Computer Vision - ECCV 2020

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

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

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

Springer (Publisher)

Published on 11. November 2020

XLII, 799 pages

E-Book

PDF with digital watermarking

System requirements

978-3-030-58565-5 (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 XX
The Average Mixing Kernel Signature
1 Introduction
2 Related Work
3 Quantum Mechanical Background
3.1 Quantum Walks and the Schrödinger equation
3.2 Average Mixing Matrix
4 Average Mixing Kernel Signature
5 Experiments
5.1 Choice of Time
5.2 Quantitative Evaluation
5.3 Qualitative Evaluation
5.4 Runtime Analysis
6 Conclusions
References
BCNet: Learning Body and Cloth Shape from a Single Image
1 Introduction
2 Related Work
3 Algorithm
3.1 Garment Model
3.2 Image to Dressed Body
3.3 Skinning Weight Network
3.4 Displacement Network
3.5 Loss Function
4 Dataset Construction
4.1 Skinning Weight Dataset
4.2 Synthetic Dataset Construction
4.3 HD Texture Dataset
5 Experiments
5.1 Analysis of BCNet
5.2 Quantitative Comparison.
5.3 Qualitative Results
6 Conclusion
References
Self-supervised Keypoint Correspondences for Multi-person Pose Estimation and Tracking in Videos
1 Introduction
2 Related Work
3 Method Overview
4 Keypoint Correspondence Network
4.1 Siamese Matching Module
4.2 Refinement Module
4.3 Training
5 Multi-person Pose Tracking
5.1 Recover Missed Detections
5.2 Tracking
6 Experiments and Results
6.1 Implementation Details
6.2 Baselines
6.3 Effect of Joint Detection Threshold and Pose Recovery
6.4 Comparison with State-of-the-Art Methods
6.5 Qualitative Results
7 Conclusion
References
Interactive Multi-dimension Modulation with Dynamic Controllable Residual Learning for Image Restoration
1 Introduction
2 Related Work
3 Method
4 Experiments
4.1 Implementation Details
4.2 Complexity Analysis
4.3 Performance Evaluation
4.4 Ablation Study
5 Conclusion
References
Polysemy Deciphering Network for Human-Object Interaction Detection
1 Introduction
2 Related Works
3 Our Method
3.1 Overview
3.2 Representation and Classification Networks
3.3 Polysemy Attention Module
3.4 Object-SPecific Verb Classification Module
3.5 Clustering-Based SP-VCM
3.6 Training and Testing
4 Experiments
4.1 Datasets and Metrics
4.2 Implementation Details
4.3 Ablation Studies
4.4 Comparisons with State-of-the-Art Methods
4.5 Qualitative Visualization Results
5 Conclusions
References
PODNet: Pooled Outputs Distillation for Small-Tasks Incremental Learning
1 Introduction
2 Related Work
3 Model
3.1 POD: Pooled Outputs Distillation Loss
3.2 Local Similarity Classifier
3.3 Complete Model Formulation
4 Experiments
4.1 Quantitative Results
4.2 Further Analysis and Ablation Studies
5 Conclusion
References
Learning Graph-Convolutional Representations for Point Cloud Denoising
1 Introduction
2 Related Work
3 Proposed Method
3.1 Architecture
3.2 Graph-Convolutional Layer
3.3 Loss Functions
4 Experimental Results
4.1 Experimental Setting
4.2 Comparisons with State-of-the-Art
4.3 Ablation Studies
4.4 Feature Analysis
4.5 Structured Noise
5 Conclusions
References
Semantic Line Detection Using Mirror Attention and Comparative Ranking and Matching
1 Introduction
2 Related Work
2.1 Line Detection
2.2 Attention Mechanisms in CNNs
2.3 Metric Learning and Order Learning
3 Proposed Algorithm
3.1 D-Net: Semantic Line Detection with Mirror Attention
3.2 R-Net and M-Net: Comparative Ranking and Matching
4 Experimental Results
4.1 Datasets
4.2 Semantic Line Detection Results
4.3 Ablation Studies
5 Applications
5.1 Dominant Parallel Lines
5.2 Reflection Symmetry Axes
6 Conclusions
References
A Differentiable Recurrent Surface for Asynchronous Event-Based Data
1 Introduction
2 Related Work
3 Method
3.1 Matrix-LSTM
4 Evaluation
4.1 Object Classification
4.2 Optical Flow Prediction
4.3 Time Performance Analysis
5 Conclusion
References
Fine-Grained Visual Classification via Progressive Multi-granularity Training of Jigsaw Patches
1 Introduction
2 Related Work
2.1 Fine-Grained Classification
2.2 Image Splitting Operations
2.3 Progressive Training
3 Approach
3.1 Progressive Training
3.2 Jigsaw Puzzle Generator
3.3 Inference
4 Experimental Results and Discussion
4.1 Implementation Details
4.2 Comparisons with State-of-the-Art Methods
4.3 Ablation Study
4.4 Discussions
4.5 Visualization
5 Conclusions
References
LiteFlowNet3: Resolving Correspondence Ambiguity for More Accurate Optical Flow Estimation
1 Introduction
2 Related Work
3 LiteFlowNet3
3.1 Preliminaries
3.2 Cost Volume Modulation
3.3 Flow Field Deformation
4 Experiments
4.1 Results
4.2 Ablation Study
5 Conclusion
References
Microscopy Image Restoration with Deep Wiener-Kolmogorov Filters
1 Introduction
2 Related Work
3 Problem Formulation
3.1 Image Formation in Microscopy
3.2 Regularization
4 Proposed Approach
4.1 Learnable Regularization Kernels
4.2 Prediction of Regularization Kernels
4.3 Prediction of Spatially Adaptive Regularization Kernels
4.4 Prediction of the Gradient of the Regularizer
5 Network Training
5.1 Dataset
5.2 Training Details
5.3 Evaluation
6 Results
7 Poisson Image Deblurring
8 Conclusion
References
ScanRefer: 3D Object Localization in RGB-D Scans Using Natural Language
1 Introduction
2 Related Work
3 Task
4 Dataset
4.1 Data Collection
4.2 Dataset Statistics
5 Method
5.1 Data Representations
5.2 Network Architecture
5.3 Loss Function
5.4 Training and Inference
6 Experiments
6.1 Task Difficulty
6.2 Quantitative Analysis
6.3 Qualitative Analysis
6.4 Ablation Studies
7 Conclusion
References
JSENet: Joint Semantic Segmentation and Edge Detection Network for 3D Point Clouds
1 Introduction
2 Related Work
2.1 3D Semantic Segmentation
2.2 2D Semantic Edge Detection
2.3 Joint Learning of Segmentation and Edge Detection
3 JSENet
3.1 Semantic Segmentation Stream
3.2 Semantic Edge Detection Stream
3.3 Joint Refinement Module
3.4 Joint Multi-task Learning
4 Experiments
4.1 Datasets and Metrics
4.2 Implementation Details
4.3 Ablation Study
4.4 Results on S3DIS & ScanNet Datasets
5 Conclusions
References
Motion-Excited Sampler: Video Adversarial Attack with Sparked Prior
1 Introduction
2 Related Work
3 Method
3.1 Problem Formulation
3.2 Motion Map Generation
3.3 Motion-Excited Sampler
3.4 Gradient Estimation and Optimization
3.5 Loss Function
4 Experiments
4.1 Experimental Setting
4.2 Comparison to State-of-the-Art
4.3 Targeted Attack
4.4 Ablation Study
5 Conclusion
References
An Inference Algorithm for Multi-label MRF-MAP Problems with Clique Size 100
1 Introduction
2 Background
3 Properties of the Multi-label to 2-Label Transformation
4 Representing Invalid Extreme Bases
4.1 Canonical Ordering and Its Properties
4.2 Elementary Invalid Extreme Base
5 The Multi-label Hybrid Algorithm
5.1 Computing Min 2 Norm by Flow
5.2 Overall Algorithm
6 Experiments
7 Conclusions
References
Dual Refinement Underwater Object Detection Network
1 Introduction
2 Related Work
3 Method
3.1 Composite Connection Backbone
3.2 Receptive Field Augmentation Module
3.3 Prediction Refinement Scheme
4 Experiments
4.1 Implementation Details
4.2 Detection Performance
4.3 Ablation Experiment
5 Conclusion
References
Multiple Sound Sources Localization from Coarse to Fine
1 Introduction
2 Related Work
3 Approach
3.1 Multi-task Training Framework
3.2 Audiovisual Feature Alignment
3.3 Sound Localization and Its Application in Separation
4 Experiments
4.1 Datasets
4.2 Implementation Details
4.3 Sound Localization
4.4 Sound Separation
5 Conclusions
References
Task-Aware Quantization Network for JPEG Image Compression
1 Introduction
2 Related Work
3 JPEG Compression
3.1 Encoder
3.2 Decoder
4 Quantization Network
4.1 Framework
4.2 Network Architecture
5 Approximation of Bitrate Measure
5.1 Symbol Prediction from RLE Model
5.2 Regression Model for Final Code Length
6 Experiments
7 Conclusion
References
Energy-Based Models for Deep Probabilistic Regression
1 Introduction
2 Background and Related Work
3 Proposed Regression Method
3.1 Formulation
3.2 Training
3.3 Prediction
4 Experiments
4.1 Object Detection
4.2 Visual Tracking
4.3 Age Estimation
4.4 Head-Pose Estimation
5 Conclusion
References
CLOTH3D: Clothed 3D Humans
1 Introduction
2 Related Work
3 Dataset
3.1 Human 3D Sequences
3.2 Garment Generation
3.3 Simulation
3.4 Additional Dataset Statistics
4 Dressed Human Generation
4.1 Data Pre-processing
4.2 SMPL-Skirt Topology
4.3 Network
5 Experiments
5.1 Ablation Study
6 Conclusions
References
Encoding Structure-Texture Relation with P-Net for Anomaly Detection in Retinal Images
1 Introduction
2 Related Work
2.1 Anomaly Detection
2.2 Structure-Texture Relation Encoding Networks
3 Method
3.1 Structure Extraction from Original Image Module
3.2 Image Reconstruction Module
3.3 Structure Extraction from Reconstructed Image Module
3.4 Objective Function
3.5 Anomaly Detection for Testing Data
4 Experiments
4.1 Implementation
4.2 Evaluation Metric
4.3 Anomaly Detection in Retinal Images
4.4 Anomaly Detection in Real World Images
5 Conclusion
References
CLNet: A Compact Latent Network for Fast Adjusting Siamese Trackers
1 Introduction
2 Related Work
2.1 Siamese Network Based Trackers
2.2 Meta-Learning
3 Siamese Tracker with Compact Latent Network
3.1 Revisiting SiamRPN for Tracking
3.2 Analysis for Siamese-Based Training Method
3.3 Compact Latent Network
3.4 Training with Diverse Sample Mining
4 Experimental Results
4.1 Implementation Details
4.2 Comparison with Other Trackers
4.3 Ablation Study
5 Conclusion
References
Occlusion-Aware Siamese Network for Human Pose Estimation
1 Introduction
2 Related Work
3 Method
3.1 Feature Erasing and Feature Reconstruction
3.2 Siamese Framework
3.3 Optimal Transport Divergence
3.4 Training and Inference
4 Experiments
4.1 Datasets
4.2 Implementations
4.3 Benchmark Results
4.4 Ablation Study
5 Conclusion
References
Learning to Predict Salient Faces: A Novel Visual-Audio Saliency Model
1 Introduction
2 Related Work
3 The Proposed Dataset
3.1 Data Collection
3.2 Database Analysis
4 The Proposed Method
4.1 Architecture
4.2 Optimization
5 Experiments and Results
5.1 Settings
5.2 Performance Comparison
5.3 Ablation Analysis
6 Conclusion
References
NormalGAN: Learning Detailed 3D Human from a Single RGB-D Image
1 Introduction
2 Related Work
3 Overveiw
4 Method
4.1 Dataset Generation
4.2 Front-View RGB-D Rectification
4.3 Back-View RGB-D Infernence
4.4 Loss Functions
5 Experiments
5.1 Ablation Study
5.2 Comparison
5.3 Training Details and Runtime Performance
6 Conclusion
References
Model-Based Occlusion Disentanglement for Image-to-Image Translation
1 Introduction
2 Related Work
3 Model-Based Disentanglement
3.1 Adversarial Disentanglement
3.2 Adversarial Parameters Estimation
3.3 Disentanglement Guidance
4 Experiments
4.1 Training Setup
4.2 Raindrops
4.3 Extension to Other Occlusion Models
4.4 Ablation Studies
5 Conclusion
References
Rotation-Robust Intersection over Union for 3D Object Detection
1 Introduction
2 Related Work
3 Approach
3.1 Rotation-Robust Intersection over Union
3.2 Discussion
4 Experiment
4.1 Datasets and Settings
4.2 Results and Analysis
4.3 Ablation Study
4.4 Applications in 2D Object Detection
5 Conclusion
References
New Threats Against Object Detector with Non-local Block
1 Introduction
2 Related Work
2.1 Non-local Neural Networks
2.2 Adversarial Examples
3 Methodology
3.1 Faster R-CNN and Non-local Block
3.2 Disappearing Attack
3.3 Appearing Attack
4 Experiments
4.1 Digital Attack
4.2 Physical Attack
4.3 Effective Attack Regions
5 Analysis of the Effective Attack Regions
6 Conclusion
References
Self-Supervised CycleGAN for Object-Preserving Image-to-Image Domain Adaptation
1 Introduction
2 Related Work
2.1 Self-supervised Learning
2.2 Unpaired Image-to-image Translation
3 Revisiting the Problem of CycleGAN
4 OP-GAN
4.1 Multi-task Self-supervised Learning
4.2 Generator and Discriminator
4.3 Objective Function
5 Experiments
5.1 Experiment Settings
5.2 Visualization of Adaptation Results
5.3 Cloudy-to-sunny Adaptation on CamVid
5.4 Night-to-day Adaptation on SYNTHIA
5.5 Multicentre Colonoscopy Adaptation
5.6 Ablation Study
6 Conclusion
References
On the Usage of the Trifocal Tensor in Motion Segmentation
1 Introduction
1.1 Related Work
1.2 Contribution
2 Proposed Methods
2.1 TriSeg
2.2 TriPairSeg
3 Experiments
3.1 Implementation Details
3.2 Existing Datasets
3.3 Novel Benchmark
4 Conclusion
References
3D-Rotation-Equivariant Quaternion Neural Networks
1 Introduction
2 Related Work
3 Approach
3.1 Quaternion Features in Neural Networks and Rotations
3.2 Rotation Equivariance
3.3 Rules for Rotation Equivariance
3.4 Rules for Permutation Invariance
3.5 Overview of the REQNN
3.6 Revisions of Traditional Neural Networks into REQNNs
4 Experiments
5 Conclusion
References
InterHand2.6M: A Dataset and Baseline for 3D Interacting Hand Pose Estimation from a Single RGB Image
1 Introduction
2 Related Works
3 InterHand2.6M
3.1 Data Capture
3.2 Annotation
3.3 Dataset Release
4 InterNet
4.1 Handedness Estimation
4.2 2.5D Right and Left Hand Pose Estimation
4.3 Right Hand-Relative Left Hand Depth Estimation
4.4 Final 3D Interacting Hand Pose
4.5 Loss Functions
5 Implementation Details
6 Experiment
6.1 Dataset and Evaluation Metric
6.2 Ablation Study
6.3 Comparison with State-of-the-Art Methods
6.4 Evaluation on InterHand2.6M
6.5 3D Interacting Hand Pose Estimation from General Images
7 Conclusion
References
Active Crowd Counting with Limited Supervision
1 Introduction
2 Related Works
2.1 Crowd Counting
2.2 Semi-supervised Learning
2.3 Active Learning
3 Method
3.1 Problem
3.2 Overview
3.3 Partition-Based Sample Selection with Weights (PSSW)
3.4 Distribution Alignment with Latent MixUp
3.5 Loss Function
4 Experiments
4.1 Experimental Setup
4.2 ShanghaiTech
4.3 UCF_CC_50
4.4 Mall
5 Discussion
References
Self-supervised Monocular Depth Estimation: Solving the Dynamic Object Problem by Semantic Guidance
1 Introduction
2 Related Work
3 Method
3.1 Self-supervised Monocular Depth Estimation
3.2 Supervised Semantic Segmentation
3.3 Semantic Guidance
4 Experimental Setup
5 Evaluation and Discussion
5.1 Depth Evaluation w.r.t. the Baselines
5.2 Ablation Studies
5.3 Semantics Evaluation
6 Conclusion
References
Hierarchical Visual-Textual Graph for Temporal Activity Localization via Language
1 Introduction
2 Related Work
3 Proposed Approach
3.1 Hierarchical Visual-Textual Graph
3.2 Sentence Localizer
3.3 Losses
4 Experiments
4.1 Datasets and Experimental Settings
4.2 Ablation Study
4.3 Comparison with State-of-the-Art Methods
4.4 Result Visualizations
5 Conclusions
References
Do Not Mask What You Do Not Need to Mask: A Parser-Free Virtual Try-On
1 Introduction
2 Problem Statement and Related Work
3 Our Approach
3.1 WUTON Architecture
3.2 Training Procedure of the Teacher Model
3.3 Training Procedure of the Student Model
4 Experiments and Analysis
4.1 Dataset
4.2 Visual Results
4.3 Quantitative Results
4.4 User Study
4.5 Runtime Analysis
4.6 The Impact of Adversarial Loss in the Teacher-Student Setting
5 Conclusion
References
NODIS: Neural Ordinary Differential Scene Understanding
1 Introduction
2 Related Work
3 Method
3.1 Scene Graphs
3.2 Models for Object and Relationship Detection
3.3 Neural Ordinary Differential Equations ch38Chen2018:NODE
3.4 Assignments by Neural Ordinary Differential Equations
3.5 Architecture
4 Experiments
4.1 Dataset, Settings, and Evaluation
4.2 Quantitative Results and Comparison
4.3 Ablation Studies
4.4 Neural ODE Analysis
4.5 Qualitative Results
5 Conclusions
References
AssembleNet++: Assembling Modality Representations via Attention Connections
1 Introduction
2 Previous Work
3 Approach
3.1 Preliminaries
3.2 Input Modalities and Semantics
3.3 Learning Weighted Connections
3.4 Attention Connectivity and Peer-Attention
3.5 One-Shot Attention Search Model
3.6 Model Implementation Details
4 Experimental Results
4.1 Using Object Modality
4.2 Attention Search
4.3 Comparison to the State-of-the-Art
4.4 Ablation
4.5 General Applicability of the Findings
5 Conclusion
References
Learning Propagation Rules for Attribution Map Generation
1 Introduction
2 Related Work
3 Method
3.1 Step 1: Forwarding the Input Image
3.2 Step 2: Backwarding the Gradients of the Input
3.3 Step 3: Mask Generation
3.4 Step 4: Forwarding the Masked Image
4 Experiments
4.1 Evaluation Protocols
4.2 Implementation Details
4.3 Compared Methods
4.4 Experimental Results
5 Conclusion
References
Reparameterizing Convolutions for Incremental Multi-Task Learning Without Task Interference
1 Introduction
2 Related Work
3 Reparameterizing CNNs for Multi-Task Learning
3.1 Problem Formulation
3.2 Task Interference
3.3 Reparameterizing Convolutions
4 Experiments
4.1 Datasets
4.2 Architecture
4.3 Evaluation Metric
4.4 Analysis of Network Module Sharing
4.5 Ablation Study
4.6 Comparison to State-of-the-Art
4.7 Incremental Learning for Multi-tasking
5 Conclusion
References
Learning Predictive Models from Observation and Interaction
1 Introduction
2 Related Work
3 Learning Predictive Models from Observation and Interaction
3.1 Graphical Model
3.2 Domain Shift
3.3 Deep Neural Network Implementation
4 Experiments
4.1 Visual Prediction for Driving
4.2 Robotic Manipulation: Prediction
4.3 Robotic Manipulation: Planning and Control
5 Conclusion
References
Unifying Deep Local and Global Features for Image Search
1 Introduction
2 Related Work
3 DELG
3.1 Design Considerations
3.2 Model
3.3 Training
4 Experiments
4.1 Experimental Setup
4.2 Results
5 Conclusions
References
Human Body Model Fitting by Learned Gradient Descent
1 Introduction
2 Related Work
3 Method
4 Experiments
4.1 Training Data Comparison
4.2 Test Datasets
4.3 Ablation and Iteration Studies
4.4 Comparison with Other Methods
4.5 Qualitative Results
4.6 Speed and Model Size
5 Conclusion
References
DDGCN: A Dynamic Directed Graph Convolutional Network for Action Recognition
1 Introduction
2 Related Work
3 Methodology
3.1 Overview
3.2 Dynamic Convolutional Sampling (DCS)
3.3 Dynamic Convolutional Weights
3.4 Directed Spatial-Temporal Graph
3.5 Network Architecture
4 Experimental Results
4.1 Datasets and Evaluation Metrics
4.2 Comparison with Existing Methods on Benchmarks
4.3 Real-Time Experiments
4.4 Ablation Study
4.5 Recognition of Incomplete Actions
5 Conclusions
References
Learning Latent Representations Across Multiple Data Domains Using Lifelong VAEGAN
1 Introduction
2 Related Works
3 The Lifelong VAEGAN
3.1 Problem Formulation
3.2 Data Generation from Prior Distributions
3.3 Training the Inference Model
4 Theoretical Analysis of the GRM
5 The Two-Step Latent Variables Optimization over Time
5.1 Supervised Learning
5.2 Semi-supervised Learning
5.3 Unsupervised Learning
6 Experimental Results
6.1 Lifelong Unsupervised Learning
6.2 Lifelong Supervised Learning
6.3 Lifelong Semi-supervised Learning
6.4 Analysis
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