.27	11.73
AtlasNet^[33]	17.77	10.36	23.40	13.40	24.16	20.24	20.82	17.52	11.62
TopNet^[7]	14.25	7.32	18.77	12.88	19.82	14.60	16.29	14.89	8.82
SoftPoolNet^[31]	11.90	4.89	18.86	10.17	15.22	12.34	14.87	11.84	6.48
SA-Net^[28]	11.22	5.27	14.45	7.78	13.67	13.53	14.22	11.75	8.84
GR-Net^[24]	10.64	6.13	16.90	8.27	12.23	10.22	14.93	10.08	5.86
MSN^[34]	9.60	5.62	11.93	8.63	11.64	10.30	13.18	9.65	5.88
PMP-Net^[27]	9.23	3.99	14.70	8.55	10.21	9.27	12.43	8.51	5.77
CR-Net^[29]	9.21	3.38	13.17	8.31	10.62	10.00	12.86	9.16	5.80
VE-PCN^[26]	8.10	3.83	12.74	7.86	8.66	7.24	11.47	7.88	4.75
VRCNet^[10]	8.23	3.59	12.06	7.68	9.37	8.76	11.22	7.65	5.48
SnowflakeNet^[51]	7.60	3.48	11.09	6.90	8.75	8.42	10.15	6.46	5.32
SpareNet^[46]	7.59	3.97	11.69	6.60	8.90	7.36	11.07	6.06	5.13

方法	Avg	Airplane	Cabinet	Car	Chair	Lamp	Sofa	Table	Boat
3D-EPN^[20]	14.21	7.38	20.95	17.64	12.48	13.90	12.63	13.52	15.21
PCN^[6]	9.64	5.50	22.70	10.63	8.70	11.00	11.34	11.68	8.59
AtlasNet^[33]	10.85	6.37	11.94	10.10	12.06	12.37	12.99	10.33	10.61
TopNet^[7]	12.15	7.61	13.31	10.90	13.82	14.44	14.78	11.12	11.12
SoftPoolNet^[31]	8.31	4.76	10.29	7.63	11.23	8.97	10.08	7.13	6.38
SA-Net^[28]	8.24	6.18	9.11	5.56	8.94	7.83	9.98	9.94	7.23
GR-Net^[24]	8.83	6.45	10.37	9.45	9.41	7.96	10.51	8.44	8.04
MSN^[34]	8.76	5.87	10.85	9.50	9.67	7.39	10.68	8.61	7.51
PMP-Net^[27]	8.73	5.56	11.24	9.64	9.51	6.95	10.83	8.72	7.25
CR-Net^[29]	8.51	4.79	9.97	8.31	9.49	8.94	10.69	7.81	5.77
VE-PCN^[26]	8.10	3.83	12.74	7.86	8.66	7.24	11.47	7.88	4.75
VRCNet^[10]	7.90	4.08	10.92	8.25	8.37	7.63	10.76	7.25	5.95
SnowflakeNet^[51]	7.21	4.29	9.16	8.08	7.89	6.07	9.23	6.55	6.40
SpareNet^[46]	7.23	4.16	9.18	7.63	7.53	7.03	9.53	6.35	6.48

评价指标	方法
评价指标	AtlasNet^[33]	PCN^[6]	TopNet^[7]	MSN^[34]	GR-Net^[24]	NSFA^[36]	CR-Net^[29]	VE-PCN^[26]	SpareNet^[46]
Consistency	0.700	1.557	0.568	1.951	0.313	0.391	0.582	12.630	0.249
Fidelity	1.759	2.235	5.354	0.434	0.816	0.347	0.337	0.258	1.461
MMD	2.108	1.336	0.636	2.259	0.568	0.426	0.394	0.372	0.368

方法		年份	思想	优点	缺点
基于PointNet++网络的方法	PU-Net^[52]	2018	基于数据驱动	提高分辨率	无法估计点云的分布
	EC-Net^[47]	2018	增加边缘感知	解决边缘问题	无法估计点云的分布
	PUGeo-Net^[54]	2020	学习法向局部参数化	从低质量输入获取结果	无法应用到不完整数据集
基于图卷积网络的方法	3PU^[55]	2019	渐进式上采样	提高细节保留	-
	PU-EVA^[57]	2021	基于边缘向量	实现任意上采样率	无法应用到没有配对的数据集
	PU-GCN^[56]	2021	设计多个上采样模块	不需要较多的参数实现	-
基于GAN网络的方法	PU-GAN^[58]	2019	基于对抗网络	增强分布的均匀性	点云数量固定

方法	CD (10^-2)	时间(ms)
PU-Net^[52]	5.56	10.04
3PU^[55]	2.98	10.86
PU-GAN^[58]	2.80	14.28
PU-GCN^[56]	2.58	8.83

方法	年份	思想	优点	缺点	数据集
DeepSDF^[59]	2019	形状条件分类器	处理复杂拓扑	消耗大量时间和数据集	ShapeNet
IF-Net^[60]	2020	对特征分类	保留细节	只适用于纹理补全	3DBodyTex.v2
SA-IFN^[61]	2021	与自注意机制结合	处理复杂模型	只适用于牙齿模型	牙齿模型数据
Vaccine-style-net^[62]	2020	连续边界函数	完整曲面模型	间接补全	ShapeNet
ShapeFormer^[63]	2022	基于Transform	处理复杂模型	耗时	ShapeNet
PatchNets^[64]	2020	基于块训练	无需多数据集	缺少局部细节	ShapeNet
DeepLS^[65]	2020	局部学习SDF	减少所需内存	无法进行细节重建	sketchup

[1]	毕春艳, 刘越. 基于深度学习的视频人体动作识别综述[J]. 图学学报, 2023, 44(4): 625-639.
[2]	曹义亲 , 周一纬 , 徐露 . 基于 E-YOLOX 的实时金属表面缺陷检测算法 [J]. 图学学报, 2023, 44(4): 677-690.
[3]	李鑫, 普园媛, 赵征鹏, 徐丹, 钱文华 . 内容语义和风格特征匹配一致的艺术风格迁移 [J]. 图学学报, 2023, 44(4): 699-709.
[4]	邵俊棋, 钱文华, 徐启豪. 基于条件残差生成对抗网络的风景图生成 [J]. 图学学报, 2023, 44(4): 710-717.
[5]	邓渭铭 , 杨铁军 , 李纯纯 , 黄琳 . 基于神经网络架构搜索的铭牌目标检测方法[J]. 图学学报, 2023, 44(4): 718-727.
[6]	余伟群, 刘佳涛, 张亚萍. 融合注意力的拉普拉斯金字塔单目深度估计 [J]. 图学学报, 2023, 44(4): 728-738.
[7]	郭印宏, 王立春, 李爽. 基于重复性和特异性约束的图像特征匹配 [J]. 图学学报, 2023, 44(4): 739-746.
[8]	毛爱坤, 刘昕明, 陈文壮, 宋绍楼. 改进YOLOv5算法的变电站仪表目标检测方法[J]. 图学学报, 2023, 44(3): 448-455.
[9]	王佳婧, 王晨, 朱媛媛, 王笑梅. 基于民国纸币的图元素匹配检索[J]. 图学学报, 2023, 44(3): 492-501.
[10]	曾武, 朱恒亮, 邢树礼, 林江宏, 毛国君. 显著性检测引导的图像数据增强方法[J]. 图学学报, 2023, 44(2): 260-270.
[11]	罗启明, 吴昊, 夏信, 袁国武. 基于Dual Dense U-Net的云南壁画破损区域预测[J]. 图学学报, 2023, 44(2): 304-312.
[12]	李洪安 , 郑峭雪 , 陶若霖 , 张敏 , 李占利 , 康宝生 . 基于深度学习的图像超分辨率研究综述[J]. 图学学报, 2023, 44(1): 1-15.
[13]	单芳湄, 王梦文, 李敏. 融合注意力机制的肠道息肉分割多尺度卷积神经网络 [J]. 图学学报, 2023, 44(1): 50-58.
[14]	邵英杰, 尹辉, 谢颖, 黄华. 草图引导的选择循环推理式人脸图像修复网络 [J]. 图学学报, 2023, 44(1): 67-76.
[15]	潘东辉, 金映含, 孙旭, 刘玉生, 张东亮. CTH-Net：从线稿和颜色点生成服装图像的 CNN-Transformer 混合网络 [J]. 图学学报, 2023, 44(1): 120-130.