Hierarchical Image-Region Labeling via Structured Learning

Size: px
Start display at page:

Download "Hierarchical Image-Region Labeling via Structured Learning"

Transcription

1 Hierarchical Image-Region Labeling via Structured Learning Julian McAuley, Teo de Campos, Gabriela Csurka, Florent Perronin XRCE September 14, 2009 McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

2 Structured Learning Most of my work has been concerned with applying machine learning to computer vision problems. In our BMVC work, we wanted to explore how recent advances in machine learning could be applied to the problem of image segmentation and classification. Structured Learning is an extension of SVMs for cases where the output is structured (as opposed to a binary output) McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

3 Structured Output Spaces Parse trees [Tsochantaridis et al 04] McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

4 Structured Output Spaces Graph matching McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

5 Structured Output Spaces Word alignment McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

6 Structured Output Spaces Ranking McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

7 Structured Output Spaces Matching McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

8 Structured Output Spaces Segmentation McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

9 Structured Learning Our risk functional is very similar to that of SVMs: (regularized) risk N (g w (X n ), y n ) 1 N n=1 } {{ } empirical risk + λω(w) }{{} regularization term N y n g w (X n ) (y, y n ) Ω(w) λ size of training set n th training instance w-parametrised classification function loss function regularisation function regularisation hyper-parameter McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

10 Structured Learning energy function SVM energy function y = g w (X n ) = arg min Φ(X n, y), w y Y y = g w (X n ) = arg min yφ(x n ), w y { 1,1} Obviously, we require that this minimisation problem is tractable. McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

11 Structured Learning For a binary classifier, minimising the risk is easy, as the problem is convex (in the separable case) For a structured classifier, minimising the risk is hard, as the risk is piecewise constant in w McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

12 Structured Learning Since we cannot minimize the risk, we minimize a convex upper bound on it. soft margin formulation subject to min w,ξ 1 N N ξ n + λω(w) n=1 Φ(X n, y) Φ(X n, y n ), w (y, y n ) ξ n for all n and y Y the problem is now convex, but the number of constraints is large. McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

13 Structured Learning Any subset of these constraints yields a convex upper bound on the correct solution We want to choose some good subset of constraints We choose those constraints which maximise ξ n McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

14 Structured Learning These can be found by the following equation: column generation ŷ n = arg min[ Φ(X n, y), w (y, y n )] y Obviously, this problem must also be tractable. McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

15 Summary Our output space is structured Our problem can be linearly parametrised The energy minimisation problem can be efficiently solved We can solve the column generation problem (usually due to a decomposable loss) McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

16 Part II: BMVC paper Hierarchical Image-Region Labeling via Structured Learning McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

17 Is this a bus or a car? McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

18 Is this a bus or a car? McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

19 A pyramidal representation allows us to combine features from multiple scales: (image stolen from [Leclerc et al]) McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

20 Our approach We would like to learn how to combine information from different classifiers, which operate at different scales McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

21 Our approach The output of our method should look something like this: We d like to use the results from small scales to help large scales, and vice-versa. McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

22 Sometimes a classifier will be wrong, but in a predictable way McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

23 bus + car = minibus Sometimes a combination of outputs will yield new insights McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

24 A bad example McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

25 A bad example McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

26 A bad example McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

27 A bad example McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

28 How to enforce consistency? We d like to encode the fact that the previous labeling is impossible Can enforce neighbourhood constraints Efficient algorithms exist for performing binary segmentation McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

29 Grid structured graphical models? Few efficient or exact techniques for multi-label segmentation Not clear how to combine data from different scales Poorly suited to structured learning if exact inference is impossible McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

30 If we form connections between overlapping regions: then our model is tree-structured; we can perform inference efficiently and exactly using the belief-propagation. McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

31 Energy minimisation problem We will have two terms; a node compatibility term: E(region, label) = Φ nodes (region, label), θ nodes and an edge compatibility term: E(r p, r c, label p, label c ) = Φ edges (r p, r c, label p, label c ), θ edges. We choose the assignment that minimizes the total energy: Y = arg min Y E(r, y(r)) + E(r p, r c, y(r p ), y(r c )). r nodes This is linear in Θ = (θ nodes ; θ edges ). (r p,r c) edges McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

32 First-order region features Rather than computing features for each image region (e.g. SIFT), we run existing classifiers on each image region, and use their probabilistic outputs. We use classifiers from [Csurka and Perronin, BMVC 2008]: a classifier which is trained to classify entire images a classifier which is trained to classify bounding-boxes a classifier which is trained on image patches in principle, any combination of classifiers could be used McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

33 We use all classifiers at all scales: Φ nodes (r, label) = (0,..., Pr,label 1 }{{,..., 0)... (0,..., Pr,label n }}{{,..., 0) } feature for first classifier feature for n th classifier Thus the model learns which classifiers are useful at which scales. McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

34 Second-order region features We want to ban the inconsistent assignments I mentioned earlier: multiple aeroplane bicycle train tvmonitor background The label of a child region must be at or below the label of its parent region on this DAG McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

35 Consistency: if label p label c, H(label p, label c ) = 0 if label p label c, 1 otherwise (label p = label c ) Regions with the same label should have similar features: Φ edges (r p, r c ; label p, label c ) = H(label p, label c ) P rp,label p P rc,label c 2 Thus we learn the extent to which classifiers are consistent at different scales McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

36 Our training data We use bounding boxes from the VOC2007/2008 datasets. McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

37 Our loss function We just use the Hamming loss: (y, y n ) = 1 r regions I {y n (r)}(y (r)). This decomposes over the nodes in our model: E(r, y(r)) I {y n (r)}(y(r)). r regions This is a fairly reasonable type of loss, though it is worth mentioning that some losses do not decompose. McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

38 We re done! We ve specified a linearly parametrised energy function We ve specified a (decomposable) loss function We ve specified an algorithm to find constraints We can now iteratively find the best value of Θ using an off-the-shelf solver such as SVM struct [Joachims] or BMRM [Teo et al]. McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

39 Our learned feature vector Node weights Image-level classifier Mid-level classifier Patch-level classifier background aeroplane bicycle bird boat bottle bus car cat chair cow diningtable dog horse motorbike person pottedplant sheep sofa train tvmonitor multiple background aeroplane bicycle bird boat bottle bus car cat chair cow diningtable dog horse motorbike person pottedplant sheep sofa train tvmonitor multiple background aeroplane bicycle bird boat bottle bus car cat chair cow diningtable dog horse motorbike person pottedplant sheep sofa train tvmonitor multiple Level 2/3 Level 1/2 Level 0/1 background aeroplane bicycle bird boat bottle bus car cat chair cow diningtable dog horse motorbike person pottedplant sheep sofa train tvmonitor multiple background aeroplane bicycle bird boat bottle bus car cat chair cow diningtable dog horse motorbike person pottedplant sheep sofa train tvmonitor multiple background aeroplane bicycle bird boat bottle bus car cat chair cow diningtable dog horse motorbike person pottedplant sheep sofa train tvmonitor multiple Level 3 Level 2 Level 1 Level 0 Edge weights Image-level classifier Mid-level classifier Patch-level classifier McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

40 Some results Correct labeling (loss = 0) McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

41 Some results Csurka and Perronin, BMVC 2008 (loss = 0.434) McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

42 Some results Our method (loss = 0.230) McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

43 Some results Training Validation Testing Csu. & Per (0.005) (0.004) (0.003) Non-learning (0.006) (0.005) (0.004) Learning (0.006) (0.006) (0.004) McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

44 Failure cases... Objects that co-occur with person are often classified as person McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

45 Stuff we could have done... Explore other possible tree-structures McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

46 Stuff we could have done... Explore other possible tree-structures McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

47 Stuff we could have done... multiple aeroplane bicycle train tvmonitor person hand face foot background Used a more complex hierarchy McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

48 Stuff we could have done... Used segmented (rather than bounding-box) data McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

49 Stuff we could have done... Our approach is something between classification and segmentation By using more computational resources, we could do better segmentation By obtaining better compound classes, we could do better classification McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

50 Conclusions We have a generic model for combining different classifiers at different scales Learning yields a significant improvement, but limits us in our choice of loss function Loads of possible directions for future work McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

51 The end McAuley et al (XRCE) Hierarchical Image-Region Labeling September 14, / 51

Beyond Sliding Windows: Object Localization by Efficient Subwindow Search

Beyond Sliding Windows: Object Localization by Efficient Subwindow Search Beyond Sliding Windows: Object Localization by Efficient Subwindow Search Christoph H. Lampert, Matthew B. Blaschko, & Thomas Hofmann Max Planck Institute for Biological Cybernetics Tübingen, Germany Google,

More information

LEARNING OBJECT SEGMENTATION USING A MULTI NETWORK SEGMENT CLASSIFICATION APPROACH

LEARNING OBJECT SEGMENTATION USING A MULTI NETWORK SEGMENT CLASSIFICATION APPROACH Università degli Studi dell'insubria Varese, Italy LEARNING OBJECT SEGMENTATION USING A MULTI NETWORK SEGMENT CLASSIFICATION APPROACH Simone Albertini Ignazio Gallo, Angelo Nodari, Marco Vanetti albertini.simone@gmail.com

More information

Deep condolence to Professor Mark Everingham

Deep condolence to Professor Mark Everingham Deep condolence to Professor Mark Everingham Towards VOC2012 Object Classification Challenge Generalized Hierarchical Matching for Sub-category Aware Object Classification National University of Singapore

More information

Semantic Pooling for Image Categorization using Multiple Kernel Learning

Semantic Pooling for Image Categorization using Multiple Kernel Learning Semantic Pooling for Image Categorization using Multiple Kernel Learning Thibaut Durand (1,2), Nicolas Thome (1), Matthieu Cord (1), David Picard (2) (1) Sorbonne Universités, UPMC Univ Paris 06, UMR 7606,

More information

CPSC340. State-of-the-art Neural Networks. Nando de Freitas November, 2012 University of British Columbia

CPSC340. State-of-the-art Neural Networks. Nando de Freitas November, 2012 University of British Columbia CPSC340 State-of-the-art Neural Networks Nando de Freitas November, 2012 University of British Columbia Outline of the lecture This lecture provides an overview of two state-of-the-art neural networks:

More information

Category-level localization

Category-level localization Category-level localization Cordelia Schmid Recognition Classification Object present/absent in an image Often presence of a significant amount of background clutter Localization / Detection Localize object

More information

Learning Representations for Visual Object Class Recognition

Learning Representations for Visual Object Class Recognition Learning Representations for Visual Object Class Recognition Marcin Marszałek Cordelia Schmid Hedi Harzallah Joost van de Weijer LEAR, INRIA Grenoble, Rhône-Alpes, France October 15th, 2007 Bag-of-Features

More information

Fine-tuning Pre-trained Large Scaled ImageNet model on smaller dataset for Detection task

Fine-tuning Pre-trained Large Scaled ImageNet model on smaller dataset for Detection task Fine-tuning Pre-trained Large Scaled ImageNet model on smaller dataset for Detection task Kyunghee Kim Stanford University 353 Serra Mall Stanford, CA 94305 kyunghee.kim@stanford.edu Abstract We use a

More information

Part 5: Structured Support Vector Machines

Part 5: Structured Support Vector Machines Part 5: Structured Support Vector Machines Sebastian Nowozin and Christoph H. Lampert Providence, 21st June 2012 1 / 34 Problem (Loss-Minimizing Parameter Learning) Let d(x, y) be the (unknown) true data

More information

Ranking Figure-Ground Hypotheses for Object Segmentation

Ranking Figure-Ground Hypotheses for Object Segmentation Ranking Figure-Ground Hypotheses for Object Segmentation João Carreira, Fuxin Li, Cristian Sminchisescu Faculty of Mathematics and Natural Science, INS, University of Bonn http://sminchisescu.ins.uni-bonn.de/

More information

MEDICAL IMAGE SEGMENTATION WITH DIGITS. Hyungon Ryu, Jack Han Solution Architect NVIDIA Corporation

MEDICAL IMAGE SEGMENTATION WITH DIGITS. Hyungon Ryu, Jack Han Solution Architect NVIDIA Corporation MEDICAL IMAGE SEGMENTATION WITH DIGITS Hyungon Ryu, Jack Han Solution Architect NVIDIA Corporation MEDICAL IMAGE SEGMENTATION WITH DIGITS Overview Prepare Dataset Configure DL Model DL Training Segmentation

More information

Object Detection Based on Deep Learning

Object Detection Based on Deep Learning Object Detection Based on Deep Learning Yurii Pashchenko AI Ukraine 2016, Kharkiv, 2016 Image classification (mostly what you ve seen) http://tutorial.caffe.berkeleyvision.org/caffe-cvpr15-detection.pdf

More information

Analysis: TextonBoost and Semantic Texton Forests. Daniel Munoz Februrary 9, 2009

Analysis: TextonBoost and Semantic Texton Forests. Daniel Munoz Februrary 9, 2009 Analysis: TextonBoost and Semantic Texton Forests Daniel Munoz 16-721 Februrary 9, 2009 Papers [shotton-eccv-06] J. Shotton, J. Winn, C. Rother, A. Criminisi, TextonBoost: Joint Appearance, Shape and Context

More information

PASCAL VOC Classification: Local Features vs. Deep Features. Shuicheng YAN, NUS

PASCAL VOC Classification: Local Features vs. Deep Features. Shuicheng YAN, NUS PASCAL VOC Classification: Local Features vs. Deep Features Shuicheng YAN, NUS PASCAL VOC Why valuable? Multi-label, Real Scenarios! Visual Object Recognition Object Classification Object Detection Object

More information

Learning Object Representations for Visual Object Class Recognition

Learning Object Representations for Visual Object Class Recognition Learning Object Representations for Visual Object Class Recognition Marcin Marszalek, Cordelia Schmid, Hedi Harzallah, Joost Van de Weijer To cite this version: Marcin Marszalek, Cordelia Schmid, Hedi

More information

Supplementary Material: Pixelwise Instance Segmentation with a Dynamically Instantiated Network

Supplementary Material: Pixelwise Instance Segmentation with a Dynamically Instantiated Network Supplementary Material: Pixelwise Instance Segmentation with a Dynamically Instantiated Network Anurag Arnab and Philip H.S. Torr University of Oxford {anurag.arnab, philip.torr}@eng.ox.ac.uk 1. Introduction

More information

Large-Scale Live Active Learning: Training Object Detectors with Crawled Data and Crowds

Large-Scale Live Active Learning: Training Object Detectors with Crawled Data and Crowds Large-Scale Live Active Learning: Training Object Detectors with Crawled Data and Crowds Sudheendra Vijayanarasimhan Kristen Grauman Department of Computer Science University of Texas at Austin Austin,

More information

Visual Recognition: Examples of Graphical Models

Visual Recognition: Examples of Graphical Models Visual Recognition: Examples of Graphical Models Raquel Urtasun TTI Chicago March 6, 2012 Raquel Urtasun (TTI-C) Visual Recognition March 6, 2012 1 / 64 Graphical models Applications Representation Inference

More information

Part 5: Structured Support Vector Machines

Part 5: Structured Support Vector Machines Part 5: Structured Support Vector Machines Sebastian Noozin and Christoph H. Lampert Colorado Springs, 25th June 2011 1 / 56 Problem (Loss-Minimizing Parameter Learning) Let d(x, y) be the (unknon) true

More information

Bag-of-features. Cordelia Schmid

Bag-of-features. Cordelia Schmid Bag-of-features for category classification Cordelia Schmid Visual search Particular objects and scenes, large databases Category recognition Image classification: assigning a class label to the image

More information

Attributes. Computer Vision. James Hays. Many slides from Derek Hoiem

Attributes. Computer Vision. James Hays. Many slides from Derek Hoiem Many slides from Derek Hoiem Attributes Computer Vision James Hays Recap: Human Computation Active Learning: Let the classifier tell you where more annotation is needed. Human-in-the-loop recognition:

More information

Object Detection. TA : Young-geun Kim. Biostatistics Lab., Seoul National University. March-June, 2018

Object Detection. TA : Young-geun Kim. Biostatistics Lab., Seoul National University. March-June, 2018 Object Detection TA : Young-geun Kim Biostatistics Lab., Seoul National University March-June, 2018 Seoul National University Deep Learning March-June, 2018 1 / 57 Index 1 Introduction 2 R-CNN 3 YOLO 4

More information

MULTIMODAL SEMI-SUPERVISED IMAGE CLASSIFICATION BY COMBINING TAG REFINEMENT, GRAPH-BASED LEARNING AND SUPPORT VECTOR REGRESSION

MULTIMODAL SEMI-SUPERVISED IMAGE CLASSIFICATION BY COMBINING TAG REFINEMENT, GRAPH-BASED LEARNING AND SUPPORT VECTOR REGRESSION MULTIMODAL SEMI-SUPERVISED IMAGE CLASSIFICATION BY COMBINING TAG REFINEMENT, GRAPH-BASED LEARNING AND SUPPORT VECTOR REGRESSION Wenxuan Xie, Zhiwu Lu, Yuxin Peng and Jianguo Xiao Institute of Computer

More information

Constrained Convolutional Neural Networks for Weakly Supervised Segmentation. Deepak Pathak, Philipp Krähenbühl and Trevor Darrell

Constrained Convolutional Neural Networks for Weakly Supervised Segmentation. Deepak Pathak, Philipp Krähenbühl and Trevor Darrell Constrained Convolutional Neural Networks for Weakly Supervised Segmentation Deepak Pathak, Philipp Krähenbühl and Trevor Darrell 1 Multi-class Image Segmentation Assign a class label to each pixel in

More information

Object Recognition II

Object Recognition II Object Recognition II Linda Shapiro EE/CSE 576 with CNN slides from Ross Girshick 1 Outline Object detection the task, evaluation, datasets Convolutional Neural Networks (CNNs) overview and history Region-based

More information

A Comparison of l 1 Norm and l 2 Norm Multiple Kernel SVMs in Image and Video Classification

A Comparison of l 1 Norm and l 2 Norm Multiple Kernel SVMs in Image and Video Classification A Comparison of l 1 Norm and l 2 Norm Multiple Kernel SVMs in Image and Video Classification Fei Yan Krystian Mikolajczyk Josef Kittler Muhammad Tahir Centre for Vision, Speech and Signal Processing University

More information

Robust Near-Isometric Matching via Structured Learning of Graphical Models

Robust Near-Isometric Matching via Structured Learning of Graphical Models Robust Near-Isometric Matching via Structured Learning of Graphical Models Julian J. McAuley NICTA/ANU julian.mcauley @nicta.com.au Tibério S. Caetano NICTA/ANU tiberio.caetano @nicta.com.au Alexander

More information

An Object Detection Algorithm based on Deformable Part Models with Bing Features Chunwei Li1, a and Youjun Bu1, b

An Object Detection Algorithm based on Deformable Part Models with Bing Features Chunwei Li1, a and Youjun Bu1, b 5th International Conference on Advanced Materials and Computer Science (ICAMCS 2016) An Object Detection Algorithm based on Deformable Part Models with Bing Features Chunwei Li1, a and Youjun Bu1, b 1

More information

Multi-class image segmentation using Conditional Random Fields and Global Classification

Multi-class image segmentation using Conditional Random Fields and Global Classification Multi-class image segmentation using Conditional Random Fields and Global Classification Nils Plath nilsp@cs.tu-berlin.de Marc Toussaint mtoussai@cs.tu-berlin.de TU Berlin, Machine Learning Group, Franklinstrasse

More information

Big and Tall: Large Margin Learning with High Order Losses

Big and Tall: Large Margin Learning with High Order Losses Big and Tall: Large Margin Learning with High Order Losses Daniel Tarlow University of Toronto dtarlow@cs.toronto.edu Richard Zemel University of Toronto zemel@cs.toronto.edu Abstract Graphical models

More information

Unified, real-time object detection

Unified, real-time object detection Unified, real-time object detection Final Project Report, Group 02, 8 Nov 2016 Akshat Agarwal (13068), Siddharth Tanwar (13699) CS698N: Recent Advances in Computer Vision, Jul Nov 2016 Instructor: Gaurav

More information

Generating Descriptions of Spatial Relations between Objects in Images

Generating Descriptions of Spatial Relations between Objects in Images Generating Descriptions of Spatial Relations between Objects in Images Adrian Muscat Communications & Computer Engineering University of Malta Msida MSD 2080, Malta adrian.muscat@um.edu.mt Anja Belz Computing,

More information

arxiv: v1 [cs.cv] 2 May 2017

arxiv: v1 [cs.cv] 2 May 2017 SHI,SIVA,XIANG: TRANSFER LEARNING BY RANKING 1 arxiv:1705.00873v1 [cs.cv] 2 May 2017 Transfer Learning by Ranking for Weakly Supervised Object Annotation Zhiyuan Shi zhiyuan.shi@eecs.qmul.ac.uk Parthipan

More information

Human detection using histogram of oriented gradients. Srikumar Ramalingam School of Computing University of Utah

Human detection using histogram of oriented gradients. Srikumar Ramalingam School of Computing University of Utah Human detection using histogram of oriented gradients Srikumar Ramalingam School of Computing University of Utah Reference Navneet Dalal and Bill Triggs, Histograms of Oriented Gradients for Human Detection,

More information

Machine Learning for NLP

Machine Learning for NLP Machine Learning for NLP Support Vector Machines Aurélie Herbelot 2018 Centre for Mind/Brain Sciences University of Trento 1 Support Vector Machines: introduction 2 Support Vector Machines (SVMs) SVMs

More information

Structured Models in. Dan Huttenlocher. June 2010

Structured Models in. Dan Huttenlocher. June 2010 Structured Models in Computer Vision i Dan Huttenlocher June 2010 Structured Models Problems where output variables are mutually dependent or constrained E.g., spatial or temporal relations Such dependencies

More information

Self-Paced Learning for Semisupervised Image Classification

Self-Paced Learning for Semisupervised Image Classification Self-Paced Learning for Semisupervised Image Classification Kevin Miller Stanford University Palo Alto, CA kjmiller@stanford.edu Abstract In this project, we apply three variants of self-paced learning

More information

Single-Shot Refinement Neural Network for Object Detection -Supplementary Material-

Single-Shot Refinement Neural Network for Object Detection -Supplementary Material- Single-Shot Refinement Neural Network for Object Detection -Supplementary Material- Shifeng Zhang 1,2, Longyin Wen 3, Xiao Bian 3, Zhen Lei 1,2, Stan Z. Li 4,1,2 1 CBSR & NLPR, Institute of Automation,

More information

Apprenticeship Learning: Transfer of Knowledge via Dataset Augmentation

Apprenticeship Learning: Transfer of Knowledge via Dataset Augmentation Apprenticeship Learning: Transfer of Knowledge via Dataset Augmentation Miroslav Kobetski and Josephine Sullivan Computer Vision and Active Perception, KTH, 114 28 Stockholm kobetski@kth.se, sullivan@nada.kth.se

More information

SVMs for Structured Output. Andrea Vedaldi

SVMs for Structured Output. Andrea Vedaldi SVMs for Structured Output Andrea Vedaldi SVM struct Tsochantaridis Hofmann Joachims Altun 04 Extending SVMs 3 Extending SVMs SVM = parametric function arbitrary input binary output 3 Extending SVMs SVM

More information

Optimizing Object Detection:

Optimizing Object Detection: Lecture 10: Optimizing Object Detection: A Case Study of R-CNN, Fast R-CNN, and Faster R-CNN Visual Computing Systems Today s task: object detection Image classification: what is the object in this image?

More information

Object recognition (part 2)

Object recognition (part 2) Object recognition (part 2) CSE P 576 Larry Zitnick (larryz@microsoft.com) 1 2 3 Support Vector Machines Modified from the slides by Dr. Andrew W. Moore http://www.cs.cmu.edu/~awm/tutorials Linear Classifiers

More information

Machine Learning. Support Vector Machines. Fabio Vandin November 20, 2017

Machine Learning. Support Vector Machines. Fabio Vandin November 20, 2017 Machine Learning Support Vector Machines Fabio Vandin November 20, 2017 1 Classification and Margin Consider a classification problem with two classes: instance set X = R d label set Y = { 1, 1}. Training

More information

Iterative Category Discovery via Multiple Kernel Metric Learning

Iterative Category Discovery via Multiple Kernel Metric Learning Int J Comput Vis (2014) 108:115 132 DOI 10.1007/s11263-013-0679-z Iterative Category Discovery via Multiple Kernel Metric Learning Carolina Galleguillos Brian McFee Gert R. G. Lanckriet Received: 26 February

More information

Support Vector Machine Learning for Interdependent and Structured Output Spaces

Support Vector Machine Learning for Interdependent and Structured Output Spaces Support Vector Machine Learning for Interdependent and Structured Output Spaces I. Tsochantaridis, T. Hofmann, T. Joachims, and Y. Altun, ICML, 2004. And also I. Tsochantaridis, T. Joachims, T. Hofmann,

More information

24 hours of Photo Sharing. installation by Erik Kessels

24 hours of Photo Sharing. installation by Erik Kessels 24 hours of Photo Sharing installation by Erik Kessels And sometimes Internet photos have useful labels Im2gps. Hays and Efros. CVPR 2008 But what if we want more? Image Categorization Training Images

More information

Harmony Potentials for Joint Classification and Segmentation

Harmony Potentials for Joint Classification and Segmentation Harmony Potentials for Joint Classification and Segmentation Josep M. Gonfaus 1,2, Xavier Boix 1, Joost van de Weijer 1,2 Andrew D. Bagdanov 1 Joan Serrat 1,2 Jordi Gonzàlez 1,2 1 Centre de Visió per Computador

More information

Sparse Coding and Dictionary Learning for Image Analysis

Sparse Coding and Dictionary Learning for Image Analysis Sparse Coding and Dictionary Learning for Image Analysis Part IV: Recent Advances in Computer Vision and New Models Francis Bach, Julien Mairal, Jean Ponce and Guillermo Sapiro CVPR 10 tutorial, San Francisco,

More information

Layered Object Detection for Multi-Class Segmentation

Layered Object Detection for Multi-Class Segmentation Layered Object Detection for Multi-Class Segmentation Yi Yang Sam Hallman Deva Ramanan Charless Fowlkes Dept. of Computer Science, University of California, Irvine {yangyi,shallman,dramanan,fowlkes}@ics.uci.edu

More information

Stealing Objects With Computer Vision

Stealing Objects With Computer Vision Stealing Objects With Computer Vision Learning Based Methods in Vision Analysis Project #4: Mar 4, 2009 Presented by: Brian C. Becker Carnegie Mellon University Motivation Goal: Detect objects in the photo

More information

Syllabus. 1. Visual classification Intro 2. SVM 3. Datasets and evaluation 4. Shallow / Deep architectures

Syllabus. 1. Visual classification Intro 2. SVM 3. Datasets and evaluation 4. Shallow / Deep architectures Syllabus 1. Visual classification Intro 2. SVM 3. Datasets and evaluation 4. Shallow / Deep architectures Image classification How to define a category? Bicycle Paintings with women Portraits Concepts,

More information

CS 1674: Intro to Computer Vision. Attributes. Prof. Adriana Kovashka University of Pittsburgh November 2, 2016

CS 1674: Intro to Computer Vision. Attributes. Prof. Adriana Kovashka University of Pittsburgh November 2, 2016 CS 1674: Intro to Computer Vision Attributes Prof. Adriana Kovashka University of Pittsburgh November 2, 2016 Plan for today What are attributes and why are they useful? (paper 1) Attributes for zero-shot

More information

Detection III: Analyzing and Debugging Detection Methods

Detection III: Analyzing and Debugging Detection Methods CS 1699: Intro to Computer Vision Detection III: Analyzing and Debugging Detection Methods Prof. Adriana Kovashka University of Pittsburgh November 17, 2015 Today Review: Deformable part models How can

More information

Domain Adapta,on in a deep learning context. Tinne Tuytelaars

Domain Adapta,on in a deep learning context. Tinne Tuytelaars Domain Adapta,on in a deep learning context Tinne Tuytelaars Work in collabora,on with T. Tommasi, N. Patricia, B. Caputo, T. Tuytelaars "A Deeper Look at Dataset Bias" GCPR 2015 A. Raj, V. Namboodiri

More information

Find that! Visual Object Detection Primer

Find that! Visual Object Detection Primer Find that! Visual Object Detection Primer SkTech/MIT Innovation Workshop August 16, 2012 Dr. Tomasz Malisiewicz tomasz@csail.mit.edu Find that! Your Goals...imagine one such system that drives information

More information

Multi-net System Configuration for Visual Object Segmentation by Error Backpropagation

Multi-net System Configuration for Visual Object Segmentation by Error Backpropagation Multi-net System Configuration for Visual Object Segmentation by Error Backpropagation Ignazio Gallo, Marco Vanetti, Simone Albertini, and Angelo Nodari Università dell Insubria, Dipartimento di Scienze

More information

Multiple cosegmentation

Multiple cosegmentation Armand Joulin, Francis Bach and Jean Ponce. INRIA -Ecole Normale Supérieure April 25, 2012 Segmentation Introduction Segmentation Supervised and weakly-supervised segmentation Cosegmentation Segmentation

More information

Complex Prediction Problems

Complex Prediction Problems Problems A novel approach to multiple Structured Output Prediction Max-Planck Institute ECML HLIE08 Information Extraction Extract structured information from unstructured data Typical subtasks Named Entity

More information

All lecture slides will be available at CSC2515_Winter15.html

All lecture slides will be available at  CSC2515_Winter15.html CSC2515 Fall 2015 Introduc3on to Machine Learning Lecture 9: Support Vector Machines All lecture slides will be available at http://www.cs.toronto.edu/~urtasun/courses/csc2515/ CSC2515_Winter15.html Many

More information

Computer Vision: Summary and Discussion

Computer Vision: Summary and Discussion 12/05/2011 Computer Vision: Summary and Discussion Computer Vision CS 143, Brown James Hays Many slides from Derek Hoiem Announcements Today is last day of regular class Second quiz on Wednesday (Dec 7

More information

Deformable Part Models

Deformable Part Models CS 1674: Intro to Computer Vision Deformable Part Models Prof. Adriana Kovashka University of Pittsburgh November 9, 2016 Today: Object category detection Window-based approaches: Last time: Viola-Jones

More information

Markov Networks in Computer Vision

Markov Networks in Computer Vision Markov Networks in Computer Vision Sargur Srihari srihari@cedar.buffalo.edu 1 Markov Networks for Computer Vision Some applications: 1. Image segmentation 2. Removal of blur/noise 3. Stereo reconstruction

More information

Learning and Recognizing Visual Object Categories Without First Detecting Features

Learning and Recognizing Visual Object Categories Without First Detecting Features Learning and Recognizing Visual Object Categories Without First Detecting Features Daniel Huttenlocher 2007 Joint work with D. Crandall and P. Felzenszwalb Object Category Recognition Generic classes rather

More information

Lecture 9. Support Vector Machines

Lecture 9. Support Vector Machines Lecture 9. Support Vector Machines COMP90051 Statistical Machine Learning Semester 2, 2017 Lecturer: Andrey Kan Copyright: University of Melbourne This lecture Support vector machines (SVMs) as maximum

More information

Markov Networks in Computer Vision. Sargur Srihari

Markov Networks in Computer Vision. Sargur Srihari Markov Networks in Computer Vision Sargur srihari@cedar.buffalo.edu 1 Markov Networks for Computer Vision Important application area for MNs 1. Image segmentation 2. Removal of blur/noise 3. Stereo reconstruction

More information

Modeling 3D viewpoint for part-based object recognition of rigid objects

Modeling 3D viewpoint for part-based object recognition of rigid objects Modeling 3D viewpoint for part-based object recognition of rigid objects Joshua Schwartz Department of Computer Science Cornell University jdvs@cs.cornell.edu Abstract Part-based object models based on

More information

2282 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 21, NO. 4, APRIL 2012

2282 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 21, NO. 4, APRIL 2012 2282 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 21, NO. 4, APRIL 2012 Task-Dependent Visual-Codebook Compression Rongrong Ji, Member, IEEE, Hongxun Yao, Member, IEEE, Wei Liu, Student Member, IEEE, Xiaoshuai

More information

Generalized RBF feature maps for Efficient Detection

Generalized RBF feature maps for Efficient Detection SREEKANTH.V et al. : GENERALIZED RBF FEATURE MAPS Generalized RBF feature maps for Efficient Detection Sreekanth Vempati v_sreekanth@research.iiit.ac.in Andrea Vedaldi vedaldi@robots.ox.ac.uk Andrew Zisserman

More information

segdeepm: Exploiting Segmentation and Context in Deep Neural Networks for Object Detection

segdeepm: Exploiting Segmentation and Context in Deep Neural Networks for Object Detection : Exploiting Segmentation and Context in Deep Neural Networks for Object Detection Yukun Zhu Raquel Urtasun Ruslan Salakhutdinov Sanja Fidler University of Toronto {yukun,urtasun,rsalakhu,fidler}@cs.toronto.edu

More information

Mixture Component Identification and Learning for Visual Recognition

Mixture Component Identification and Learning for Visual Recognition Mixture Component Identification and Learning for Visual Recognition Omid Aghazadeh, Hossein Azizpour, Josephine Sullivan, and Stefan Carlsson Computer Vision and Active Perception laboratory (CVAP), KTH,

More information

EECS 442 Computer vision Databases for object recognition and beyond Segments of this lectures are courtesy of Prof F. Li, R. Fergus and A.

EECS 442 Computer vision Databases for object recognition and beyond Segments of this lectures are courtesy of Prof F. Li, R. Fergus and A. EECS 442 Computer vision Databases for object recognition and beyond Segments of this lectures are courtesy of Prof F. Li, R. Fergus and A. Zisserman Why are we talking about datasets? This is computer

More information

Learning to Localize Objects with Structured Output Regression

Learning to Localize Objects with Structured Output Regression Learning to Localize Objects with Structured Output Regression Matthew Blaschko and Christopher Lampert ECCV 2008 Best Student Paper Award Presentation by Jaeyong Sung and Yiting Xie 1 Object Localization

More information

arxiv: v3 [cs.cv] 15 Sep 2018

arxiv: v3 [cs.cv] 15 Sep 2018 DPATCH: An Adversarial Patch Attack on Object Detectors Xin Liu1, Huanrui Yang1, Ziwei Liu2, Linghao Song1, Hai Li1, Yiran Chen1, arxiv:1806.02299v3 [cs.cv] 15 Sep 2018 2 1 Duke University The Chinese

More information

Video Semantic Indexing using Object Detection-Derived Features

Video Semantic Indexing using Object Detection-Derived Features Video Semantic Indexing using Object Detection-Derived Features Kotaro Kikuchi, Kazuya Ueki, Tetsuji Ogawa, and Tetsunori Kobayashi Dept. of Computer Science, Waseda University, Japan Abstract A new feature

More information

The PASCAL Visual Object Classes Challenge 2010 (VOC2010) Development Kit

The PASCAL Visual Object Classes Challenge 2010 (VOC2010) Development Kit The PASCAL Visual Object Classes Challenge 2010 (VOC2010) Development Kit Mark Everingham John Winn May 8, 2010 Contents 1 Challenge 3 2 Data 3 2.1 Classification/Detection Image Sets.................

More information

Optimizing Intersection-Over-Union in Deep Neural Networks for Image Segmentation

Optimizing Intersection-Over-Union in Deep Neural Networks for Image Segmentation Optimizing Intersection-Over-Union in Deep Neural Networks for Image Segmentation Md Atiqur Rahman and Yang Wang Department of Computer Science, University of Manitoba, Canada {atique, ywang}@cs.umanitoba.ca

More information

Gradient of the lower bound

Gradient of the lower bound Weakly Supervised with Latent PhD advisor: Dr. Ambedkar Dukkipati Department of Computer Science and Automation gaurav.pandey@csa.iisc.ernet.in Objective Given a training set that comprises image and image-level

More information

Generative and discriminative classification techniques

Generative and discriminative classification techniques Generative and discriminative classification techniques Machine Learning and Category Representation 013-014 Jakob Verbeek, December 13+0, 013 Course website: http://lear.inrialpes.fr/~verbeek/mlcr.13.14

More information

A Family of Contextual Measures of Similarity between Distributions with Application to Image Retrieval

A Family of Contextual Measures of Similarity between Distributions with Application to Image Retrieval A Family of Contextual Measures of Similarity between Distributions with Application to Image Retrieval Florent Perronnin, Yan Liu and Jean-Michel Renders Xerox Research Centre Europe (XRCE) Textual and

More information

Object Segmentation by Alignment of Poselet Activations to Image Contours

Object Segmentation by Alignment of Poselet Activations to Image Contours Object Segmentation by Alignment of Poselet Activations to Image Contours Thomas Brox 1, Lubomir Bourdev 2,3, Subhransu Maji 2, and Jitendra Malik 2 1 University of Freiburg, Germany 2 University of California

More information

Top-down Saliency with Locality-constrained Contextual Sparse Coding

Top-down Saliency with Locality-constrained Contextual Sparse Coding CHOLAKKAL et al.: TOP-DOWN SALIENCY WITH LCCSC 1 Top-down Saliency with Locality-constrained Contextual Sparse Coding Hisham Cholakkal hisham002@ntu.edu.sg Deepu Rajan http://www3.ntu.edu.sg/home/asdrajan

More information

Sparse coding for image classification

Sparse coding for image classification Sparse coding for image classification Columbia University Electrical Engineering: Kun Rong(kr2496@columbia.edu) Yongzhou Xiang(yx2211@columbia.edu) Yin Cui(yc2776@columbia.edu) Outline Background Introduction

More information

SVM: Multiclass and Structured Prediction. Bin Zhao

SVM: Multiclass and Structured Prediction. Bin Zhao SVM: Multiclass and Structured Prediction Bin Zhao Part I: Multi-Class SVM 2-Class SVM Primal form Dual form http://www.glue.umd.edu/~zhelin/recog.html Real world classification problems Digit recognition

More information

Real-Time Object Segmentation Using a Bag of Features Approach

Real-Time Object Segmentation Using a Bag of Features Approach Real-Time Object Segmentation Using a Bag of Features Approach David ALDAVERT a,1, Arnau RAMISA c,b, Ramon LOPEZ DE MANTARAS b and Ricardo TOLEDO a a Computer Vision Center, Dept. Ciencies de la Computació,

More information

Discriminative models for multi-class object layout

Discriminative models for multi-class object layout Discriminative models for multi-class object layout Chaitanya Desai Deva Ramanan Charless Fowlkes Department of Computer Science U C Irvine {desaic,dramanan,fowlkes}@ics.uci.edu Abstract those subtle statistical

More information

Linear combinations of simple classifiers for the PASCAL challenge

Linear combinations of simple classifiers for the PASCAL challenge Linear combinations of simple classifiers for the PASCAL challenge Nik A. Melchior and David Lee 16 721 Advanced Perception The Robotics Institute Carnegie Mellon University Email: melchior@cmu.edu, dlee1@andrew.cmu.edu

More information

arxiv: v1 [cs.cv] 21 Sep 2008

arxiv: v1 [cs.cv] 21 Sep 2008 Robust Near-Isometric Matching via Structured Learning of Graphical Models Julian J. McAuley, Tibério S. Caetano and Alexander J. Smola October 25, 28 arxiv:89.368v [cs.cv] 2 Sep 28 Abstract Models for

More information

DPATCH: An Adversarial Patch Attack on Object Detectors

DPATCH: An Adversarial Patch Attack on Object Detectors DPATCH: An Adversarial Patch Attack on Object Detectors Xin Liu1, Huanrui Yang1, Ziwei Liu2, Linghao Song1, Hai Li1, Yiran Chen1 1 Duke 1 University, 2 The Chinese University of Hong Kong {xin.liu4, huanrui.yang,

More information

Semantic Segmentation without Annotating Segments

Semantic Segmentation without Annotating Segments Chapter 3 Semantic Segmentation without Annotating Segments Numerous existing object segmentation frameworks commonly utilize the object bounding box as a prior. In this chapter, we address semantic segmentation

More information

Hierarchical Learning for Object Detection. Long Zhu, Yuanhao Chen, William Freeman, Alan Yuille, Antonio Torralba MIT and UCLA, 2010

Hierarchical Learning for Object Detection. Long Zhu, Yuanhao Chen, William Freeman, Alan Yuille, Antonio Torralba MIT and UCLA, 2010 Hierarchical Learning for Object Detection Long Zhu, Yuanhao Chen, William Freeman, Alan Yuille, Antonio Torralba MIT and UCLA, 2010 Background I: Our prior work Our work for the Pascal Challenge is based

More information

Image classification using object detectors

Image classification using object detectors Image classification using object detectors Thibaut Durand, Nicolas Thome, Matthieu Cord, Sandra Avila To cite this version: Thibaut Durand, Nicolas Thome, Matthieu Cord, Sandra Avila. Image classification

More information

Lecture 9: Support Vector Machines

Lecture 9: Support Vector Machines Lecture 9: Support Vector Machines William Webber (william@williamwebber.com) COMP90042, 2014, Semester 1, Lecture 8 What we ll learn in this lecture Support Vector Machines (SVMs) a highly robust and

More information

CSE373: Data Structures & Algorithms Lecture 23: Applications. Linda Shapiro Spring 2016

CSE373: Data Structures & Algorithms Lecture 23: Applications. Linda Shapiro Spring 2016 CSE373: Data Structures & Algorithms Lecture 23: Applications Linda Shapiro Spring 2016 Announcements Spring 2016 CSE373: Data Structures & Algorithms 2 Other Data Structures and Algorithms Quadtrees:

More information

arxiv: v2 [cs.cv] 22 Sep 2014

arxiv: v2 [cs.cv] 22 Sep 2014 arxiv:47.6v2 [cs.cv] 22 Sep 24 Analyzing the Performance of Multilayer Neural Networks for Object Recognition Pulkit Agrawal, Ross Girshick, Jitendra Malik {pulkitag,rbg,malik}@eecs.berkeley.edu University

More information

CSE 573: Artificial Intelligence Autumn 2010

CSE 573: Artificial Intelligence Autumn 2010 CSE 573: Artificial Intelligence Autumn 2010 Lecture 16: Machine Learning Topics 12/7/2010 Luke Zettlemoyer Most slides over the course adapted from Dan Klein. 1 Announcements Syllabus revised Machine

More information

Exploiting Depth from Single Monocular Images for Object Detection and Semantic Segmentation

Exploiting Depth from Single Monocular Images for Object Detection and Semantic Segmentation APPEARING IN IEEE TRANSACTIONS ON IMAGE PROCESSING, OCTOBER 2016 1 Exploiting Depth from Single Monocular Images for Object Detection and Semantic Segmentation Yuanzhouhan Cao, Chunhua Shen, Heng Tao Shen

More information

HIERARCHICAL JOINT-GUIDED NETWORKS FOR SEMANTIC IMAGE SEGMENTATION

HIERARCHICAL JOINT-GUIDED NETWORKS FOR SEMANTIC IMAGE SEGMENTATION HIERARCHICAL JOINT-GUIDED NETWORKS FOR SEMANTIC IMAGE SEGMENTATION Chien-Yao Wang, Jyun-Hong Li, Seksan Mathulaprangsan, Chin-Chin Chiang, and Jia-Ching Wang Department of Computer Science and Information

More information

SUPPORT VECTOR MACHINES

SUPPORT VECTOR MACHINES SUPPORT VECTOR MACHINES Today Reading AIMA 18.9 Goals (Naïve Bayes classifiers) Support vector machines 1 Support Vector Machines (SVMs) SVMs are probably the most popular off-the-shelf classifier! Software

More information

Computer Vision: Making machines see

Computer Vision: Making machines see Computer Vision: Making machines see Roberto Cipolla Department of Engineering http://www.eng.cam.ac.uk/~cipolla/people.html http://www.toshiba.eu/eu/cambridge-research- Laboratory/ Vision: what is where

More information

Introduction to Machine Learning

Introduction to Machine Learning Introduction to Machine Learning Maximum Margin Methods Varun Chandola Computer Science & Engineering State University of New York at Buffalo Buffalo, NY, USA chandola@buffalo.edu Chandola@UB CSE 474/574

More information