Computer vision - ACCV 2020 : 15th Asian conference on computer vision, Kyoto, Japan, November 30 - December 4, 2020 : revised selected papers, Part 1 / / Hiroshi Ishikawa [and three others], editors

• Record Nr.: UNISA996464413903316
• Pubbl/distr/stampa: Cham, Switzerland : , : Springer, , [2021]; ©2021
• ISBN: 3-030-69525-5
• Descrizione fisica: 1 online resource (755 pages) : illustrations
• Collana: Lecture Notes in Computer Science ; ; v.12622
• Disciplina: 006.37
• Soggetti: Computer vision; Optical data processing; Artificial intelligence
• Lingua di pubblicazione: Inglese
• Formato: Materiale a stampa
• Livello bibliografico: Monografia
• Nota di bibliografia: Includes bibliographical references.
• Nota di contenuto: Intro -- Preface -- Organization -- Contents - Part I -- 3D Computer Vision -- Weakly-Supervised Reconstruction of 3D Objects with Large Shape Variation from Single In-the-Wild Images -- 1 Introduction -- 2 Related Work -- 3 Method -- 3.1 Overview -- 3.2 Inference with Multi-scale Features -- 3.3 Shape-Sensitive Geometric Constraints -- 3.4 Training Data Augmentation via Background Manipulation -- 4 Experiments -- 4.1 Implementation Details -- 4.2 Datasets and Protocols -- 4.3 Model Analysis -- 4.4 Ablation Study -- 4.5 Comparison to State-of-the-Arts -- 5 Conclusion -- References -- HPGCNN: Hierarchical Parallel Group Convolutional Neural Networks for Point Clouds Processing -- 1 Introduction -- 2 Related Work -- 3 Hierarchical Parallel Group Convolutional Neural Network -- 3.1 Hierarchical Parallel Group Convolution -- 3.2 Hierarchical Parallel Group Convolutional Neural Network -- 4 Experiments -- 4.1 Shape Recognition -- 4.2 Semantic Segmentation -- 5 Conclusion -- References -- 3D Object Detection and Pose Estimation of Unseen Objects in Color Images with Local Surface Embeddings -- 1 Introduction -- 2 Related Work -- 2.1 3D Object Detection and Pose Estimation from Color Images -- 2.2 Training on Synthetic Images for 6D Pose Estimation -- 2.3 6D Pose Estimation Without Retraining -- 3 ; Method -- 3.1 Local Surface Embeddings -- 3.2 Predicting the Local Surface Embeddings for New Images -- 3.3 Pose Estimation Algorithm -- 4 Evaluation -- 4.1 Dataset -- 4.2 LSE Prediction Network Architecture and Training -- 4.3 Metrics -- 4.4 Results -- 4.5 Robustness to Texture -- 5 Conclusion -- References -- Reconstructing Creative Lego Models -- 1 Introduction -- 2 Related Work -- 3 Method -- 3.1 Triangle Mesh Generation -- 3.2 Flat Surface Extraction -- 3.3 Colour Information Extraction -- 3.4 Rotational Group Detection -- 3.5 Cuboid Detection.; 3.6 Chiselling -- 3.7 Snapping -- 4 Evaluation and Results -- 4.1 Angular DBSCAN -- 4.2 Lego Reconstruction -- 5 Conclusion -- References -- Multi-view Consistency Loss for Improved Single-Image 3D Reconstruction of Clothed People -- 1 Introduction -- 2 Related Work -- 2.1 Single View 3D Human Reconstruction -- 2.2 Datasets for 3D Human Reconstruction -- 3 Single-View Human Shape with Multi-view Training -- 3.1 Learning Architecture -- 3.2 3D Human Representation -- 3.3 The Proposed Loss Functions -- 3.4 3DVH Dataset -- 4 Experimental Evaluation -- 4.1 Implementation Details -- 4.2 Comparison -- 4.3 Generalization to Real Images of People -- 4.4 Ablation Study -- References -- Learning Global Pose Features in Graph Convolutional Networks for 3D Human Pose Estimation -- 1 Introduction -- 2 Related Work -- 3 Approach -- 3.1 Revisit GCN -- 3.2 Learning Global Pose Features -- 3.3 Network Architecture -- 4 Experiments -- 4.1 Datasets and Evaluation Protocols -- 4.2 Ablation Study -- 4.3 Comparison with the State of the Art -- 5 Conclusion -- References -- SGNet: Semantics Guided Deep Stereo Matching -- 1 Introduction -- 2 Related Work -- 3 Approach -- 3.1 Architecture Overview -- 3.2 Semantic Branch -- 3.3 Residual Module -- 3.4 Confidence Module -- 3.5 Loss Module -- 4 Experiments and Analysis -- 4.1 Ablation Studies -- 4.2 Comparing with Other Methods -- 5 Conclusion -- References -- Reconstructing Human Body Mesh from Point Clouds by Adversarial GP Network -- 1 Introduction -- 2 Related Work -- 3 Method -- 3.1 Network Architecture -- 3.2 Local and Global Spatial Consistency -- 3.3 Training Loss -- 4 Experimental Results -- 4.1 Datasets -- 4.2 Evaluation Protocol -- 4.3 Implementation Details -- 4.4 Comparison with Baselines -- 5 Conclusion -- References.; SDP-Net: Scene Flow Based Real-Time Object Detection and Prediction from Sequential 3D Point Clouds -- 1 Introduction -- 2 Related Work -- 2.1 Single-Frame 3D Object Detection -- 2.2 Multi-frame 3D Object Detection -- 2.3 Object Prediction and Tracking -- 3 SDP-Net -- 3.1 Network Structure -- 3.2 Loss Function -- 4 Experiments -- 4.1 Dataset and Implementation Details -- 4.2 Evaluation Results -- 5 Conclusions -- References -- SAUM: Symmetry-Aware Upsampling Module for Consistent Point Cloud Completion -- 1 Introduction -- 2 Related Works -- 3 Two-Branch Network -- 3.1 Encoder-Decoder Network -- 3.2 Symmetry-Aware Upsampling Module (SAUM) -- 4 Experimental Setting -- 4.1 Datasets -- 4.2 Implementation Details -- 4.3 Evaluation Metrics -- 5 Results -- 6 Conclusion -- References -- Faster Self-adaptive Deep Stereo -- 1 Introduction -- 2 Related Work -- 3 Methodology -- 3.1 The Overall Framework -- 3.2 Knowledge Reverse Distillation -- 3.3 The Adapt-or-Hold Mechanism -- 4 Experiment -- 4.1 Implementation Details -- 4.2 Benchmark Comparison -- 4.3 Ablation Study -- 5 Conclusion and Future Work -- References -- AFN: Attentional Feedback Network Based 3D Terrain Super-Resolution -- 1 Introduction -- 2 Related Work -- 2.1 Terrain Modeling -- 2.2 Super-Resolution of Images -- 2.3 DEM Super-Resolution with Neural Networks -- 3 Method -- 3.1 Proposed Attentional Feedback Network Architecture -- 4 Experimental Setup -- ; 4.1 Datasets -- 4.2 Implementation Details -- 5 Results and Discussions -- 5.1 Ablation Studies -- 6 Conclusion -- References -- Bi-Directional Attention for Joint Instance and Semantic Segmentation in Point Clouds -- 1 Introduction -- 2 Related Works -- 3 Motivation -- 4 Methodology and Implementation -- 4.1 Methodology -- 4.2 Implementation -- 5 Experiments -- 5.1 Experiments Setting -- 5.2 S3DIS Results -- 5.3 PartNet Results.; 5.4 ScanNetV2 Results -- 6 Discussion -- 6.1 Ablation Study -- 6.2 Mechanism Study -- 6.3 Efficiency Study -- 7 Conclusion -- References -- Anatomy and Geometry Constrained One-Stage Framework for 3D Human Pose Estimation -- 1 Introduction -- 2 Related Work -- 2.1 One Stage Methods -- 2.2 Two Stage Methods -- 3 Method -- 3.1 Kinematic Computation -- 3.2 Kinematic Structure Model -- 3.3 Bone-Length and Bone-Symmetry Loss -- 3.4 Multi-view Consistency Constraints -- 3.5 Loss Function -- 4 Experiment -- 4.1 Datasets -- 4.2 Implementation Details -- 4.3 Experiment Results on Human3.6M Dataset -- 4.4 Experiment Results on MPI-INF-3DHP Dataset -- 4.5 Qualitative Results on MPII Dataset -- 5 Conclusion -- References -- DeepVoxels++: Enhancing the Fidelity of Novel View Synthesis from 3D Voxel Embeddings -- 1 Introduction -- 2 Related Work -- 3 Method -- 3.1 View-Dependent Patch Feature Extraction -- 3.2 Recurrent-Concurrent Voxel Feature Aggregation -- 3.3 View-Dependent Patch Rendering -- 3.4 Implementation Details -- 4 Experiments and Discussion -- 4.1 Dataset and Metrics -- 4.2 Evaluating 360 Novel View Synthesis -- 4.3 Ablation Studies -- 5 Conclusion and Limitations -- References -- Dehazing Cost Volume for Deep Multi-view Stereo in Scattering Media -- 1 Introduction -- 2 Related Work -- 2.1 Multi-view Stereo -- 2.2 Dehazing -- 2.3 3D Reconstruction in Scattering Media -- 3 Multi-view Stereo in Scattering Media -- 3.1 Image Formation Model -- 3.2 Overview -- 3.3 Dehazing Cost Volume -- 3.4 Network Architecture and Loss Function -- 4 Experiments -- 4.1 Dataset -- 4.2 Training Details -- 4.3 Results -- 4.4 Experiments with Actual Data -- 4.5 Discusssion on Errors of Scattering Parameters -- 5 Conclusion -- References -- Homography-Based Egomotion Estimation Using Gravity and SIFT Features -- 1 Introduction -- 2 Background -- 2.1 DOF Analysis.; 2.2 Orientation- and Scale-Covariant Feature Constraints -- 3 Pose Estimation -- 3.1 Points on a Horizontal Plane -- 3.2 Points on a Vertical Plane -- 4 Experiments -- 4.1 Computational Complexity -- 4.2 Synthetic Evaluation -- 4.3 Camera Mounted to a Moving Vehicle -- 4.4 Smart Phone Images -- 5 Conclusions -- References -- Mapping of Sparse 3D Data Using Alternating Projection -- 1 Introduction -- 2 Related Work -- 3 Problem Statement -- 4 Alternating Projection (AP) -- 5 Implementation Details -- 5.1 Pre-processing -- 5.2 Full Meta-algorithm with RANSAC -- 6 Experiments -- 6.1 Sensitivity Analysis -- 7 Discussion -- References -- Best Buddies Registration for Point Clouds -- 1 Introduction -- 2 Related Work -- 3 Method -- 4 Experiments -- 4.1 Performance Evaluation Setup -- 4.2 Comparing Accuracy and Robustness Between BBR Variants -- 4.3 Accuracy -- 4.4 Robustness -- 4.5 Odometry -- 5 Conclusions -- References -- Project to Adapt: Domain Adaptation for Depth Completion from Noisy and Sparse Sensor Data -- 1 Introduction -- 2 Related Work -- 2.1 Unimodal Approaches -- 2.2 Depth Completion from RGB and LiDAR -- 3 Method -- 3.1 Data Generation via Projections -- 3.2 RGB Adaptation -- 3.3 Filtering Projection Artifacts for Supervision -- 3.4 Summary of Losses -- 4 Experiments -- 4.1 Ablation Study -- 4.2 Method Evaluation -- 5 Conclusions -- References -- Dynamic Depth Fusion and Transformation for Monocular 3D Object Detection -- 1 ; Introduction -- 2 Related Work -- 3 Methodology -- 3.1 Approach Overview -- 3.2 Adaptive Depth-Guided Instance Normalization -- 3.3 Dynamic Depth Transformation -- 3.4 3D Bounding Box Estimation -- 3.5 Feature Extraction -- 3.6 Implementation Details -- 4 Experiments -- 4.1 Results on KITTI -- 4.2 Ablation Study -- 5 Conclusion -- References.; Attention-Aware Feature Aggregation for Real-Time Stereo Matching on Edge Devices.