Computer Vision - ECCV 2020

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

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

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

Springer (Publisher)

Published on 3. November 2020

XLI, 797 pages

E-Book

PDF with digital watermarking

System requirements

978-3-030-58536-5 (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 II
Diffraction Line Imaging
1 Introduction
2 Background on Diffraction Gratings
3 Diffraction-Based 2D Positioning
3.1 Image Formation Model
3.2 Learning to Recover Horizontal Coordinates
4 Expanding the FOV Using Multiple ROIs
5 Reducing Sparse Point Uncertainty
5.1 Horizontal Cylindrical Lens
5.2 Double-Axis Diffraction
6 Hardware and Calibration
7 Experimental Evaluation Details
8 Concluding Remarks
References
Transforming and Projecting Images into Class-Conditional Generative Networks
1 Introduction
2 Related Work
3 Image Projection Methods
3.1 Basic Loss Function
3.2 Object Localization
3.3 Transformation Model and Loss
3.4 Optimization Algorithms
3.5 Fine-Tuning
4 Results
5 Discussion
References
Suppress and Balance: A Simple Gated Network for Salient Object Detection
1 Introduction
2 Related Work
2.1 Salient Object Detection
2.2 Multiscale Feature Extraction
2.3 Gated Mechanisms
3 Proposed Method
3.1 Network Overview
3.2 Gated Dual Branch
3.3 Folded Atrous Spatial Pyramid Pooling
3.4 Supervision
4 Experiments
4.1 Experimental Setup
4.2 Performance Comparison with State-of-the-Art
4.3 Ablation Studies
5 Conclusions
References
Visual Memorability for Robotic Interestingness via Unsupervised Online Learning
1 Introduction
2 Related Work
3 Visual Memory
3.1 Memory Writing
3.2 Memory Reading
4 Learning
4.1 Long-Term Learning
4.2 Short-Term Learning
4.3 Online Learning
5 Experiments
6 Ablation Study
6.1 Writing Protocol
6.2 Memory Capacity
6.3 Translational Invariance
6.4 Losing Interest
7 Conclusion
References
Post-training Piecewise Linear Quantization for Deep Neural Networks
1 Introduction
2 Related Work
3 Quantization Schemes
3.1 Uniform Quantization
3.2 Piecewise Linear Quantization (PWLQ)
3.3 Error Analysis
4 Hardware Impact
5 Experiments
5.1 Ablation Study on ImageNet
5.2 Comparison to Existing Approaches
5.3 Other Applications
6 Conclusion
References
Joint Disentangling and Adaptation for Cross-Domain Person Re-Identification
1 Introduction
2 Related Work
3 Method
3.1 Disentangling Module
3.2 Adaptation Module
3.3 Discussion
3.4 Optimization
4 Experiments
4.1 Implementation Details
4.2 Quantitative Results
4.3 Qualitative Results
5 Conclusion
References
In-Home Daily-Life Captioning Using Radio Signals
1 Introduction
2 Related Work
3 RF Signal Preliminary
4 RF-Diary
4.1 RF Signal Encoding
4.2 Floormap Encoding
4.3 Caption Generation
5 Multi-modal Feature Alignment Training
5.1 Video Encoding
5.2 Alignment of Paired Data
5.3 Alignment of Unpaired Data
6 Experiments
6.1 Quantitative Results
6.2 Ablation Study
6.3 Qualitative Result
6.4 Additional Notes on Privacy
7 Conclusion
References
Self-challenging Improves Cross-Domain Generalization
1 Introduction
2 Related Work
3 Method
3.1 Self-challenging Algorithm
3.2 Theoretical Evidence
3.3 Engineering Specification and Extensions
4 Experiments
4.1 Datasets
4.2 Ablation Study
4.3 Cross-Domain Evaluation
5 Discussion
6 Conclusion
References
A Competence-Aware Curriculum for Visual Concepts Learning via Question Answering
1 Introduction
2 Related Work
2.1 Neural-Symbolic Visual Question Answering
2.2 Curriculum Learning and Machine Teaching
3 Methodology
3.1 Neural-Symbolic Concept Learner
3.2 Background of Item Response Theory (IRT)
3.3 Multi-dimensional IRT Using Model Responses
3.4 Variational Bayesian Inference for mIRT
3.5 Training Samples Selection Strategy
4 Experiments
4.1 Experimental Setup
4.2 Training Process and Model Performance
4.3 Multi-dimensional IRT
4.4 Concept Learner
4.5 Question Selection Strategy
5 Conclusions and Discussions
References
Multitask Learning Strengthens Adversarial Robustness
1 Introduction
2 Related Work
3 Adversarial Setting
3.1 Multitask Learning Objective
3.2 Adversarial Multitask Attack Objective
3.3 Adversarial Single-Task Attack Objective
4 Theoretical Analysis
5 Experiments
5.1 Datasets
5.2 Attack Methods
5.3 Multitask Models Against Multitask Attack
5.4 Multitask Models Against Single-Task Attacks
5.5 Multitask Learning Complements Adversarial Training
6 Conclusion
References
S2DNAS: Transforming Static CNN Model for Dynamic Inference via Neural Architecture Search
1 Introduction
2 Related Work
3 Our Approach
3.1 Overview of S2DNAS
3.2 The Details of S2D
3.3 The Details of NAS
4 Experiments
4.1 Experiment Settings
4.2 Classification Results
4.3 Discussion
5 Conclusion
References
Improving Deep Video Compression by Resolution-Adaptive Flow Coding
1 Introduction
2 Related Work
2.1 Image Compression
2.2 Video Compression
3 Methodology
3.1 System Overview
3.2 Problem Formulation
3.3 Resolution-Adaptive Flow Coding (RaFC)
4 Experiment
4.1 Experimental Setup
4.2 Experimental Results
4.3 Ablation Study and Model Analysis
5 Conclusion
References
Motion Capture from Internet Videos
1 Introduction
2 Related Work
3 Methods
3.1 Pose-Based Video Synchronization
3.2 Motion Recovery
3.3 Iterative Optimization
4 Experiments
4.1 Motion Capture from Internet Videos
4.2 Quantitative Evaluation
5 Summary
References
Appearance-Preserving 3D Convolution for Video-Based Person Re-identification
1 Introduction
2 Related Work
3 Appearance-Preserving 3D Convolution
3.1 The Framework
3.2 Appearance-Preserving Module
3.3 Discussion
3.4 Combining AP3D with I3D and P3D Blocks
4 AP3D for Video-Based ReID
4.1 Network Architectures
4.2 Objective Function
5 Experiments
5.1 Datasets and Evaluation Protocol
5.2 Implementation Details
5.3 Comparison with Related Approaches
5.4 Ablation Study
5.5 Visualization
5.6 Comparison with State-of-the-Art Methods
6 Conclusion
References
Solving the Blind Perspective-n-Point Problem End-to-End with Robust Differentiable Geometric Optimization
1 Introduction
2 Related Work
3 An End-to-End Blind PnP Solver
3.1 Problem Formulation
3.2 Bi-level Optimization
3.3 Declarative Layers for Blind PnP
3.4 Network Architecture
3.5 Learning from Pose-Labelled Data
4 Results
4.1 Synthetic Data Experiments
4.2 Real Data Experiments
5 Conclusion
References
Exploiting Deep Generative Prior for Versatile Image Restoration and Manipulation
1 Introduction
2 Related Work
3 Method
3.1 Deep Generative Prior
3.2 Discriminator Guided Progressive Reconstruction
4 Applications
4.1 Image Restoration
4.2 Image Manipulation
5 Conclusion
References
Deep Spatial-Angular Regularization for Compressive Light Field Reconstruction over Coded Apertures
1 Introduction
2 Related Work
2.1 LF Reconstruction from SAIs
2.2 LF Reconstruction from Coded Measurements
3 Proposed Method
3.1 Learning Coded Apertures
3.2 Reconstruction with Deep Spatial-Angular Regularization
3.3 Training and Implementation Details
4 Experiments
4.1 Comparison with State-of-the-Art Methods
4.2 Ablation Study
5 Conclusion and Future Work
References
Video-Based Remote Physiological Measurement via Cross-Verified Feature Disentangling
1 Introduction
2 Related Work
2.1 Video-Based Remote Physiological Measurement
2.2 Disentangle Representation Learning
3 Proposed Method
3.1 Multi-scale Spatial Temporal Map
3.2 Cross-Verified Feature Disentangling
3.3 Multi-task Physiological Measurement
4 Experiments
4.1 Databases and Experimental Settings
4.2 Intra-database Testing
4.3 Cross-Database Testing
4.4 Ablation Study
5 Conclusions
References
Combining Implicit Function Learning and Parametric Models for 3D Human Reconstruction
1 Introduction
2 Related Work
2.1 Parametric Modelling for Humans
2.2 Implicit Functions for Humans
2.3 Summary: Implicit vs Parametric Modelling
3 Method
3.1 Training Data Preparation
3.2 IP-Net: Overview
3.3 Registering SMPL to IP-Net Predictions
4 Dataset and Experiments
4.1 Dataset
4.2 Outer Surface Reconstruction
4.3 Comparison to Baselines
4.4 Body Shape Under Clothing
4.5 Why is Correspondence Prediction Important?
4.6 Why Not Independent Networks for Inner and Outer Surfaces?
4.7 Using IP-Net Correspondences to Register Scans
4.8 Registration from Point Clouds Obtained from a Single View
4.9 Hand Registration
5 Conclusions
References
Orientation-Aware Vehicle Re-Identification with Semantics-Guided Part Attention Network
1 Introduction
2 Related Work
3 Proposed Method
3.1 Semantics-Guided Part Attention Network
3.2 Part Feature Extraction
3.3 Co-occurrence Part-Attentive Distance Metric
3.4 Model Learning Scheme
4 Experiments
4.1 Datasets and Evaluation Metrics
4.2 Implementation Details
4.3 Ablation Studies and Visualization
4.4 Comparison with the State-of-the-Arts
5 Conclusion
References
Mining Cross-Image Semantics for Weakly Supervised Semantic Segmentation
1 Introduction
2 Related Work
3 Methodology
3.1 Co-attention Classification Network
3.2 Co-attention Classifier Guided WSSS Learning
3.3 Detailed Network Architecture
4 Experiment
4.1 Experiment 1: Learn WSSS only from PASCAL VOC ch21everingham2015pascal Data
4.2 .26em plus .1em minus .1emExperiment 2: Learn WSSS With Extra Simple Single-Label Data
4.3 Experiment 3: Learn WSSS with Extra Web-Sourced Data
4.4 Ablation Studies
5 Conclusion
References
CoReNet: Coherent 3D Scene Reconstruction from a Single RGB Image
1 Introduction
2 Related Work
3 Proposed Approach
3.1 3D Volume Representation
3.2 Core Model Architecture
3.3 Ray Traced Skip Connections
3.4 IoU Training Loss
3.5 Mesh Reconstruction
4 Experiments
4.1 Single Object Reconstruction on ShapeNet
4.2 Single Object Reconstruction on Pix3D
4.3 Multiple Object Reconstruction
5 Conclusions
References
Layer-Wise Conditioning Analysis in Exploring the Learning Dynamics of DNNs
1 Introduction
2 Preliminaries
3 Layer-Wise Conditioning Analysis for DNNs
3.1 Efficient Computation
3.2 Connection to Proximal Back-Propagation
4 Exploring Batch Normalized Networks
4.1 Stabilizing Training
4.2 Improved Conditioning
5 Training Very Deep Residual Networks
5.1 Proposed Solution
5.2 Revisiting the Pre-activation Residual Network
6 Conclusion and Future Work
References
RAFT: Recurrent All-Pairs Field Transforms for Optical Flow
1 Introduction
2 Related Work
3 Approach
3.1 Feature Extraction
3.2 Computing Visual Similarity
3.3 Iterative Updates
3.4 Supervision
4 Experiments
4.1 Sintel
4.2 KITTI
4.3 Ablations
4.4 Timing and Parameter Counts
4.5 Video of Very High Resolution
5 Conclusions
References
Domain-Invariant Stereo Matching Networks
1 Introduction
2 Related Work
2.1 Deep Neural Networks for Stereo Matching
2.2 Adaptation and Self-supervised Learning
2.3 Cross-Domain Generalization
3 Proposed DSMNet
3.1 Domain Normalization
3.2 Structure-Preserving Graph-Based Filtering
3.3 Network Architecture
4 Experimental Results
4.1 Datasets
4.2 Ablation Study
4.3 Component Analysis and Comparisons
4.4 Cross-Domain Evaluations
4.5 Fine-Tuning
5 Extension for Optical Flow
6 Conclusion
References
DeepHandMesh: A Weakly-Supervised Deep Encoder-Decoder Framework for High-Fidelity Hand Mesh Modeling
1 Introduction
2 Related Works
3 Hand Model
4 Encoder
4.1 Hand Pose Vector
4.2 Network Architecture
5 Decoder
5.1 Hand Model Refinement
5.2 Hand Model Deformation
6 Training DeepHandMesh
7 Implementation Details
8 Experiment
8.1 Dataset
8.2 Ablation Study
8.3 Comparison with State-of-the-Art Methods
8.4 3D Hand Mesh Estimation from General Images
9 Discussion
10 Conclusion
References
Content Adaptive and Error Propagation Aware Deep Video Compression
1 Introduction
2 Related Work
2.1 Image Compression
2.2 Video Compression
3 Motivations Related to Learning Based Video Compression System
3.1 Error Propagation
3.2 The Content Adaptive Coding Scheme
4 Proposed Method
4.1 Introduction of the DVC Framework
4.2 The Error Propagation Aware Training Strategy
4.3 The Online Encoder Updating Scheme
5 Experiments
5.1 Experimental Setup
5.2 Comparison with the State-of-the-art Methods
5.3 Ablation Study
5.4 Discussion
6 Conclusion
References
Towards Streaming Perception
1 Introduction
2 Related Work
3 Proposed Evaluation
3.1 Formal Definition
3.2 Emergent Tracking and Forecasting
3.3 Computational Constraints
3.4 Challenges for Practical Implementation
4 Solutions and Analysis
4.1 Setup
4.2 Detection-Only
4.3 Forecasting
4.4 Visual Tracking
5 Discussion
6 Conclusion and Future Work
References
Towards Automated Testing and Robustification by Semantic Adversarial Data Generation
1 Introduction
2 Related Work
3 Synthesizing Semantic Adversarial Objects
3.1 Synthesizer Design
3.2 Synthesizing Semantic Adversaries
4 Experiments and Results
4.1 Setup and Datasets
4.2 Semantic Adversary for Automated Testing
4.3 Semantic Adversary for Data Augmentation
5 Conclusions
References
Adversarial Generative Grammars for Human Activity Prediction
1 Introduction
2 Related Work
3 Approach
3.1 Preliminaries
3.2 Learning the Starting Non-terminal
3.3 Grammar Learning
3.4 Adversarial Grammar Learning
4 Experiments
4.1 Datasets
4.2 Human Pose Forecasting
4.3 Activity Forecasting in Videos
5 Conclusion
References
GDumb: A Simple Approach that Questions Our Progress in Continual Learning
1 Introduction
2 Problem Formulation, Assumptions, and Trends
2.1 Simplifying Assumptions in Continual Learning
2.2 Recent Trends in Continual Learning
3 Greedy Sampler and Dumb Learner (GDumb)
4 Experiments
4.1 Results and Discussions
4.2 Resources Needed
4.3 Potential Future Extensions
5 Conclusion
References
Learning Lane Graph Representations for Motion Forecasting
1 Introduction
2 Related Work
3 Lane Graph Representations for Motion Forecasting
3.1 ActorNet: Extracting Traffic Participant Representations
3.2 MapNet: Extracting Structured Map Representation
3.3 FusionNet
3.4 Prediction Header
3.5 Learning
4 Experimental Evaluation
4.1 Experimental Settings
4.2 Results
5 Conclusion
References
What Matters in Unsupervised Optical Flow
1 Introduction
2 Related Work
3 Preliminaries on Unsupervised Optical Flow
4 Key Components of Unsupervised Optical Flow
5 Experiments
6 Results
7 Conclusion
References
Synthesis and Completion of Facades from Satellite Imagery
1 Introduction
2 Related Work
3 Facade Synthesis
3.1 Selection
3.2 Classification and Parameter Estimation
3.3 Completion
4 Results
5 Conclusions and Future Work
References
Mapillary Planet-Scale Depth Dataset
1 Introduction
2 Dataset
2.1 Global Model-Wise Camera Calibration
2.2 Image Search
2.3 Reconstruction and Multi-view Stereo
3 Training with Multiple Cameras
4 Experiments
5 Conclusion
References
V2VNet: Vehicle-to-Vehicle Communication for Joint Perception and Prediction
1 Introduction
2 Related Work
3 Perceiving the World by Leveraging Multiple Vehicles
3.1 Which Information Should Be Transmitted
3.2 Leveraging Multiple Vehicles
3.3 Learning
4 V2V-Sim: A Dataset for V2V Communication
5 Experimental Evaluation
6 Conclusion
References
Training Interpretable Convolutional Neural Networks by Differentiating Class-Specific Filters
1 Introduction
2 Related Works
3 Method
3.1 Ideally Class-Specific Filters
3.2 Problem Formulation
3.3 Optimization
4 Experiment
4.1 Effectiveness of CSG Training
4.2 Study on Class-Specificity
4.3 Correlation Between Filters
5 Application
5.1 Localization
5.2 Adversarial Sample Detection
6 Conclusion
References
EagleEye: Fast Sub-net Evaluation for Efficient Neural Network Pruning
1 Introduction
2 Related Work
3 Methodology
3.1 Motivation
3.2 Adaptive Batch Normalization
3.3 Correlation Measurement
3.4 EagleEye Pruning Algorithm
4 Experiments
4.1 Quantitative Analysis of Correlation
4.2 Generality of the Adaptive-BN-Based Evaluation Method
4.3 Efficiency of Our Proposed Method
4.4 Effectiveness of Our Proposed Method
5 Discussion and Conclusions
References
Intrinsic Point Cloud Interpolation via Dual Latent Space Navigation
1 Introduction
2 Related Work
3 Motivation and Background
3.1 Metric Interpolation in a Learned Space
4 Method
4.1 Overview
4.2 Architecture
4.3 Navigating the Restricted Latent Space
4.4 Interpretation
4.5 Unsupervised Training
5 Results
5.1 Shape Interpolation
5.2 Shape Reconstruction
6 Conclusion, Limitations and Future Work
References
Cross-Domain Cascaded Deep Translation
1 Introduction
2 Related Work
3 Method
4 Experiments
4.1 Ablation Study
4.2 Comparison to Other Methods
4.3 Unconditional Generation via Deep Feature Synthesis
5 Conclusions
References
``Look Ma, No Landmarks!'' - Unsupervised, Model-Based Dense Face Alignment
1 Introduction
1.1 Related Work
2 3DMM Parameters from Image-Model Correspondence
2.1 Interpolating a 3DMM to UV and Pixel Space
2.2 Least Squares Shape-from-correspondence
2.3 Least Squares Inverse Rendering
3 Self-supervised Learning of Dense Correspondence
3.1 Per-Pixel Confidence
3.2 Uncalibrated Shape-from-correspondence
3.3 In-Network Least Squares Inverse Rendering
3.4 Losses
4 Training
5 Experiments
6 Conclusion
References
Online Invariance Selection for Local Feature Descriptors
1 Introduction
2 Related Work
3 Learning the Best Invariance for Local Descriptors
3.1 Disentangling Invariance for Local Descriptors
3.2 Online Selection of the Best Invariance
3.3 Training Details
4 Experimental Results
4.1 Metrics
4.2 Method Validation
4.3 Descriptor Evaluation on HPatches
4.4 Evaluation in Challenging and Cross-Modal Situations
4.5 Application to Localization in Challenging Conditions
5 Conclusion
References
Rethinking Image Inpainting via a Mutual Encoder-Decoder with Feature Equalizations*-4mm
1 Introduction
2 Related Works
3 Proposed Algorithm
3.1 Mutual Encoder-Decoder
3.2 Feature Equalizations
3.3 Loss Functions
3.4 Visualizations
4 Experiments
4.1 Visual Evaluations
4.2 Numerical Evaluations
5 Ablation Study
6 Concluding Remarks
References
TextCaps: A Dataset for Image Captioning with Reading Comprehension
1 Introduction
2 Related Work
3 5044341En44FigaPrint.eps TextCaps Dataset
3.1 Dataset Collection
3.2 Dataset Analysis
4 Benchmark Evaluation
4.1 Baselines
4.2 Experimental Setup
4.3 Results
5 Conclusion
References
It Is Not the Journey But the Destination: Endpoint Conditioned Trajectory Prediction
1 Introduction
2 Related Work
2.1 Context-Based Prediction
2.2 Multimodal Trajectory Prediction
2.3 Conditioned-on-Goal
3 Proposed Method
3.1 Endpoint VAE
3.2 Endpoint Conditioned Trajectory Prediction
3.3 Loss Functions
4 Experiments
4.1 Datasets
4.2 Implementation Details
4.3 Quantitative Results
4.4 Qualitative Results
5 Conclusion
References
Learning What to Learn for Video Object Segmentation
1 Introduction
2 Related Work
3 Method
3.1 Video Object Segmentation as Few-Shot Learning
3.2 Learning What to Learn
3.3 Few-Shot Learner
3.4 Video Object Segmentation Architecture
3.5 Inference
3.6 Training
3.7 Bounding Box Initialization
4 Experiments
4.1 Ablative Analysis
4.2 State-of-the-Art Comparison
5 Conclusions
References
Correction to: Rethinking Image Inpaintingvia a Mutual Encoder-Decoder with FeatureEqualizations
Correction to:Chapter "Rethinking Image Inpainting via a MutualEncoder-Decoder with Feature Equalizations"in: A. Vedaldi et al. (Eds.): Computer Vision - ECCV 2020,LNCS 12347, https://doi.org/10.1007/978-3-030-58536-5_43
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