Note: Intro -- Preface -- Organization -- Awards 2016 -- Contents -- Biomedical Image Processing and Analysis -- A Quantitative Assessment of Image Normalization for Classifying Histopathological Tissue of the Kidney -- 1 Introduction -- 2 Medical Background -- 3 Experimental Study -- 3.1 Image Representations -- 3.2 Stain Normalization Methods -- 3.3 Evaluation Protocol -- 4 Results -- 5 Discussion -- 6 Conclusion -- References -- Classification and Detection -- Deep Learning for Vanishing Point Detection Using an Inverse Gnomonic Projection -- 1 Introduction -- 1.1 Related Work -- 2 Approach -- 2.1 Parametrisation -- 2.2 Network Architecture -- 2.3 Training Data -- 2.4 Vanishing Point Refinement -- 2.5 Horizon Line and Orthogonal Vanishing Point Estimation -- 3 Experiments -- 3.1 Horizon Estimation -- 3.2 Additional Applications -- 4 Conclusion -- References -- Learning Where to Drive by Watching Others -- 1 Introduction -- 2 Related Work -- 3 Unsupervised Learning of Drivable Surfaces -- 3.1 Self-supervision by Experiential Reflection -- 3.2 Learning Drivable Surfaces -- 4 Experimental Results -- 4.1 Datasets -- 4.2 Zero-Shot Learning -- 4.3 Transfer Learning -- 4.4 Self-supervision from the Wild: YouTube Dashcam Dataset -- 5 Conclusions -- References -- Learning Dilation Factors for Semantic Segmentation of Street Scenes -- 1 Introduction -- 2 Learning Dilation Factors for Convolutions -- 2.1 Forward Pass -- 2.2 Backward Pass -- 2.3 Network Architectures -- 2.4 Implementation Details -- 3 Experiments -- 3.1 Cityscapes -- 3.2 CamVid -- 4 Conclusion -- References -- Learning to Filter Object Detections -- 1 Introduction -- 2 Related Work -- 3 Problem Formulation -- 4 Filtering Network Architecture -- 5 Learning Objectives -- 5.1 Approximate Non-maximum Suppression Objective -- 5.2 Network Detection Objective -- 6 Experiments -- 7 Conclusion. , References -- Computational Photography -- Motion Deblurring in the Wild -- 1 Introduction -- 1.1 Related Work -- 2 Blurry Images in the Wild -- 3 The Multiscale Convolutional Neural Network -- 4 Experiments -- 5 Conclusions -- References -- Robust Multi-image HDR Reconstruction for the Modulo Camera -- 1 Introduction -- 2 Related Work -- 3 The Modulo Camera -- 4 Robust HDR Reconstruction -- 5 Robust Capture Protocol -- 6 Evaluation -- 7 Conclusion -- References -- Trainable Regularization for Multi-frame Superresolution -- 1 Introduction -- 2 Multi-frame Superresolution -- 3 SR Method Description -- 4 Data Acquisition -- 4.1 Acquisition Setup -- 4.2 Real Acquisitions -- 4.3 Simulated Acquisitions for Learning -- 5 Experiments and Results -- 6 Conclusion -- References -- Image and Video Processing -- A Comparative Study of Local Search Algorithms for Correlation Clustering -- 1 Introduction -- 1.1 Contribution -- 2 Related Work -- 3 Local Search Algorithms -- 3.1 Greedy Additive Edge Contraction (GAEC) -- 3.2 Greedy Fixation (GF) -- 3.3 Cut, Glue and Cut (CGC) -- 3.4 Kernighan-Lin Algorithm with Joins (KLj) -- 4 Empirical Comparison -- 4.1 Image Segmentation (seg-2d) -- 4.2 Volume Image Segmentation (seg-3d-300, Seg-3d-450) -- 4.3 Clustering Images of Hand-Written Digits (MNIST) -- 4.4 Clustering of Social Networks (Epinions, Slashdot) -- 5 Conclusion -- References -- Combined Precise Extraction and Topology of Points, Lines and Curves in Man-Made Environments -- 1 Introduction -- 1.1 Related Work -- 1.2 Contribution -- 2 Characteristics of Different Image Features -- 3 Detection Process -- 3.1 Extraction of Edges -- 3.2 Fitting the Geometrical Primitives -- 3.3 Calculation of Intersection Features -- 3.4 Building a Topology -- 4 Evaluation -- 4.1 Detection Accuracy of Intersection Features -- 4.2 Comparison to ELSDc -- 4.3 Detection of Contours. , 5 Conclusion -- References -- Recurrent Residual Learning for Action Recognition -- 1 Introduction -- 2 Related Work -- 3 Recurrent Residual Network -- 3.1 ResNet -- 3.2 Type of Temporal Skip Connection -- 3.3 Temporal Context -- 4 Experimental Setup -- 4.1 Baseline Experiments -- 4.2 Effect of Type and Position of the Recurrent Connection -- 4.3 Effect of Temporal Context -- 4.4 Comparison with the State of the Art -- 5 Conclusion -- References -- A Local Spatio-Temporal Approach to Plane Wave Ultrasound Particle Image Velocimetry -- 1 Introduction -- 2 Spatio-Temporal Filter Bank -- 3 Local Flow Estimation -- 4 Experimental Results -- 4.1 One-Dimensional Synthetic Ground Truth Videos -- 4.2 Laminar Pipe Flow: Ground Truth Data -- 4.3 Ultrasound Particle Image Velocimetry: In Vitro Data -- 5 Conclusion -- References -- Machine Learning and Pattern Recognition -- Object Boundary Detection and Classification with Image-Level Labels -- 1 Introduction -- 2 Obtaining Pixel-Level Scores from Image-Wise Predictions -- 2.1 Gradient -- 2.2 Deconvolution -- 2.3 Layer-Wise Relevance Propagation -- 3 Experiments -- 3.1 Performance on the SBD Task -- 3.2 Shortcomings of Visualization Methods -- 4 Conclusion -- References -- Semantic Segmentation of Outdoor Areas Using 3D Moment Invariants and Contextual Cues -- 1 Introduction -- 2 Related Work -- 3 Local Features Based on 3D Moment Invariants -- 3.1 3D Surface Moments -- 3.2 Local 3D Moment Invariants -- 4 Leveraging Context Information Using Random Forests -- 4.1 Random Forests for Semantic Segmentation -- 4.2 Contextual Cues from Local Neighborhoods -- 4.3 Cascaded Random Forests -- 5 Experiments -- 5.1 Segmentation of Individual Trees -- 5.2 Analysis of Forestal Areas -- 5.3 Urban Scenes -- 6 Conclusions -- References -- Neuron Pruning for Compressing Deep Networks Using Maxout Architectures -- 1 Introduction. , 2 Related Work -- 3 Fundamentals -- 4 Compressing Networks with Maxout Architectures -- 4.1 Neuron Pruning -- 4.2 Weight Pruning -- 5 Evaluation -- 5.1 Handwritten Digit Recognition -- 5.2 Face Verification -- 6 Conclusion -- References -- A Primal Dual Network for Low-Level Vision Problems -- 1 Introduction -- 2 Related Work -- 3 Method -- 3.1 Network Structure -- 3.2 Extension to Related Problems (from the TV-family) -- 4 Evaluation -- 4.1 From Energy Optimization to Algorithm Learning -- 5 Conclusion -- References -- End-to-End Learning of Video Super-Resolution with Motion Compensation -- 1 Introduction -- 2 Related Work -- 2.1 Image Super-Resolution -- 2.2 Video Super-Resolution -- 2.3 Motion Estimation -- 3 Video Super-Resolution with Patch-Based Training -- 4 Video Super-Resolution with Image-Based Training -- 5 Combined Warping and Upsampling Operation -- 6 Conclusions -- References -- Convolutional Neural Networks for Movement Prediction in Videos -- 1 Introduction -- 2 Related Work -- 3 Scenarios and Data Acquisition -- 3.1 Table Tennis Setup -- 3.2 Squash Setup -- 4 Trajectory Prediction Model -- 4.1 Network Architecture -- 5 Results -- 5.1 Table Tennis Scenario -- 5.2 Squash Scenario -- 5.3 Qualitative Results -- 6 Conclusion -- References -- Finding the Unknown: Novelty Detection with Extreme Value Signatures of Deep Neural Activations -- 1 Introduction -- 2 Related Work -- 3 Extreme Value Signatures -- 4 Experiments -- 4.1 Baseline Methods -- 4.2 Multi-class Novelty Detection on ImageNet Subsets -- 4.3 Computation Time Analysis -- 4.4 Large-Scale Multi-class Novelty Detection on ImageNet -- 4.5 Visualizing Class-Indicative Image Parts with EVS -- 5 Conclusion -- References -- Improving Facial Landmark Detection via a Super-Resolution Inception Network -- 1 Introduction -- 2 Related Work -- 3 Landmark Detection on Low Resolution Images. , 4 Proposed Super-Resolution Network -- 4.1 Network Architecture -- 4.2 Training -- 4.3 Results -- 5 Super Resolution for Facial Landmark Detection -- 5.1 Results -- 6 Conclusion -- References -- Mathematical Foundations, Statistical Data Analysis and Models -- Diverse M-Best Solutions by Dynamic Programming -- 1 Introduction -- 2 Related Work -- 3 Optimal Second Best Tree Solutions -- 4 Optimal M-Best Tree Solutions -- 5 Approximate Diverse M-Best Solutions -- 6 Applications and Experiments -- 7 Conclusion -- References -- Adaptive Regularization in Convex Composite Optimization for Variational Imaging Problems -- 1 Introduction -- 1.1 Related Work -- 2 Variational Model with Adaptive Parameter Balancing -- 2.1 Optimization with ADMM Algorithm -- 3 Applications -- 3.1 Image Segmentation -- 3.2 Motion Estimation -- 4 Numerical Results -- 4.1 Image Segmentation -- 4.2 Motion Estimation -- 5 Conclusion and Discussion -- References -- Variational Networks: Connecting Variational Methods and Deep Learning -- 1 Introduction -- 2 Variational Networks -- 2.1 Relation to Incremental Gradient Methods -- 2.2 Relation to Residual Networks -- 3 Variational Networks for Image Reconstruction -- 3.1 Problem Formulation and Parametrization -- 3.2 Training -- 4 Experiments -- 4.1 Energy Minimization with VNs -- 4.2 Approximate Incremental Minimization with VNs -- 4.3 VNs in a Reaction Diffusion Setup -- 5 Conclusion -- References -- Gradient Flows on a Riemannian Submanifold for Discrete Tomography -- 1 Introduction -- 2 Approach -- 3 Optimization -- 4 Numerical Experiments -- 5 Conclusion and Future Work -- References -- Model Selection for Gaussian Process Regression -- 1 Introduction -- 1.1 Existing Model Selection Criteria for Gaussian Processes -- 1.2 Our Contributions -- 2 Posterior Agreement Applied to Model Selection for Gaussian Process Regression. , 2.1 General Model Selection Framework Using Posterior Agreement.