基于图像视频的三维建模是3D数字技术的核心内容，可以重建真实3D场景和人物，广泛应用于机器人及自动驾驶等领域，属于跨学科研究领域，具有很高的研究和应用价值。本书围绕图像视频三维建模的最新研究技术和方法展开，重点关注具有挑战性问题进行了系统研究和介绍，包括3D物体建模、3D人脸面部建模、3D人体姿态建模及通用建模的相关学习算法，是一本系统介绍三维建模先进方法的研究专著。本书中描述的所有算法都来自我们的研究成果，与最先进方法进行了比较，验证了有效性和先进性。本书将使人工智能及信息计算机等领域的研究人员、专业人士和研究生受益，对跨学科研究也非常有用。

吴素萍，教授，一直从事计算机相关的教学科研工作，研究方向为机器视觉，人工智能，大数据和并行分布计算。担任中国计算机学会高性能计算专业委员。为国家人工智能，大数据重点研发项目评审专家。

Chapter 1 Introduction 1
1.1 3D Object Modeling 2
1.1.1 Single-View 3D Reconstruction 2
1.1.2 Multi-View 3D Reconstruction Method 3
1.2 3D Face Modeling 5
1.2.1 3D Face Keypoint Detection 5
1.2.2 3D Face Reconstruction 6
1.3 3D Human Body Modeling 9
1.3.1 3D Human Pose Estimation 9
1.3.2 3D human Body Reconstruction 10
1.4 3D Reconstruction Modeling 12
1.5 Outline of the Work 13
Bibliography 14
Chapter 2 3D Object Modeling 17
2.1 Single-View 3D Object Modeling 18
2.1.1 Multi-Scale Edge-Guided Learning for 3D Reconstruction 18
2.1.2 Multi-Granularity Relationship Reasoning Network for
High-Fidelity 3D Shape Reconstruction 44
2.1.3 3D Shape Reconstruction Based on Dynamic
Multi-Branch Information Fusion 67
2.1.4 Hierarchical Feature Learning Network for 3D Object
Reconstruction 78
2.2 Multi-View 3D Object Modeling 94
2.2.1 High-Resolution Multi-View Stereo with Dynamic Depth
Edge Flow 94
2.2.2 Global Contextual Complementary Network for Multi-View
Stereo 103
2.2.3 Attention-Guided Multi-View Stereo Network for
Depth Estimation 114
2.2.4 Self-Supervised Edge Structure Learning for Multi-View
Stereo and Parallel Optimization 126
2.2.5 Layered Decoupled Complementary Networks for
Multi-View Stereo 138
2.2.6 Global Balanced Networks for Multi-View Stereo 149
Bibliography 157
Chapter 3 3D Face Keypoint Detection 172
3.1 Learning Relation-Sensitive Structured Network for
Robust Face Alignment 173
3.1.1 Introduction 173
3.1.2 Proposed Method 174
3.1.3 Experiments 178
3.1.4 Conclusion 181
3.2 Multi-Agent Deep Collaboration Learning for Face Alignment
under Different Perspectives 182
3.2.1 Introduction 182
3.2.2 Proposed Method 184
3.2.3 Experiments 186
3.2.4 Conclusion 189
3.3 Towards Accurate 3D Face Alignment under Extreme
Scenarios Via Multi-Granularity Perturbation Relearning 190
3.3.1 Introduction 190
3.3.2 Proposed Method 192
3.3.3 Loss Function 196
3.3.4 Experiments 197
3.3.5 Conclusion 200
Bibliography 201
Chapter 4 3D Face Reconstruction 205
4.1 Towards Rich-Detail 3D Face Reconstruction and Dense
Alignment via Multi-Scale Detail Augmentation 206
4.1.1 Introduction 206
4.1.2 Proposed Method 207
4.1.3 Experiments 211
4.1.4 Conclusion 214
4.2 Multi-Attribute Regression Network for Face Reconstruction 215
4.2.1 Introduction 215
4.2.2 Proposed Method 217
4.2.3 Experiments 220
4.2.4 Conclusion 227
4.3 Geometry Normal Consistency Loss for 3D Face
Reconstruction and Dense Alignment 228
4.3.1 Introduction 228
4.3.2 Proposed Method 230
4.3.3 Experiments 234
4.3.4 Conclusion 238
4.4 Complementary Learning Network for 3D Face
Reconstruction and Alignment 238
4.4.1 Introduction 238
4.4.2 Proposed Method 240
4.4.3 Experiments 243
4.4.4 Conclusion 249
4.5 Graph Structure Reasoning Network for Face Alignment and
Reconstruction 250
4.5.1 Introduction 250
4.5.2 Proposed Method 252
4.5.3 Experiments 256
4.5.4 Conclusion 260
4.6 Unsupervised Shape Enhancement and Factorization Machine
Network for 3D Face Reconstruction 261
4.6.1 Introduction 261
4.6.2 Proposed Method 263
4.6.3 Experiments 266
4.6.4 Conclusion 270
4.7 A Detail Geometry Learning Network for High-Fidelity Face
Reconstruction 271
4.7.1 Introduction 271
4.7.2 Proposed Method 273
4.7.3 Experiments 277
4.7.4 Conclusion 281
4.8 A Bi-Directional Optimization Network for De-Obscured
3D High-Fidelity Face Reconstruction 282
4.8.1 Introduction 282
4.8.2 Proposed Method 284
4.8.3 Experiments 289
4.8.4 Conclusion 294
Bibliography 294
Chapter 5 3D Human Pose Estimation 304
5.1 Multi-Hybrid Extractor Network for 3D Human Pose Estimation 305
5.1.1 Introduction 305
5.1.2 Proposed Method 306
5.1.3 Experiments 310
5.1.4 Conclusion 312
5.2 3D Human Pose Estimation Based on Center of Gravity 312
5.2.1 Introduction 312
5.2.2 Proposed Method 315
5.2.3 Experiments 319
5.2.4 Conclusion 324
5.3 Edge-Angle Structure Constraint Loss for 3D Human Pose Estimation 324
5.3.1 Introduction 324
5.3.2 Related Works 325
5.3.3 Proposed Method 326
5.3.4 Experiments 329
5.3.5 Conclusion 333
Bibliography 333
Chapter 6 3D Human Body Reconstruction 339
6.1 Two-Stage Co-Segmentation Network Based on Discriminative
Representation for Recovering Human Mesh from Videos 340
6.1.1 Introduction 340
6.1.2 Related Works 342
6.1.3 Proposed Method 343
6.1.4 Experiments 350
6.1.5 Conclusion 356
6.2 Frame-Level Feature Tokenization Learning for Human
Body Pose and Shape Estimation 356
6.2.1 Introduction 356
6.2.2 Related Works 358
6.2.3 Proposed Method 359
6.2.4 Experiments 363
6.2.5 Conclusion 368
6.3 Time-Frequency Awareness Network for Human Mesh
Recovery from Videos 369
6.3.1 Introduction and Related Works 369
6.3.2 Proposed Method 371
6.3.3 Experiments 375
6.3.4 Conclusion 378
6.4 Spatio-Temporal Tendency Reasoning for Human Body Pose and
Shape Estimation from Videos 378
6.4.1 Introduction and Related Works 378
6.4.2 Proposed Method 380
6.4.3 Experiments 385
6.4.4 Conclusion 389
Bibliography 389

Chapter 7 3D Reconstruction Modeling 394
7.1 Replay Attention and Data Augmentation Network for
3D Face and Object Reconstruction 395
7.1.1 Introduction 395
7.1.2 Related Works 398
7.1.3 Proposed Method 402
7.1.4 Experiments 408
7.1.5 Conclusion 417
7.2 A Lightweight Grouped Low-Rank Tensor Approximation
Network for 3D Mesh Reconstruction from Videos 417
7.2.1 Introduction 417
7.2.2 Proposed Method 420
7.2.3 Experiments 423
7.2.4 Conclusion 429
Bibliography 429