Computer Vision - ECCV 2020

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

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

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

Springer (Publisher)

Published on 6. October 2020

XLIII, 803 pages

E-Book

PDF with digital watermarking

System requirements

978-3-030-58526-6 (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 XXIX
Procrustean Regression Networks: Learning 3D Structure of Non-rigid Objects from 2D Annotations
1 Introduction
2 Related Works
3 Method
3.1 Procrustean Regression
3.2 PR Loss for Neural Networks
3.3 Design of f and g
3.4 Network Structure
4 Experiments
4.1 Datasets
4.2 Implementation Details
4.3 Results
5 Conclusion
References
Learning to Learn Parameterized Classification Networks for Scalable Input Images
1 Introduction
2 Related Work
3 Approach
3.1 Network Architecture
3.2 Scale Distillation
3.3 Inference
4 Experiments
4.1 Main Results
4.2 Ablation Studies
5 Conclusion
References
Stereo Event-Based Particle Tracking Velocimetry for 3D Fluid Flow Reconstruction
1 Introduction
2 Related Work
3 Proposed Method
3.1 System Overview
3.2 Image Formation Model
3.3 Event-Based Particle Tracking
3.4 Stereo Matching
3.5 3D Velocimetry Reconstruction
4 Experiments
4.1 Synthetic Data
4.2 Captured Data
5 Conclusions
References
Simplicial Complex Based Point Correspondence Between Images Warped onto Manifolds
1 Introduction
2 Preliminaries
3 Building a Simplicial Complex on a Curved Manifold
4 Assignment Algorithm
5 Experiments
6 Conclusion
References
Representation Learning on Visual-Symbolic Graphs for Video Understanding
1 Introduction
2 Related Work
3 Method
3.1 Visual Context Module
3.2 Semantic Context Module
4 Experiments
4.1 Experiments on CAD-120
4.2 Experiments on Charades
4.3 Experiments on ActivityNet Entities
5 Conclusions
References
Distance-Normalized Unified Representation for Monocular 3D Object Detection
1 Introduction
2 Related Work
2.1 2D Object Detection
2.2 Monocular 3D Object Detection
2.3 Cascaded Point Regression
3 Proposed UR3D
3.1 Basic Framework
3.2 Distance-Normalized Unified Representation
3.3 Distance-Guided NMS
3.4 Fully Convolutional Cascaded Point Regression
3.5 Implementation Details
4 Experiments
4.1 KITTI
4.2 Ablation Study
5 Conclusions
References
Sequential Deformation for Accurate Scene Text Detection
1 Introduction
2 Related Work
3 Methodology
3.1 Sequential Deformation Module
3.2 Auxiliary Character Counting Supervision
3.3 Mask R-CNN with SDM
4 Experiments
4.1 Datasets
4.2 Implementation Details
4.3 Ablation Study
4.4 Comparative Results on Public Benchmarks
4.5 Discussion for SDM's Adaptability
5 Conclusion
References
Where to Explore Next? ExHistCNN for History-Aware Autonomous 3D Exploration
1 Introduction
2 Related Work
3 Proposed Method
3.1 Representation for 3D Reconstruction History
3.2 ExHistCNN
3.3 NBV for 3D Exploration
4 Experiments
4.1 Dataset Generation
4.2 ExHistCNN Ablation Study
4.3 Autonomous 3D Exploration Performance
5 Conclusion
References
Semi-supervised Segmentation Based on Error-Correcting Supervision
1 Introduction
2 Related Work
2.1 Supervised Semantic Segmentation
2.2 Weakly-Supervised Segmentation
2.3 Semi-supervised Segmentation
3 Error-Correcting Supervision
3.1 Error-Correcting Network
3.2 Supervised Training with an Auxiliary Objective
3.3 Semi-supervised Step
4 Experiments and Analysis
4.1 Cityscapes
4.2 Pascal VOC 2012
4.3 Model Architecture
4.4 Setup
4.5 Results
4.6 Ablation Study
4.7 Comparison with Existing Methods
4.8 Relation Between Correction and Truth
5 Conclusion
References
Quantum-Soft QUBO Suppression for Accurate Object Detection
1 Introduction
2 Related Work
3 Approach
3.1 Quadratic Unconstrained Binary Optimization
3.2 Quantum Annealing
3.3 Quantum-Soft QUBO Suppression
4 Experiments
4.1 Datasets and Evaluation Metrics
4.2 Implementation Details
4.3 Ablation Study
4.4 Results
5 Conclusion
References
Label-Similarity Curriculum Learning
1 Introduction
2 Related Work
3 Method
3.1 Document Embedding for Defining Label Similarity
3.2 Label Encoding
3.3 Loss Function
4 Experiments
5 Results and Discussion
6 Conclusions
References
Recurrent Image Annotation with Explicit Inter-label Dependencies
1 Introduction
2 Related Work
3 Approach
3.1 Background: CNN-RNN Framework
3.2 Multi-order RNN
3.3 Discussion
4 Experiments
4.1 Datasets
4.2 Evaluation Metrics
4.3 Implementation Details
4.4 Results and Discussion
5 Summary and Conclusion
References
Cross-Attention in Coupled Unmixing Nets for Unsupervised Hyperspectral Super-Resolution
1 Introduction
2 Related Work
2.1 Conventional Methods
2.2 DL-Based Methods
3 Coupled Unmixing Nets with Cross-Attention
3.1 Method Overview
3.2 Problem Formulation
3.3 Network Architecture
3.4 Network Training
4 Experimental Results
4.1 Ablation Study
4.2 Comparative Experiments
5 Conclusion
References
SimPose: Effectively Learning DensePose and Surface Normals of People from Simulated Data
1 Introduction
2 Related Work
3 Simulated Multi-person Dense Pose Dataset
3.1 Human 3D Models
4 Our Approach
4.1 Human Pose Estimation Meta-architectures
4.2 Human Pose Estimation Tasks
4.3 Multi-task Learning and Batch Mixture Normalization
5 Evaluation
5.1 Experimental Setup
5.2 Evaluation of SimPose's UV Prediction
5.3 Evaluation of SimPose's Surface Normal Prediction
6 Conclusion
References
ByeGlassesGAN: Identity Preserving Eyeglasses Removal for Face Images
1 Introduction
2 Related Works
2.1 Face Attributes Manipulation
2.2 Image Completion
3 ByeGlassesGAN
3.1 Proposed Framework
3.2 Objective Function
3.3 Network Architecture
4 Synthesis of Face Images with Eyeglasses
5 Experimental Results
5.1 Implementation Details
5.2 Qualitative Results
5.3 Quantitative Results
6 Face Recognition Evaluation
7 Conclusions
References
Differentiable Joint Pruning and Quantization for Hardware Efficiency
1 Introduction
2 Related Work
3 Differentiable Joint Pruning and Quantization
3.1 Quantization with Learnable Mapping
3.2 Structured Pruning via VIBNet Gates
3.3 Evaluation Metrics
3.4 Joint Optimization of Pruning and Quantization
3.5 Power-of-Two Quantization
4 Experiments
4.1 Comparison of DJPQ with Quantization only Schemes
4.2 Comparison of DJPQ with Two-Stage Optimization
4.3 Comparison of DJPQ with Others Under Bit Restriction
4.4 Analysis of Learned Distributions
5 Conclusion
A Quantization scheme in DJPQ
B Experimental details
B.1 Comparison of DJPQ with DQ
B.2 DJPQ Results for MobileNetV2
B.3 Experimental Setup
References
Learning to Generate Customized Dynamic 3D Facial Expressions
1 Introduction
2 Related Work
3 Learnable Mesh Operators: Background
3.1 Spiral Convolution Networks
3.2 Mesh Unpooling Operations
4 Model
5 Experiments
5.1 Dynamic 3D Face Database
5.2 Dynamic Facial Expressions
5.3 Classification of Generated 4D Expressions
5.4 Loss per Frame
5.5 Interpolation on the Latent Space
5.6 Expression Generation In-the-Wild
6 Limitations and Future Work
7 Conclusion
References
LandscapeAR: Large Scale Outdoor Augmented Reality by Matching Photographs with Terrain Models Using Learned Descriptors
1 Introduction
2 Related Work
2.1 Visual Localization
2.2 Local Descriptors
2.3 Cross-Domain Matching
3 Method
3.1 Dataset Generation
3.2 Weakly Supervised Cross-domain Patch Sampling
3.3 Architecture
3.4 Training
3.5 Pose Estimation
4 Experiments
4.1 Test Datasets
4.2 Ablation Studies
4.3 Comparison with State-of-the-Art
5 Applications
6 Conclusion and Future Work
References
Learning Disentangled Feature Representation for Hybrid-Distorted Image Restoration
1 Introduction
2 Related Work
2.1 Image Restoration on Single Distortion
2.2 Image Restoration on Hybrid Distortion
3 Approach
3.1 Primary Knowledge
3.2 Feature Disentanglement
3.3 Feature Aggregation Module
3.4 Auxiliary Module
3.5 Overview of Whole Framework
3.6 Loss Function
4 Experiments
4.1 Dataset for Hybrid-Distorted Image Restoration
4.2 Datasets for Single Distortion Restoration
4.3 Implementation Details
4.4 Comparison with State-of-the-Arts
4.5 Interpretative Experiment
4.6 Experiments on Single Distortion
4.7 Ablation Studies
5 Conclusion
References
Jointly De-Biasing Face Recognition and Demographic Attribute Estimation
1 Introduction
2 Related Work
3 Methodology
3.1 Problem Definition
3.2 Algorithm Design
3.3 Network Architecture
3.4 Adversarial Training and Disentanglement
4 Experiments
4.1 Datasets and Pre-processing
4.2 Implementation Details
4.3 De-Biasing Face Verification
4.4 De-Biasing Demographic Attribute Estimation
4.5 Analysis of Disentanglement
4.6 Face Verification on Public Testing Datasets
5 Conclusion
References
Regularized Loss for Weakly Supervised Single Class Semantic Segmentation
1 Introduction
2 Related Work
3 Our Approach
3.1 Regularized Losses
3.2 Loss and Segmentation Accuracy Correlation
3.3 Complete Loss Function
3.4 Training
4 Experimental Results
4.1 Saliency Datasets
4.2 Cosegmentation
4.3 Semantic Segmentation
5 Conclusions
References
Spike-FlowNet: Event-Based Optical Flow Estimation with Energy-Efficient Hybrid Neural Networks
1 Introduction
2 Related Work
3 Method
3.1 Spiking Neuron Model
3.2 Spiking Input Event Representation
3.3 Self-Supervised Loss
3.4 Spike-FlowNet Architecture
3.5 Backpropagation Training in Spike-FlowNet
4 Experimental Results
4.1 Dataset and Training Details
4.2 Algorithm Evaluation Metric
4.3 Average End-Point Error (AEE) Results
4.4 Computational Efficiency
5 Conclusion
References
Forgetting Outside the Box: Scrubbing Deep Networks of Information Accessible from Input-Output Observations
1 Introduction
2 Related Work
3 Out of the Box Forgetting
3.1 Information Theoretic Formalism
3.2 Bound for Activations
3.3 Close Form Bound for Gaussian Scrubbing
4 An NTK-Inspired Forgetting Procedure
4.1 Relation Between NTK and Fisher Forgetting
5 Experiments
5.1 Datasets
5.2 Models and Training
5.3 Baselines
5.4 Readout Functions
5.5 Results
6 Discussion
References
Inherent Adversarial Robustness of Deep Spiking Neural Networks: Effects of Discrete Input Encoding and Non-linear Activations
1 Introduction
2 Background
2.1 Adversarial Attack
2.2 Spiking Neural Network (SNN)
3 Experiments
3.1 Dataset and Models
3.2 Training Procedure
3.3 Adversarial Input Generation Methodology
4 Results
4.1 ANN vs SNN
4.2 ANN-Crafted vs SNN-Crafted Attack
5 Conclusions
References
Synthesizing Coupled 3D Face Modalities by Trunk-Branch Generative Adversarial Networks
1 Introduction
2 Related Work
2.1 3D Face Modeling
2.2 Photorealistic Face Synthesis
2.3 Boosting Face Recognition by Synthetic Training Data
3 Approach
3.1 UV Maps for Shape, Texture and Normals
3.2 Trunk-Branch GAN to Generate Coupled Texture, Shape and Normals
3.3 Expression Augmentation by Conditional GAN
3.4 Photorealistic Rendering with Generated UV Maps
4 Results
4.1 Qualitative Results
4.2 Pose-Invariant Face Recognition
5 Conclusion
References
Learning to Learn Words from Visual Scenes
1 Introduction
2 Related Work
3 Learning to Learn Words
3.1 Episodes
3.2 Model
3.3 Learning Objectives
3.4 Information Flow
3.5 Inference
4 Experiments
4.1 Datasets
4.2 Baselines
4.3 Acquisition of New Words
4.4 Acquisition of Familiar Words
4.5 Compositionality
4.6 Retrieval
4.7 Analysis
5 Discussion
References
On Transferability of Histological Tissue Labels in Computational Pathology
1 Introduction
1.1 Related Works
2 Transferring Diagnostically-Relevant Labels
2.1 Source Domain: ADP
2.2 Task 1: Tissue Classification
2.3 Task 2: Disease Classification
3 YCbCr Color Augmentation
4 HistoNet for ADP Source Domain
4.1 Neural Architecture Search (NAS) for HistoNet
4.2 Choice of Pixel Resolution
4.3 Network Performance Comparison
5 Experiments on HTT Transferability
5.1 Image Modifications and Pixel-Resolution Adjustment
5.2 Transferability in CRC Dataset for Tissue Classification
5.3 Transferability in HMT Dataset for Tissue Classification
5.4 Transferability in GlaS Dataset for Cancer Classification
5.5 Alternative Source Domain Choice
6 Cancer Detection on WSI Level
7 Concluding Remarks
References
Learning Actionness via Long-Range Temporal Order Verification
1 Introduction
2 Related Work
2.1 Self-supervised Learning
2.2 Learning from Instructional Videos
2.3 Action Proposals
3 Unsupervised Learning of Actionness Score
3.1 Models for Actionness and Order Verification
3.2 Training with Ordering Verification
4 Experiments
4.1 Experimental Setup
4.2 Ablations Studies
4.3 Actionness Score for Practical Applications
4.4 Qualitative Results
5 Conclusion
References
Fully Embedding Fast Convolutional Networks on Pixel Processor Arrays
1 Introduction
2 SCAMP-5 Overview
3 Related Work
4 Parallel Convolutional Layer Computation
4.1 Computational Layout on PE Array
4.2 In-Pixel Filter Weights
4.3 Parallel ReLU and Max Pooling
4.4 Further Convolutional Layers
4.5 Feature Map Shrinking and Duplication
5 Parallel Fully Connected Layer Computation
5.1 Activation Value Summation
6 Results
6.1 MNIST Network Training
6.2 Inference on SCAMP-5 Hardware
7 Conclusions
References
Character Region Attention for Text Spotting
1 Introduction
2 Related Work
3 Methodology
3.1 Overview
3.2 Detection Stage
3.3 Sharing Stage
3.4 Recognition Stage
4 Experiment
4.1 Datasets
4.2 Training Strategy
4.3 Experimental Results
4.4 Ablation Study
4.5 Discussions
5 Conclusion
References
Stable Low-Rank Tensor Decomposition for Compression of Convolutional Neural Network
1 Introduction
1.1 Why CPD
1.2 Why Not Standard CPD
1.3 Contributions
2 Stable Tensor Decomposition Method
2.1 CP Decomposition of Convolutional Kernel
2.2 Degeneracy and Its Effect to CNN Stability
2.3 Stabilization Method
2.4 Tucker Decomposition with Bound Constraint
3 Implementation
4 Experiments
4.1 Layer-Wise Study
4.2 Full Model Compression
5 Discussion and Conclusions
References
Dual Mixup Regularized Learning for Adversarial Domain Adaptation
1 Introduction
2 Related Work
2.1 Interpolation-Based Regularization
2.2 Domain Adaptation
3 Method
3.1 Adversarial Domain Adaptation
3.2 Dual Mixup Regularization
3.3 Training Procedure
3.4 Discussion
4 Experiments
4.1 Setup
4.2 Results
4.3 Further Analysis
5 Conclusion
References
Robust and On-the-Fly Dataset Denoising for Image Classification
1 Introduction
2 Problem Setup
2.1 Entropy-Based Assumption over Noisy Labels
3 Denoising Datasets On-the-Fly with Counterfactual Thresholds
3.1 Separating Mislabeled Examples via SGD
3.2 Thresholds that Classify Mislabeled Examples
3.3 A Practical Algorithm for Robust Training
4 Experiments
4.1 CIFAR-10 and CIFAR-100
4.2 ImageNet
4.3 WebVision
4.4 Clothing1M
4.5 Ablation Studies
5 Related Work
6 Discussion
References
Imaging Behind Occluders Using Two-Bounce Light
1 Introduction
1.1 Contributions
2 Related Work
2.1 Non-Line-of-Sight Imaging
2.2 Shape from Shadows
3 Imaging Behind Occluders
3.1 The Elementary Measurement
3.2 Method
4 Reconstructing the Shape of Hidden Objects
4.1 Implementation
4.2 Stationary Hidden Objects
4.3 Moving Hidden Object
5 Overcoming Geometric and Photometric Challenges of Imaging Behind Occluders
5.1 Sources of Error in Imaging Behind Occluders
5.2 Robust Carving
5.3 Implementation
6 Applications
7 Conclusion
References
Improving Object Detection with Selective Self-supervised Self-training
1 Introduction
2 Augmenting COCO Detection with Web Images
3 Selective Self-supervised Self-training
3.1 Self-training for Object Detection (SOD)
3.2 Selective Self-training for Object Detection (S2OD)
3.3 Selective Self-supervised Self-training Object Detection (S4OD)
4 Related Work
5 Experiments
5.1 Augmenting COCO-backpack(chair) with Web-backpack(chair)
5.2 Semi-supervised Object Detection on PASCAL VOC
6 Conclusion
References
Deep Local Shapes: Learning Local SDF Priors for Detailed 3D Reconstruction
1 Introduction
2 Related Work
2.1 Traditional Shape Representations
2.2 Learned Shape Representations
2.3 Local Shape Priors
3 Review of DeepSDF
4 Deep Local Shapes
4.1 Shape Border Consistency
4.2 Deep Local Shapes Training and Inference
4.3 Point Sampling
5 Experiments
5.1 Object Reconstruction
5.2 Scene Reconstruction
6 Conclusion
References
Info3D: Representation Learning on 3D Objects Using Mutual Information Maximization and Contrastive Learning
1 Introduction
2 Background
3 Methodology
3.1 Local Chunks
3.2 Geometric Transformation
3.3 InfoNCE Objective
4 Experiments
4.1 Representation Learning on Aligned Shapes
4.2 Representation Learning on Rotated Shapes
4.3 Ablation Studies
5 Conclusion
References
Adversarial Data Augmentation via Deformation Statistics
1 Related Work
2 Method
2.1 Baseline Method: AdvAffine
2.2 Proposed Method: AdvEigAug
3 Experiments
3.1 Registration and Segmentation Networks
3.2 Experimental Design
4 Results and Discussion
5 Conclusion
References
Neural Predictor for Neural Architecture Search
1 Introduction
2 Related Work
3 Neural Predictor
3.1 Hyper-parameters in the Workflow
3.2 Modeling by Graph Convolutional Networks
4 Experiments
4.1 NASBench-101
4.2 ImageNet Experiments
References
Learning Permutation Invariant Representations Using Memory Networks
1 Introduction
2 Related Work
3 Background
4 Proposed Approach
4.1 Motivation
4.2 Model Components
4.3 Model Architecture
4.4 Analysis
5 Experiments
5.1 Toy Datasets
5.2 Real World Datasets
6 Conclusions
References
Feature Space Augmentation for Long-Tailed Data
1 Introduction
2 Related Work
2.1 Learning with Balanced Loss
2.2 Data Synthesis and Augmentation
2.3 Transfer Learning
3 The Problem of Long Tail
3.1 Two Reasons of Model Performance Drop
3.2 Class Activation Map and Feature Decomposition
4 Method
4.1 Initial Feature Learning
4.2 Feature Space Augmentation
4.3 Fine Tuning with Online Augmented Samples
5 Experiments
5.1 Long-Tailed CIFAR
5.2 ImageNet-LT and Places-LT Dataset
5.3 iNaturalist
5.4 Ablation Analysis
6 Conclusion
References
Laying the Foundations of Deep Long-Term Crowd Flow Prediction
1 Introduction
2 Related Work
3 Method
3.1 Preliminaries
3.2 CAGE Representation
3.3 Crowd-Flow Prediction
4 Experimental Preliminaries
4.1 Synthetic Datasets
4.2 Real-World Datasets
4.3 Evaluation Protocols
5 Experiments and Evaluation
5.1 Varied Density and No Compression
5.2 Varied Compression Rate with Varied Density
5.3 Results on Simulated Crowd Flow
5.4 Results on Real-World Data
5.5 Case Study
6 Conclusion
References
Weakly-Supervised Action Localization with Expectation-Maximization Multi-Instance Learning
1 Introduction
2 Related Work
3 Method
3.1 E Step
3.2 M Step
3.3 Overall Algorithm
3.4 Comparison with Previous Methods
3.5 Inference
4 Experiments
4.1 Experimental Setup
4.2 Comparison with State-of-the-Art Approaches
4.3 Ablation Studies
5 Conclusion
References
Fairness by Learning Orthogonal Disentangled Representations
1 Introduction
2 Related Work
3 Methodology
3.1 Fairness Definition
3.2 Problem Formulation
3.3 Fairness by Learning Orthogonal and Disentangled Representations
3.4 Overall Objective Function
4 Experiments
4.1 Experimental Setup
4.2 Comparison with State of the Art
4.3 Ablative Study
4.4 Qualitative Analysis
4.5 Sensitivity Analysis
5 Conclusions and Future Work
References
Self-supervision with Superpixels: Training Few-Shot Medical Image Segmentation Without Annotation
1 Introduction
2 Related Work
2.1 Few-Shot Semantic Segmentation
2.2 Self-supervised Learning in Semantic Segmentation
2.3 Superpixel Segmentation
3 Method
3.1 Problem Formulation
3.2 Network Architecture
3.3 Superpixel-Based Self-supervised Learning
4 Experiments
4.1 Quantitative and Qualitative Results
4.2 Ablation Studies
5 Conclusion
References
On Diverse Asynchronous Activity Anticipation
1 Introduction
2 Related Research
3 Technical Approach
3.1 Overview
3.2 Dual Token Embedding
3.3 Sequence Generation
3.4 Adversarial Learning and the Discriminator
4 Empirical Evaluation
4.1 Datasets
4.2 Implementation Details.
4.3 Anticipation Results Across Datasets
4.4 Analysis of Adversarial Learning
4.5 Analysis of Number of Samples
4.6 Analysis of Normalized Distance Regularization and Diversity
4.7 Visualization of Diversity and Quality
5 Summary
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