Computer Vision - ECCV 2020

Name: Computer Vision - ECCV 2020 | 16th European Conference, Glasgow, UK, August 23-28, 2020, Proceedings, Part XXVII
Brand: Springer
Price: 96.29 EUR
Availability: OnlineOnly

16th European Conference, Glasgow, UK, August 23-28, 2020, Proceedings, Part XXVII

Andrea Vedaldi Horst Bischof Thomas Brox Jan-Michael Frahm(Editor)

Springer (Publisher)

Published on 18. November 2020

XLII, 789 pages

E-Book

PDF with digital watermarking

System requirements

978-3-030-58583-9 (ISBN)

€96.29incl. 7% vat

System requirements

for PDF with digital watermarking

E-Book Single Licence

Available for download

Description

More details

Other editions

Content

Intro
Foreword
Preface
Organization
Contents - Part XXVII
Teaching Cameras to Feel: Estimating Tactile Physical Properties of Surfaces from Images
1 Introduction
2 Related Work
3 Surface Property Synesthesia Dataset
4 Methods
4.1 Mapping Vision to Touch
4.2 Viewpoint Selection
5 Experiments
5.1 Cross-Modal Experiments
5.2 Viewing Angle Selection Experiments
6 Conclusion
References
Accurate Optimization of Weighted Nuclear Norm for Non-Rigid Structure from Motion
1 Introduction
1.1 Related Work and Contributions
2 Bilinear Parameterization Penalties
2.1 Extreme Points and Optimality
3 Non-square Matrices
4 Linear Objectives - Weighted Nuclear Norms
5 Experiments
5.1 Pseudo Object Space Error (pOSE) and Non-Rigid Structure from Motion
5.2 Low-Rank Matrix Recovery with pOSE Errors
5.3 Non-Rigid Structure Recovery
6 Conclusions
References
Proposal-Based Video Completion
1 Introduction
2 Related Work
3 Proposal-Based Video Completion
3.1 Overview
3.2 3D Inpainting Network
3.3 Proposal Generation
3.4 Proposals Fusion
3.5 Training
3.6 Implementation Details
4 Experimental Results
4.1 Experimental Setting
4.2 Fixed Region Inpainting
4.3 Video Object Removal
4.4 Ablation Study
4.5 Failure Cases
5 Conclusion
References
HGNet: Hybrid Generative Network for Zero-Shot Domain Adaptation
1 Introduction
2 Related Work
3 The Proposed Method
3.1 Preliminaries and Motivation
3.2 Adaptive Feature Separation
3.3 Hybrid Generation
3.4 Training and Inference
4 Experiments
4.1 Datasets and Comparisons
4.2 Implementation Details
4.3 Experimental Results
4.4 Ablation Study
5 Conclusion
References
Beyond Monocular Deraining: Stereo Image Deraining via Semantic Understanding
1 Introduction
2 Related Work
2.1 Single Image Deraining
2.2 Video Deraining
2.3 Stereo Deraining
3 The Semantic-Aware Deraining Module
3.1 The Consolidation of Different Tasks
3.2 Image Deraining and Scene Segmentation
3.3 Semantic-Rethinking Loop
4 The Paired Rain Removal Network
4.1 Network Architecture
4.2 SFNet
4.3 VFNet
4.4 Objective Functions
5 Experiments
5.1 Datasets
5.2 Implementation Details
5.3 Ablation Study
5.4 Stereo Deraining
5.5 Monocular Deraining
5.6 Evaluation on Real-World Images
6 Conclusion
References
DBQ: A Differentiable Branch Quantizer for Lightweight Deep Neural Networks
1 Introduction
2 Related Work
3 Differentiable Branched Quantizer (DBQ)
3.1 Formulation of DBQ
3.2 Differentiability
3.3 Implementation Details
4 Experimental Results
4.1 Complexity Metrics
4.2 CIFAR-10 Results
4.3 ImageNet Results
4.4 Visual Wake Words Results
5 Conclusion
References
All at Once: Temporally Adaptive Multi-frame Interpolation with Advanced Motion Modeling
1 Introduction
2 Related Work
3 Proposed Method
3.1 Algorithm Overview
3.2 Cubic Flow Prediction
3.3 Motion Estimation
3.4 Temporal Pyramidal Network
3.5 Loss Functions
4 Experiments
4.1 Implementation Details
4.2 Evaluation Datasets
4.3 Comparison with the State-of-the-Arts
4.4 Ablation Studies
4.5 Efficiency Analysis
5 Conclusions
References
A Broader Study of Cross-Domain Few-Shot Learning
1 Introduction
2 Related Work
3 Proposed Benchmark
4 Cross-Domain Few-Shot Learning Formulation
5 Evaluated Methods for Cross-Domain Few-Shot Learning
5.1 Meta-learning Based Methods
5.2 Transfer Learning Based Methods
6 Evaluation Setup
7 Experimental Results
7.1 Meta-learning Based Results
7.2 Transfer Learning Based Results
7.3 Benchmark Summary
8 Conclusion
References
Practical Poisoning Attacks on Neural Networks
1 Introduction
2 System and Adversarial Model
3 Attack Methodology
4 Experimental Evaluation
4.1 Experiment Setup
4.2 Experiments Evaluation
5 Related Work
6 Conclusion
References
Unsupervised Domain Adaptation in the Dissimilarity Space for Person Re-identification
1 Introduction
2 Unsupervised Domain Adaptation for ReID
3 Proposed Method
3.1 Supervised Loss
3.2 Dissimilarity-Based Maximum Mean Discrepancy (D-MMD)
4 Results and Discussion
4.1 Experimental Methodology
4.2 Ablation Study
4.3 Comparison with State-of-Art Methods
5 Conclusion
References
Learn Distributed GAN with Temporary Discriminators
1 Introduction
1.1 Advantages of Distributed GAN Learning
1.2 Temporary Datasets Challenge for Distributed GAN
2 Related Work
2.1 Generative Adversarial Networks (GANs)
2.2 Federated Learning
2.3 Lifelong Learning
3 Method
3.1 Problem Definition
3.2 TDGAN Framework
3.3 Loss Function of TDGAN
3.4 Theoretical Guarantees of TDGAN Loss
4 Experiments
4.1 Experimental Set-Up
4.2 Results on Homogeneous Tasks
4.3 Results on Heterogeneous Tasks
5 Conclusion
References
SemifreddoNets: Partially Frozen Neural Networks for Efficient Computer Vision Systems
1 Introduction
2 Related Work
3 SemifreddoNets
3.1 Vertically Frozen Neural Networks
3.2 Backbone Model Architecture
3.3 Model Head
3.4 Repeatable Blocks
4 Results
5 Ablation Study
6 Conclusions
References
Improving Adversarial Robustness by Enforcing Local and Global Compactness
1 Introduction
2 Related Works
3 Proposed Method
3.1 Local Compactness
3.2 Global Compactness
3.3 Clustering Assumption and Label Supervision
3.4 Generating Adversarial Examples
3.5 Putting it all Together
4 Experiments
4.1 Experimental Setting
4.2 Experimental Results
5 Conclusion
References
TopoAL: An Adversarial Learning Approach for Topology-Aware Road Segmentation
1 Introduction
2 Related Work
3 Method
3.1 Formalization
3.2 Discriminator Network: D
3.3 Generator Network: G
3.4 Training
4 Experiments
4.1 Datasets, Metrics, and Baselines
4.2 Implementation Details
4.3 Boosting the Performance of Existing Architectures
4.4 Comparison Against the State-of-the-Art
5 Conclusion
References
Channel Selection Using Gumbel Softmax
1 Pruning and Conditional Computation
1.1 Gating Neural Networks
1.2 Our Loss
2 Related Work
2.1 Conditional Computation
2.2 Pruning
2.3 Regularization and Architecture Search
3 Technical Considerations
3.1 Gate Polarization
3.2 Training Considerations
3.3 Inference Strategies
3.4 Architectural Considerations
4 Experiments
4.1 Training Parameters
4.2 Results on ImageNet
4.3 Results on CIFAR-10
4.4 Analysis and Ablation Studies
5 DenseNet Extensions
6 Conclusion
References
Exploiting Temporal Coherence for Self-Supervised One-Shot Video Re-identification
1 Introduction
2 Related Works
3 Methodology
3.1 Progressive Pseudo-labeling and Its Drawbacks
3.2 Temporal Coherence as Self-supervision
3.3 Temporal Consistency Progressive Learning
4 Experiments
5 Conclusion
References
An Efficient Training Framework for Reversible Neural Architectures
1 Introduction
2 Background
2.1 Reversible Neural Architectures
2.2 Scheduling for Training
3 Method
3.1 Memory Centric and Computation Centric Modes
3.2 Formulation
3.3 Algorithm and Framework
3.4 Various Mini-Batch Size
4 Experiments
4.1 Settings
4.2 Profiling
4.3 RevNet
4.4 Inplace ABN
4.5 Reformer
4.6 Various Mini-Batch Sizes
5 Conclusions
References
Box2Seg: Attention Weighted Loss and Discriminative Feature Learning for Weakly Supervised Segmentation
1 Introduction
2 Related Work
3 Proposed Approach
3.1 Feed Forward Network Architecture
3.2 Box and GrabCut Based Losses
3.3 Attention Weighted Segmentation Loss
3.4 Discriminative Feature Learning
3.5 Training Box2Seg
4 Experiments and Results
4.1 Implementation Details
4.2 Quantitative and Qualitative Evaluation
4.3 Ablation Studies
4.4 Attention Weighted Loss in the Fully-Supervised Setting
4.5 Semi-supervised Semantic Segmentation
5 Conclusions
References
FreeCam3D: Snapshot Structured Light 3D with Freely-Moving Cameras
1 Introduction
2 Related Work
2.1 Active Depth Sensing Techniques
2.2 Indoor Localization
2.3 Deep Optics
3 Forward Model
3.1 Projector 2D Pattern Design
3.2 Depth Encoding with the Phase Mask
3.3 Image Warping
3.4 Dataset Generation
4 Reconstruction Algorithm
4.1 Image Preprocessing
4.2 Reconstruction Network
4.3 Loss Function
4.4 Training Details
4.5 Camera Pose Estimation
5 Simulation Results
5.1 Optimized Mask Design and Testing Results
5.2 Ablation Study and Comparisons
6 Experiment Results
7 Discussion and Conclusion
References
One-Pixel Signature: Characterizing CNN Models for Backdoor Detection
1 Introduction
2 Related Work
3 Backdoored CNN Detection with One-Pixel Signature
3.1 Trojan Attack Problem Settings
3.2 Neural Cleanse ch20NeuralCleanse and ABS ch20ABS
3.3 Overview of Our Backdoor Detection Method
4 One-Pixel Signature
4.1 Basic Formulation
4.2 Visualization and Illustration
4.3 Theoretical Justification
5 Experiments
6 Ablation Study
7 Conclusion
References
Learning to Transfer Learn: Reinforcement Learning-Based Selection for Adaptive Transfer Learning
1 Introduction
2 Related Work
3 Learning from Source and Target Datasets
4 Learning to Transfer Learn Framework
5 Experiments
5.1 Datasets and Implementation Details
5.2 Similar Domain Transfer Learning
5.3 Dissimilar Domain Transfer Learning
5.4 Computational Cost of Training
6 Conclusions
References
Structure-Aware Generation Network for Recipe Generation from Images
1 Introduction
2 Related Work
2.1 Image Captioning
2.2 Multimodal Food Computing
2.3 Language Parsing
3 Method
3.1 Overview
3.2 ON-LSTM Revisit
3.3 Recipe2tree Module
3.4 Img2tree Module
3.5 Tree2recipe Module
3.6 Recipe Generation
4 Experiments
4.1 Dataset and Evaluation Metrics
4.2 Implementation Details
4.3 Baselines
4.4 Main Results
4.5 Qualitative Results
5 Conclusion
References
A Simple and Effective Framework for Pairwise Deep Metric Learning
1 Introduction
2 Related Work
3 DML as a DRO-Based Binary Classification Problem
3.1 General DRO-Based Framework
3.2 Proposed Three Variants of Our Framework
3.3 Recovering the Method Based on SOTA Pair-Based Loss
4 Experiments
4.1 Quantitative Results
4.2 Ablation Study
5 Conclusion
References
Meta-rPPG: Remote Heart Rate Estimation Using a Transductive Meta-learner
1 Introduction
2 Related Work
3 Methodology
3.1 Network Architecture
3.2 Transductive Meta-learning
3.3 Posing rPPG Estimation as an Ordinal Regression Task
4 Experiments
4.1 Dataset and Experimental Settings
4.2 Evaluation on MAHNOB-HCI and UBFC-rPPG
5 Conclusion
References
A Recurrent Transformer Network for Novel View Action Synthesis
1 Introduction
2 Related Work
3 Proposed Method
3.1 Action Transformation
3.2 Action Key-Point Detection
3.3 Appearance Transformer Network
3.4 Action Synthesis
3.5 Implementation and Training Details
4 Experiments
4.1 Dataset
4.2 Evaluation
4.3 Ablation Study
4.4 Novel View with Novel Actor
4.5 Limitations and Failure Cases
5 Conclusion
References
Multi-view Action Recognition Using Cross-View Video Prediction
1 Introduction
2 Related Work
3 Method
3.1 Representation Learning Network (RL-NET)
3.2 Video Rendering Network (VR-NET)
3.3 Action Recognition
3.4 Training and Implementation Details
4 Experiments
4.1 Representation Learning via Rendering
4.2 Action Recognition
4.3 Transfer Learning for Action Recognition
4.4 Ablations and Discussion
5 Conclusion
References
Learning Discriminative Feature with CRF for Unsupervised Video Object Segmentation
1 Introduction
2 Related Work
3 The Proposed Method
3.1 Network Architecture
3.2 Discriminative Feature Module
3.3 Attention Module with CRF
4 Experiments
4.1 Unsuperviesed Video Object Segmentation Task
4.2 Image Object Co-Segmentation Task
4.3 Ablation Study
4.4 Model Analysis
5 Conclusion
References
SMART: Simultaneous Multi-Agent Recurrent Trajectory Prediction
1 Introduction
2 Related Work
3 Simulated Dataset
4 SMART
4.1 Problem Formulation
4.2 Method
5 Experiments
6 Conclusion
References
Label-Driven Reconstruction for Domain Adaptation in Semantic Segmentation
1 Introduction
2 Related Work
3 Algorithm
3.1 Overview
3.2 Target-to-Source Translation
3.3 Semantic Segmentation
3.4 Image Reconstruction from the Label Space
4 Experiments
4.1 Datasets
4.2 Network Architecture
4.3 Implementation Details
4.4 GTA5Cityscapes
4.5 SYNTHIACityscapes
4.6 Ablation Study
5 Conclusion
References
Efficient Outdoor 3D Point Cloud Semantic Segmentation for Critical Road Objects and Distributed Contexts
1 Introduction
2 Related Works
2.1 Deep Learning on Point Clouds
2.2 Attention Mechanism
3 Methods
3.1 Attention-Based Dynamic Point Convolution (ADConv)
3.2 Self-Attention Global Context (SAGC) Module
3.3 Network Architectures
4 Experiments
4.1 Training Details
4.2 Indoor Point Cloud Segmentation on S3DIS Dataset
4.3 Outdoor Point Cloud Segmentation on NPM3D Dataset
4.4 Effectiveness Analysis of Attention Mechanisms
4.5 Efficiency of ADConvnet-SAGC
4.6 Ablation Analysis
5 Conclusion
References
Attributional Robustness Training Using Input-Gradient Spatial Alignment
1 Introduction
2 Related Work
3 Attributional Robustness Training: Methodology
3.1 Attribution Manipulation
3.2 Attributional Robustness Training (ART)
3.3 Connection to Adversarial Robustness
3.4 Downstream Task: Weakly Supervised Object Localization (WSOL)
4 Experiments and Results
4.1 Attributional and Adversarial Robustness
4.2 Weakly Supervised Image Localization
5 Discussion and Ablation Studies
6 Conclusion
References
Reducing the Sim-to-Real Gap for Event Cameras
1 Introduction
2 Related Works
2.1 Video Reconstruction
2.2 Optic Flow
2.3 Input Representations
3 Method
3.1 Event Camera Contrast Threshold
3.2 Training Data
3.3 Sequence Length
3.4 Loss
3.5 Data Augmentation
3.6 Architecture
3.7 High Quality Frames Dataset
4 Experiments
4.1 Evaluation
4.2 Contrast Thresholds
4.3 Training Noise and Sequence Length
5 Discussion
References
Spatial Geometric Reasoning for Room Layout Estimation via Deep Reinforcement Learning
1 Introduction
2 Related Work
3 Approach
3.1 3D Layout Representation
3.2 Incremental Layout Estimation
4 Experiments
4.1 Datasets
4.2 Implementation Details
4.3 Results and Analysis
5 Conclusion
References
Learning Data Augmentation Strategies for Object Detection
1 Introduction
2 AutoAugment for Object Detection
2.1 Search Space Definition
2.2 Controller Settings
3 Experiments
3.1 Understanding the Policies Found by AutoAugment
3.2 Data Augmentation Policy Found by AutoAugment Systematically Improves Object Detection
3.3 Data Augmentation Policy Found by AutoAugment Push the State-of-the-Art on Object Detection Models
3.4 Data Augmentation Policy Found by AutoAugment Transfers to Other Detection Datasets
4 Analysis
4.1 Data Augmentation Policy Found by AutoAugment Mimics the Performance of Larger Annotated Datasets
4.2 Data Augmentation Improves Model Regularization
5 Related Work
6 Discussion
A Appendix
A.1 AutoAugment Controller Training Details
References
DA-NAS: Data Adapted Pruning for Efficient Neural Architecture Search
1 Introduction
2 Related Work
3 Understanding Network Training Process
4 Data Adapted NAS
4.1 Expanding Search Space
4.2 Searching with Constrains
4.3 Training Strategy
5 Experiment
5.1 Setup
5.2 Compared with State-of-the-Art Methods
5.3 Ablation Study
5.4 Transferring to Key-Point Localization Task
5.5 Visualization
6 Conclusion
References
A Closer Look at Generalisation in RAVEN
1 Introduction
2 Background and Related Work
2.1 Raven's Progressive Matrices and Neural Networks
2.2 Disentanglement and Scene Decomposition
3 Preliminary Investigation
4 Architectures
4.1 ResNet Baseline
4.2 Frame-Relational ResNet (Rel-Base)
4.3 Object-Relational ResNet (Rel-AIR)
5 Experiments
5.1 Data
5.2 Results on PGM
5.3 Results on RAVEN
6 Discussion
7 Conclusion
References
Supervised Edge Attention Network for Accurate Image Instance Segmentation
1 Introduction
2 Related Work
2.1 Instance Segmentation
2.2 Object Detection
2.3 Attention Mechanism
3 Our Approach
3.1 Fully Convolutional Box Head
3.2 Supervised Attention Module
3.3 Loss Function
4 Experiments
4.1 Evaluation Metrics
4.2 Training Details
4.3 Main Results
4.4 Ablation Study
5 Conclusions
References
Discriminative Partial Domain Adversarial Network
1 Introduction
2 Related Works
3 Discriminative Partial Domain Adversarial Network
3.1 Discriminative Partial Domain Adversarial Framework
3.2 Hard Binary Weights
3.3 Positive Partial Domain Adaptation
3.4 Negative Partial Domain Adaptation
3.5 Discriminative Partial Domain Adversarial Network
3.6 Theoretical Analysis
4 Experiment
4.1 Datasets and Protocols
4.2 Experimental Results
4.3 Analysis and Discussion
5 Conclusion
References
Differentiable Programming for Hyperspectral Unmixing Using a Physics-Based Dispersion Model
1 Introduction
2 Related Work
3 Method
3.1 Dispersion Model
3.2 Differentiable Programming for End-to-End Spectral Unmixing
4 Experimental Results
5 Discussion
References
Deep Cross-Species Feature Learning for Animal Face Recognition via Residual Interspecies Equivariant Network
1 Introduction
2 Related Work
3 Approach
3.1 Basic Idea
3.2 Residual Interspecies Feature Equivariant
3.3 Animal Facial Feature Fusion
4 Implementation
4.1 Preprocessing
4.2 Animal Face Verification
4.3 Stem Network
5 Experiments
5.1 Experimental Setting
5.2 Comparison with Our Baselines
5.3 Ablation Study of RiseNet
6 Conclusion
References
Guidance and Evaluation: Semantic-Aware Image Inpainting for Mixed Scenes
1 Introduction
2 Related Work
2.1 Deep Learning-Based Inpainting
2.2 Structural Information-Guided Inpainting
3 Approach
3.1 Semantic Guidance Network (SG-Net)
3.2 Semantic Guidance and Evaluation Network (SGE-Net)
3.3 Training Loss Function
4 Experiments
4.1 Setting
4.2 Image Inpainting Results
4.3 Ablation Study
5 Conclusion
References
Sound2Sight: Generating Visual Dynamics from Sound and Context
1 Introduction
2 Related Works
3 Proposed Method
4 Experiments
4.1 Experimental Results
5 Conclusions
References
3D-CVF: Generating Joint Camera and LiDAR Features Using Cross-view Spatial Feature Fusion for 3D Object Detection
1 Introduction
2 Related Work
2.1 LiDAR-Only 3D Object Detection
2.2 LiDAR and Camera Fusion-Based 3D Object Detection
3 Proposed 3D Object Detector
3.1 Overall Architecture
3.2 Cross-view Feature Mapping
3.3 Gated Camera-LiDAR Feature Fusion
3.4 3D-RoI Fusion-Based Refinement
3.5 Training Loss Function
4 Experiments
4.1 KITTI
4.2 nuScenes
4.3 Ablation Study
4.4 Performance Evaluation Based on Object Distance
5 Conclusions
References
NoiseRank: Unsupervised Label Noise Reduction with Dependence Models
1 Introduction
2 Related Work
3 Unsupervised Label Noise Reduction and Model Training Framework
3.1 Vector Representation
3.2 Label Noise Detection
3.3 Iterative Training
4 Experiments
4.1 Experiment Setup and Hyper-parameters
4.2 Label Noise Detection Experiments
4.3 Classification Experiments
4.4 Interpretability Analysis
5 Conclusion
References
Fast Adaptation to Super-Resolution Networks via Meta-learning
1 Introduction
2 Related Works
3 Meta-learning for Super-Resolution
3.1 Exploiting Patch-Recurrence for Deep SR
3.2 Handling Unknown SR Kernel for Deep SR
3.3 Proposed Method
4 Experimental Results
4.1 Implementation Details
4.2 MLSR with Fixed Bicubic SR Kernel
4.3 MLSR with Unseen SR Kernel
4.4 SR with Large Scaling Factor
5 Conclusion
References
TP-LSD: Tri-Points Based Line Segment Detector
1 Introduction
2 Related Work
2.1 Hand-Crafted Feature Based Methods
2.2 Deep Edge and Line Segment Detection
2.3 Object Detection
3 Tri-Points Representation
4 Methods
4.1 TP Extraction Branch
4.2 Line Segmentation Branch
4.3 Training and Inference
5 Evaluation Metrics
6 Experiments
6.1 Analysis of TP-LSD
6.2 Comparison with Other Methods
7 Conclusion
References
Author Index

System requirements

Save as PDF Copy link into clipboard

Schweitzer Fachinformationen

Computer Vision - ECCV 2020

Description

More details

Other editions

Additional editions

Content

System requirements