Computer Vision - ECCV 2020

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

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

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

Springer (Publisher)

Published on 12. November 2020

XLII, 791 pages

E-Book

PDF with digital watermarking

System requirements

978-3-030-58529-7 (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 XIX
High-Resolution Image Inpainting with Iterative Confidence Feedback and Guided Upsampling
1 Introduction
2 Related Work
3 Approach
3.1 Data Generation
3.2 Inpainting Model
3.3 Guided Upsampling
4 Experiments
4.1 Implementation Details
4.2 Comparison with State-of-the-Art Methods
4.3 Ablation Study
5 Conclusion
References
Online Ensemble Model Compression Using Knowledge Distillation
1 Introduction
2 Related Works
2.1 Model Compression Using Knowledge Distillation
2.2 Online Knowledge Distillation
2.3 Intermediate Representation Knowledge Distillation
3 Methodology
3.1 Ensemble Student Model Compression
3.2 Intermediate Knowledge Distillation
3.3 Knowledge Distillation Based Training
4 Experiments
4.1 Evaluation of Our Online Model Compression Framework
5 Ablation Studies
6 Discussion
7 Conclusion
References
Deep Learning-Based Pupil Center Detection for Fast and Accurate Eye Tracking System
1 Introduction
2 Related Works
3 Method
3.1 Overview
3.2 Face Detection Network
3.3 Glasses Removal Network
3.4 Segmentation Network
4 Experiments
4.1 Datasets and Experimental Environment
4.2 Quantitative Results
4.3 Qualitative Results
4.4 Ablation Study
5 Conclusions
References
Efficient Residue Number System Based Winograd Convolution
1 Introduction
2 Related Work
3 Residue Number System (RNS)
3.1 Convolution in RNS
4 Winograd Convolution
5 Winograd Convolution over Residue Number System
6 Fast Convolution via Integral Arithmetic for Convolutional Neural Networks (CNN)
7 Performance Analysis
8 Experiments
9 Conclusions
References
Robust Tracking Against Adversarial Attacks
1 Introduction
2 Related Works
2.1 Deep Visual Tracking
2.2 Adversarial Attacks and Defense
3 Proposed Algorithm
3.1 Adversarial Example Generation
3.2 Adversarial Defense
3.3 Deployment of Deep Trackers
4 Experiments
4.1 Ablation Studies
4.2 Benchmark Performance
4.3 Qualitative Evaluations
5 Concluding Remarks
References
Single-Shot Neural Relighting and SVBRDF Estimation
1 Introduction
2 Related Works
3 Method
3.1 Motivation for Our Design
3.2 Joint SVBRDF Estimation and Relighting
3.3 Training Details
4 Experiments
4.1 Ablation Study
4.2 Generalization to Real Data
4.3 Comparative Study
4.4 Environment Illumination Editing
4.5 Variance Under Different Environment Maps
5 Conclusion and Future Work
References
Unsupervised 3D Human Pose Representation with Viewpoint and Pose Disentanglement
1 Introduction
2 Related Works
2.1 Modeling Human 3D Poses
2.2 Learning Pose Representations
3 Our Approach
3.1 Problem Formulation
3.2 Sequential Bidirectional Recursive Network
3.3 Learning Framework Based on SeBiReNet
4 Experiments
4.1 Experimental Setup
4.2 Evaluation on Pose Denoising
4.3 Evaluation on Unsupervised Cross-view Action Recognition
4.4 Ablation Study
4.5 Extension Evaluation on 3D Pose Estimation
5 Conclusion
References
Angle-Based Search Space Shrinking for Neural Architecture Search
1 Introduction
2 Related Work
3 Search Space Shrinking
3.1 Elaborately Shrunk Search Space Is Better
3.2 Angle-Based Metric
3.3 Angle-Based Shrinking Method
4 Experiments
4.1 Empirical Study on Angle-Based Metric
4.2 NAS Algorithms with Angle-Based Shrinking
5 Conclusion and Future Work
References
RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition
1 Introduction
2 Related Work
3 Methodology
3.1 Background
3.2 Decoder Dissection
3.3 RobustScanner
4 Experiments
4.1 Datasets
4.2 Implementation Details
4.3 Comparison with State-of-the-Art Approaches
4.4 Ablation Study
5 Conclusions
References
Towards Fast, Accurate and Stable 3D Dense Face Alignment
1 Introduction
2 Methodology
2.1 Preliminary of 3DMM
2.2 Meta-Joint Optimization
2.3 Landmark-Regression Regularization
2.4 3D Aided Short-video-synthesis
3 Experiments
3.1 Datasets and Evaluation Protocols
3.2 Ablation Study
3.3 Evaluations of Accuracy and Stability
3.4 Evaluations of Speed
3.5 Analysis of Meta-Joint Optimization
4 Conclusion
References
Iterative Feature Transformation for Fast and Versatile Universal Style Transfer
1 Introduction
2 Related Works
3 Method
3.1 General Framework - Background
3.2 New Iterative Transformation with Analytical Gradient Descent
3.3 Applications that Demonstrate the Versatility of Our Method
4 Experiments
4.1 Single-Style Transfer
4.2 Photo-Realistic Style Transfer - Quantitative Analysis
4.3 Other Applications
5 Conclusion
References
CATCH: Context-Based Meta Reinforcement Learning for Transferrable Architecture Search
1 Introduction
2 Related Work
3 CATCH Framework
3.1 Context Encoding
3.2 Network Sampling
3.3 Network Scoring and Evaluation
3.4 Optimization of CATCHer
4 Experiments
4.1 Implementation Details
4.2 Benchmark on NAS-Bench-201
4.3 Experiments on Residual Block-Based Search Space
5 Ablation Study
6 Conclusion and Discussion
References
Toward Faster and Simpler Matrix Normalization via Rank-1 Update
1 Introduction
2 Background
2.1 Matrix Normalization
2.2 Compact Bilinear Pooling
3 Rank-1 Update Normalization (RUN)
4 Experiments
4.1 Datasets and Evaluation Metrics
4.2 Implementation Details
4.3 Ablation Study on RUN with Original Bilinear Pooling
4.4 Ablation Study on RUN with Compact Bilinear Pooling
4.5 Comparison with Other Pooling Methods
5 Conclusion
References
Accurate Polarimetric BRDF for Real Polarization Scene Rendering
1 Introduction
2 Related Work
2.1 Shape from Polarization
2.2 Polarimetric BRDF
3 Basics of Polarization
3.1 Surface Normal from Polarized Images
3.2 Stokes Vector and Mueller Matrix
4 Our Polarimetric BRDF Based on Measurement
4.1 Polarization Measurement System
4.2 Polarimetric BRDF Model
4.3 Evaluation of Our Polarimetric BRDF
5 Polarization Renderer
6 Shape from Polarization by CNN Trained with Synthesized Polarization Images
7 Conclusion
References
Lensless Imaging with Focusing Sparse URA Masks in Long-Wave Infrared and Its Application for Human Detection
1 Introduction
2 Related Work
3 Proposed Method
3.1 Mask Design
3.2 Scene Reconstruction
3.3 Diffraction Simulation for Imaging Matrix Estimation
4 Implementation and Results
4.1 System Overview
4.2 Mask Prototyping
4.3 Implementation of the Scene Reconstruction
4.4 Implementation of the Imaging Matrix Simulation
4.5 Robustness to the Alignment Error
5 Experiments in Inference Task
5.1 Human Detection Algorithm Pipeline
5.2 Performance Evaluation with Synthetic Dataset
5.3 Experiment with Real-World Dataset
6 Conclusion
References
Topology-Preserving Class-Incremental Learning
1 Introduction
2 Related Work
2.1 Multi-task Incremental Learning
2.2 Class-Incremental Learning
3 Topology-Preserving Class-Incremental Learning
3.1 Problem Definition
3.2 Overall Framework
3.3 Topology Modelling via Elastic Hebbian Graph
3.4 Topology-Preserving Constraint
3.5 Optimization
3.6 Comparison with the Distillation-Based Approaches
4 Experiments
4.1 Datasets and Experimental Setups
4.2 Comparison Results
4.3 Analysis of the TPCIL Components
4.4 Sensitivity Study of the Hyper-parameter
5 Conclusion
References
Inter-Image Communication for Weakly Supervised Localization
1 Introduction
2 Related Work
3 Methodology
3.1 Object Seed Vectors
3.2 Stochastic Consistency
3.3 Global Consistency
4 Experiments
4.1 Experiment Setup
4.2 Comparison with the State-of-the-Arts
4.3 Ablation Study
5 Conclusion
References
UFO2: A Unified Framework Towards Omni-supervised Object Detection
1 Introduction
2 Related Work
3 UFO2
3.1 Unified Model
3.2 Proposal Refinement
4 Partial Labels Simulation
5 Experiments
5.1 Evaluation of Single Labels
5.2 Qualitative Results
5.3 Ablation Study
5.4 Omni-supervised Learning
6 Conclusions
A Extensions: Learning to Detect Everything
B Annotation Policies
C Additional Visualization of Partial Labels
D Additional Qualitative Results
References
iCaps: An Interpretable Classifier via Disentangled Capsule Networks
1 Introduction
2 Related Work
2.1 Capsule Networks
2.2 Disentanglement
2.3 Interpretable Methods
3 iCaps: An Interpretable Classifier via Disentangled Capsule Networks
3.1 Disentanglement Between Class-Relevant and Class-Irrelevant Information
3.2 Latent Traversal
3.3 Interpretable Classifier Based on Learned Concepts
4 Experiments
4.1 Informativeness
4.2 Distinctness
4.3 Explainability
5 Conclusion and Future Work
References
Detecting Natural Disasters, Damage, and Incidents in the Wild
1 Introduction
2 Related Work
3 Incidents Dataset
4 Incident Model
5 Experiments on the Incidents Dataset
6 Detecting Incidents in Social Media Images
6.1 Incident Detection from Flickr Images
6.2 Incident Detection from Twitter Images
6.3 Temporal Monitoring of Incidents on Twitter
7 Conclusion
References
Dynamic ReLU
1 Introduction
2 Related Work
3 Dynamic ReLU
3.1 Dynamic Activation
3.2 Definition and Implementation of Dynamic ReLU
3.3 Relation to Prior Work
4 Variations of Dynamic ReLU
4.1 Network Structure and Complexity
4.2 Ablations
5 Experimental Results
5.1 ImageNet Classification
5.2 Inspecting DY-ReLU: Is It Dynamic?
5.3 Ablation Studies on ImageNet
5.4 COCO Single-Person Keypoint Detection
6 Conclusion
References
Acquiring Dynamic Light Fields Through Coded Aperture Camera
1 Introduction
2 Proposed Method
2.1 Notations and Problem Formulation
2.2 Reconstruction of Dynamic Light Field
2.3 Network Architecture
2.4 Dataset and Training Procedure
3 Experiments
3.1 Quantitative Evaluation
3.2 Experiment Using Physical Coded Aperture Camera
4 Conclusions
References
Gait Recognition from a Single Image Using a Phase-Aware Gait Cycle Reconstruction Network
1 Introduction
2 Related Work
2.1 Gait Recognition from Low Frame-Rate Videos
2.2 Gait Representation
3 Gait Recognition Using PA-GCRNet
3.1 Overview
3.2 PA-GCR
3.3 Combining PA-GCR with GaitSet
3.4 Unified Loss Function
4 Experiments
4.1 Datasets
4.2 Implementation Details
4.3 Visualizing Gait Cycle Reconstruction
4.4 Comparison with State-of-the-Art Methods
4.5 Ablation Study
5 Conclusion
References
Informative Sample Mining Network for Multi-domain Image-to-Image Translation
1 Introduction
2 Related Work
3 Proposed Method
3.1 Adversarial Importance Weighting
3.2 Multi-hop Sample Training
3.3 Implementation Details
4 Experiments
4.1 Dataset
4.2 Facial Attribute Transfer
4.3 Season Transfer
4.4 Edge&Photo Transfer
4.5 Ablation Study
5 Conclusion
References
Spherical Feature Transform for Deep Metric Learning
1 Introduction
2 Related Work
3 Proposed Approach
3.1 Review of Feature Transform
3.2 Review of Spherical-Homoscedasticity
3.3 Spherical Feature Transform
3.4 Theoretical Analysis
3.5 Training Scheme
4 Experiments
4.1 Quantitative Results
4.2 Ablation Study
5 Conclusion
References
Semantic Equivalent Adversarial Data Augmentation for Visual Question Answering
1 Introduction
2 Related Work
3 Method
3.1 VQA Model
3.2 Data Augmentation
3.3 Adversarial Training with Augmented Examples
4 Experiments
4.1 Experiments Setup
4.2 Results
4.3 Analysis
4.4 Ablation Studies
4.5 Model Robustness
4.6 Human Evaluation of Semantic Consistency
5 Conclusion
References
Unsupervised Multi-view CNN for Salient View Selection of 3D Objects and Scenes
1 Introduction
2 Related Work
3 Salient View Selection via UMVCNN
3.1 Multi-view Representation of a 3D Object
3.2 UMVCNN Architecture
3.3 Salient View Selection
3.4 Implementation
4 Experimental Results
4.1 Datasets
4.2 Qualitative Results
4.3 Quantitative Results
4.4 Evaluations over the Variants of UMVCNN
5 Conclusions
References
Representation Sharing for Fast Object Detector Search and Beyond
1 Introduction
2 Related Work
2.1 Object Detection and Instance Segmentation
2.2 Neural Architecture Search
3 Fast Diverse-Transformation Search
3.1 Search Space of FAD
3.2 Fast Search with Representation Sharing
4 Experiments
4.1 Object Detection
4.2 Instance Segmentation
5 Conclusion
References
Peeking into Occluded Joints: A Novel Framework for Crowd Pose Estimation
1 Introduction
2 Related Works
3 OPEC-Net: Occluded Pose Estimation and Correction
3.1 Initial Pose Estimation from Heatmap-Based Modules
3.2 GCN-Based Joints Correction
3.3 Loss Functions
4 Our Occluded Pose Dataset
5 Experiments
5.1 Experiments Settings
5.2 Performance Comparison on Our OCPose Dataset
5.3 Comparison Against State-of-the-Arts on Other Benchmarks
5.4 Alabtion Studies
6 Conclusion
References
RubiksNet: Learnable 3D-Shift for Efficient Video Action Recognition
1 Introduction
2 Related Work
3 Technical Approach
3.1 RubiksShift: Learnable 3D Shift
3.2 RubiksNet: Model Architecture Design
4 Experiments and Analysis
4.1 Experimental Setup
4.2 Benchmark Comparisons and Analysis
4.3 Ablations and Analysis
5 Conclusion
References
Deep Hashing with Active Pairwise Supervision
1 Introduction
2 Related Work
3 Approach
3.1 Deep Hashing with Active Pairwise Supervision
3.2 Structural Risk Minimization for Active Hashing
3.3 Designing the Acquisition Function via Structural Risk Minimization
4 Experiments
4.1 Datasets and Implementation Details
4.2 Ablation Study
4.3 Comparison with the State-of-the-art Methods
4.4 Visualization
5 Conclusion
References
Graph Edit Distance Reward: Learning to Edit Scene Graph
1 Introduction
2 Related Work
3 Method
3.1 Our Model
3.2 Graph Edit Distance
3.3 Policy Gradient + GED reward
4 CRIR Dataset
4.1 Image Generation
4.2 Query Generation
5 Experiment
5.1 Experiment on CSS Dataset
5.2 Experiment on CRIR Dataset
6 Conclusion
References
Malleable 2.5D Convolution: Learning Receptive Fields Along the Depth-Axis for RGB-D Scene Parsing
1 Introduction
2 Related Works
3 Malleable 2.5D Convolution
3.1 Learning Receptive Fields Along the Depth-Axis
3.2 Kernel Rebalancing
3.3 Understanding Malleable 2.5D Convolution
4 Experiments
4.1 Datasets
4.2 Implementation Details
4.3 Main Results
4.4 Ablation Studies
5 Conclusion
References
Feature-Metric Loss for Self-supervised Learning of Depth and Egomotion
1 Introduction
2 Related Work
3 Method
3.1 Geometry Models
3.2 Cross-View Reconstruction
3.3 Single-View Reconstruction
3.4 Overall Pipeline
3.5 Implementation Details
4 Experiments
4.1 Depth Evaluation
4.2 Odometry Evaluation
4.3 Ablation Study
5 Conclusion
References
Propagating Over Phrase Relations for One-Stage Visual Grounding
1 Introduction
2 Related Work
3 Approach
3.1 Image and Language Representation
3.2 Relational Propagation
3.3 Prediction and Loss
4 Experiments
4.1 Dataset and Evaluation
4.2 Implementation
4.3 Comparison with the State of the Art
4.4 Ablation Study
4.5 Qualitative Evaluation
5 Conclusions
References
Adversarial Semantic Data Augmentation for Human Pose Estimation
1 Introduction
2 Related Work
2.1 Human Pose Estimation
2.2 Data Augmentation
2.3 Adversarial Learning
3 Methodology
3.1 Semantic Data Augmentation
3.2 Adversarial Learning
4 Experiments
4.1 Datasets and Evaluation Protocols
4.2 Implementation Details
4.3 Quantitative Results
4.4 Qualitative Results
4.5 Ablation Studies
5 Conclusions
References
Free View Synthesis
1 Introduction
2 Related Work
3 Method
3.1 Preprocessing
3.2 Selection of Source Images
3.3 Mapping and Blending
4 Experimental Evaluation
5 Conclusion
References
Face Anti-Spoofing via Disentangled Representation Learning
1 Introduction
2 Related Work
3 Disentanglement Framework
3.1 Disentanglement Process
3.2 Auxiliary Supervision
4 Experimental Results
4.1 Experimental Setting
4.2 Experimental Comparison
4.3 Ablation Study
5 Further Exploration
6 Conclusions
References
Prime-Aware Adaptive Distillation
1 Introduction
2 Related Work
3 Introducing Sample Weighting to KD
3.1 A General Formulation of KD
3.2 Sample Weighting and a Few Baselines
4 Prime-Aware Adaptive Distillation
5 Experiment
5.1 Classification
5.2 Metric Learning
5.3 Object Detection
5.4 Analysis
6 Conclusion
References
Meta-learning with Network Pruning
1 Introduction
2 Related Work
3 Method
3.1 Problem Setup
3.2 Meta-Learning with Model Capacity Constraint
3.3 Generalization Analysis
4 Algorithm
4.1 Main Algorithm: Reptile with Iterative Network Pruning
4.2 Two Substantialized Implementations
5 Experiments
5.1 Few-Shot Classification Performances
5.2 On the Impact of Hyperparameters
5.3 Performance on More Complex Networks
6 Conclusion
A Proofs of Results
A.1 Proof of Theorem 1
A.2 Proof of Corollary 1
B Detailed Experimental Settings
B.1 Model
B.2 Datasets
B.3 Detailed Experimental Settings
C Additional Experimental Results
C.1 Results on Omniglot dataset
C.2 Results on MiniImageNet dataset
C.3 Results on TieredImageNet dataset
References
Spiral Generative Network for Image Extrapolation
1 Introduction
2 Related Work
2.1 Generative Adversarial Networks
2.2 Image Extrapolation
3 Spiral Generative Network
3.1 ImagineGAN
3.2 SliceGAN
3.3 Spiral Loss Design
3.4 Case of Unknown Margin
3.5 Implementation Details
4 Experiments
4.1 Quantitative Comparison
4.2 Qualitative Comparison
4.3 Why Spiral Is Necessary
4.4 Analysis of SliceGAN
4.5 Efficacy of Hue-Color Loss
5 Conclusion and Limitations
References
SceneSketcher: Fine-Grained Image Retrieval with Scene Sketches
1 Introduction
2 Related Work
3 Methods
3.1 Overview
3.2 Scene Graph Generation
3.3 Graph Encoder
3.4 Graph Similarity Function
3.5 Category-Wise IoU
3.6 Loss Function
4 Experiments
4.1 Datasets
4.2 Evaluation Metrics
4.3 Comparison with Baselines
4.4 Ablation Study
4.5 Results on Our Extended Scene Sketch Database
5 Conclusion
References
Few-Shot Compositional Font Generation with Dual Memory
1 Introduction
2 Related Works
2.1 Few-Shot Image-to-Image Translation
2.2 Automatic Font Generation
3 Preliminary: Complete Compositional Scripts
4 Dual Memory-Augmented Font Generation Network
4.1 Architecture Overview
4.2 Learning
5 Experiments
5.1 Datasets
5.2 Comparison Methods and Evaluation Metrics
5.3 Main Results
5.4 More Analysis
6 Conclusions
References
PUGeo-Net: A Geometry-Centric Network for 3D Point Cloud Upsampling
1 Introduction
2 Related Work
3 Proposed Method
3.1 Overview
3.2 Hierarchical Feature Learning and Recalibration
3.3 Parameterization-Based Point Expansion
3.4 Updating Samples via Local Shape Approximation
3.5 Joint Loss Optimization
4 Experimental Results
4.1 Experiment Settings
4.2 Comparison with State-of-the-Art Methods
4.3 Effectiveness of Normal Prediction
4.4 Robustness Analysis
4.5 Ablation Study
5 Conclusion and Future Work
References
Handcrafted Outlier Detection Revisited
1 Introduction
2 Related Work
3 Method
3.1 Preliminaries and Core Assumptions
3.2 Seed Points Selection
3.3 Local Neighborhood Selection, Filtering and Validation
3.4 Adaptive Assumption Relaxation
4 Experiments
4.1 Evaluation Pipeline
4.2 Datasets
4.3 Comparison with the State of the Art
4.4 Ablation Studies
5 Conclusions
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