(Downloads - 0)
For more info about our services contact : help@bestpfe.com
Table of contents
Introduction
1 Big (Visual) Data
2 Motivation
3 Contributions
1 Background
1.1 Image Representations for Classication
1.1.1 Visual Bag-of-Words
1.1.2 Deep representations
1.2 Metric Learning for Computer Vision
1.3 Supervised Distance Metric Learning
1.3.1 Notations
1.3.2 Distance and similarity metrics
1.3.3 Learning scheme
1.3.4 Review of popular metric learning approaches
1.4 Training Information in Metric Learning
1.4.1 Binary similarity labels
1.4.2 Richer provided information
1.4.3 Quadruplet-wise approaches
1.5 Regularization in Metric Learning
1.5.1 Representative regularization terms
1.5.2 Other regularization methods in Computer Vision
1.6 Summary
2 Quadruplet-wise Distance Metric Learning
2.1 Motivation
2.2 Quadruplet-wise Similarity Learning Framework
2.2.1 Quadruplet-wise Constraints
2.2.2 Full matrix Mahalanobis distance metric learning
2.2.3 Simplication of the model by optimizing over vectors
2.3 Quadruplet-wise (Qwise) Optimization
2.3.1 Full matrix metric optimization
2.3.2 Vector metric optimization
2.3.3 Implementation details
2.4 Experimental Validation on Relative Attributes
2.4.1 Integrating quadruplet-wise constraints
2.4.2 Classication experiments
2.5 Experimental Validation on Hierarchical Information
2.5.1 Formulation of our metric and constraints
2.5.2 Experiments
2.6 Conclusion
3 Fantope Regularization
3.1 Introduction
3.2 Regularization Scheme
3.2.1 Regularization term linearization
3.2.2 Optimization scheme
3.3 Theoretical Analysis
3.3.1 Concavity analysis
3.3.2 (Super-)Gradient of the regularizer
3.4 Experimental Validation
3.4.1 Synthetic example
3.4.2 Real-world experiments
3.5 Discussion
3.6 Conclusion
4 Discovering Important Semantic Regions in Webpages
4.1 Introduction
4.2 Constraint Formalization
4.2.1 Automatic generation of constraints
4.2.2 Similarity information provided by human users
4.2.3 Distance metric formulation
4.3 Visual and Structural Comparisons of Webpages
4.3.1 Regular grid segmentation
4.3.2 Structural segmentation
4.3.3 Integration of structural distance metrics
4.4 Experimental Results
4.4.1 Dataset
4.4.2 Setup parameter
4.4.3 Evaluation protocol
4.4.4 Learning results without human supervision
4.4.5 Supervised learning results
4.4.6 Structural segmentation maps
4.4.7 Summary
4.5 Conclusion
A Positive Semidenite Cone
A.1 Denitions
A.2 Rank of a Matrix
A.3 Projection onto the PSD Cone
B Solver for the Vector Optimization Problem
B.1 Primal Form of the Optimization Problem
B.2 Loss Functions
B.3 Gradient and Hessian Matrices
Bibliography
