Part Based Models Andrew Harp. Part Based Models Physical arrangement of features.

Slides:



Advertisements
Similar presentations
Pseudo-Relevance Feedback For Multimedia Retrieval By Rong Yan, Alexander G. and Rong Jin Mwangi S. Kariuki
Advertisements

Active Appearance Models
Object Recognition from Local Scale-Invariant Features David G. Lowe Presented by Ashley L. Kapron.
Random Forest Predrag Radenković 3237/10
Clustering Basic Concepts and Algorithms
Sensor-Based Abnormal Human-Activity Detection Authors: Jie Yin, Qiang Yang, and Jeffrey Junfeng Pan Presenter: Raghu Rangan.
Human Identity Recognition in Aerial Images Omar Oreifej Ramin Mehran Mubarak Shah CVPR 2010, June Computer Vision Lab of UCF.
Ľubor Ladický1 Phil Torr2 Andrew Zisserman1
Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.
Recognition by Probabilistic Hypothesis Construction P. Moreels, M. Maire, P. Perona California Institute of Technology.
Multi-Local Feature Manifolds for Object Detection Oscar Danielsson Stefan Carlsson Josephine Sullivan
Zhenwen Dai Jӧrg Lücke Frankfurt Institute for Advanced Studies,
Hidden Variables, the EM Algorithm, and Mixtures of Gaussians Computer Vision CS 543 / ECE 549 University of Illinois Derek Hoiem 03/15/12.
A. Darwiche Learning in Bayesian Networks. A. Darwiche Known Structure Complete Data Known Structure Incomplete Data Unknown Structure Complete Data Unknown.
Uncertainty Representation. Gaussian Distribution variance Standard deviation.
Bag-of-features models Many slides adapted from Fei-Fei Li, Rob Fergus, and Antonio Torralba.
Model: Parts and Structure. History of Idea Fischler & Elschlager 1973 Yuille ‘91 Brunelli & Poggio ‘93 Lades, v.d. Malsburg et al. ‘93 Cootes, Lanitis,
Overview Full Bayesian Learning MAP learning
Beyond bags of features: Part-based models Many slides adapted from Fei-Fei Li, Rob Fergus, and Antonio Torralba.
Object Recognition Using Genetic Algorithms CS773C Advanced Machine Intelligence Applications Spring 2008: Object Recognition.
A Study of Approaches for Object Recognition
Beyond bags of features: Adding spatial information Many slides adapted from Fei-Fei Li, Rob Fergus, and Antonio Torralba.
Multi-Class Object Recognition Using Shared SIFT Features
Object Recognition by Parts Object recognition started with line segments. - Roberts recognized objects from line segments and junctions. - This led to.
Bag-of-features models
Lecture 17: Parts-based models and context CS6670: Computer Vision Noah Snavely.
Object Class Recognition by Unsupervised Scale-Invariant Learning R. Fergus, P. Perona, and A. Zisserman Presented By Jeff.
K-means Clustering. What is clustering? Why would we want to cluster? How would you determine clusters? How can you do this efficiently?
AN ANALYSIS OF SINGLE- LAYER NETWORKS IN UNSUPERVISED FEATURE LEARNING [1] Yani Chen 10/14/
Predicting Matchability - CVPR 2014 Paper -
Accurate, Dense and Robust Multi-View Stereopsis Yasutaka Furukawa and Jean Ponce Presented by Rahul Garg and Ryan Kaminsky.
Object Recognition by Parts Object recognition started with line segments. - Roberts recognized objects from line segments and junctions. - This led to.
Introduction to machine learning
Generic object detection with deformable part-based models
Exercise Session 10 – Image Categorization
Face Alignment Using Cascaded Boosted Regression Active Shape Models
Introduction Belief propagation: known to produce accurate results for stereo processing/ motion estimation High storage requirements limit the use of.
CSE 185 Introduction to Computer Vision Pattern Recognition.
PageRank for Product Image Search Kevin Jing (Googlc IncGVU, College of Computing, Georgia Institute of Technology) Shumeet Baluja (Google Inc.) WWW 2008.
COMMON EVALUATION FINAL PROJECT Vira Oleksyuk ECE 8110: Introduction to machine Learning and Pattern Recognition.
Bag-of-features models. Origin 1: Texture recognition Texture is characterized by the repetition of basic elements or textons For stochastic textures,
2 2  Background  Vision in Human Brain  Efficient Coding Theory  Motivation  Natural Pictures  Methodology  Statistical Characteristics  Models.
A Regression Approach to Music Emotion Recognition Yi-Hsuan Yang, Yu-Ching Lin, Ya-Fan Su, and Homer H. Chen, Fellow, IEEE IEEE TRANSACTIONS ON AUDIO,
Unsupervised Learning: Clustering Some material adapted from slides by Andrew Moore, CMU. Visit for
Object Detection with Discriminatively Trained Part Based Models
BING: Binarized Normed Gradients for Objectness Estimation at 300fps
Deformable Part Models (DPM) Felzenswalb, Girshick, McAllester & Ramanan (2010) Slides drawn from a tutorial By R. Girshick AP 12% 27% 36% 45% 49% 2005.
Template matching and object recognition. CS8690 Computer Vision University of Missouri at Columbia Matching by relations Idea: –find bits, then say object.
MSRI workshop, January 2005 Object Recognition Collected databases of objects on uniform background (no occlusions, no clutter) Mostly focus on viewpoint.
MACHINE LEARNING 8. Clustering. Motivation Based on E ALPAYDIN 2004 Introduction to Machine Learning © The MIT Press (V1.1) 2  Classification problem:
Information Retrieval Lecture 6 Introduction to Information Retrieval (Manning et al. 2007) Chapter 16 For the MSc Computer Science Programme Dell Zhang.
Gesture Recognition in a Class Room Environment Michael Wallick CS766.
Discussion of Pictorial Structures Pedro Felzenszwalb Daniel Huttenlocher Sicily Workshop September, 2006.
Pictorial Structures and Distance Transforms Computer Vision CS 543 / ECE 549 University of Illinois Ian Endres 03/31/11.
Learning and Acting with Bayes Nets Chapter 20.. Page 2 === A Network and a Training Data.
Carl Vondrick, Aditya Khosla, Tomasz Malisiewicz, Antonio Torralba Massachusetts Institute of Technology
Learning Kernel Classifiers 1. Introduction Summarized by In-Hee Lee.
Hidden Variables, the EM Algorithm, and Mixtures of Gaussians Computer Vision CS 543 / ECE 549 University of Illinois Derek Hoiem 02/22/11.
Learning in Bayesian Networks. Known Structure Complete Data Known Structure Incomplete Data Unknown Structure Complete Data Unknown Structure Incomplete.
CMA Coastline Matching Algorithm SSIP’99 - Project 10 Team H.
Machine Learning in Practice Lecture 9 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute.
A Discriminatively Trained, Multiscale, Deformable Part Model Yeong-Jun Cho Computer Vision and Pattern Recognition,2008.
CSCI 631 – Foundations of Computer Vision March 15, 2016 Ashwini Imran Image Stitching.
Ch 1. Introduction Pattern Recognition and Machine Learning, C. M. Bishop, Updated by J.-H. Eom (2 nd round revision) Summarized by K.-I.
Machine Learning overview Chapter 18, 21
Object Localization Goal: detect the location of an object within an image Fully supervised: Training data labeled with object category and ground truth.
Object Recognition by Parts
Image Segmentation Techniques
Object Recognition by Parts
Presentation transcript:

Part Based Models Andrew Harp

Part Based Models Physical arrangement of features

Code Based on the Simple Parts and Structure Object Detector by R. Fergus Allows user training on N images Supports a variety of models ▫Simple part based model ▫efficient model by Felzenszwalb and Huttenlocher ▫Naïve bayes ▫Probabilistic Latent Semantic Analysis

Steps Preprocessing ▫Resizes images and ground-truth information Training ▫User clicks on ordered features in random training examples Feature recognition ▫Heat maps of features are created in test image ▫Features are considered to be local maxima Object detection ▫Feature configurations are scored using model ▫Best score is selected as recognized object Evaluation ▫Bounding boxes are generated around guessed points and compared to ground truth data ▫RPC curves are computed

Training

Model Creation

Feature Recognition

My additions Expectation Maximization ▫Run trials iteratively HOG filtering ▫Compare HOG descriptor correlations to last known filters Heatmap shifting ▫Try to shift feature heatmaps over

Expectation Maximization (EM)

EM Algorithm Seeded with N initial training examples K iterations Uses X best matches from the previous iteration to extract new filters Model composition variance initially determined by user input beyond initial training, unsupervised ▫ground truth data is hidden from it ▫test data selection will influence model

EM Training Uses best scoring images from previous iteration as training examples Repeats guessed part positions as if it was ground truth information

EM Algorithm

Filters Over iterations

Variance Over Iterations

HOG filtering

HOG filtering (cont) Without HOG filteringWith HOG filtering Between iterations 1 and 2, HOG Filtering threw out 5 degenerate training examples in the top 20 scorers.

Heatmap Shifting Fast

Heatmap Shifting

Heatmap shifting

Observations Model Variance spikes initially as problem space is explored with messy filters, but fitness of matches with low variance forces it down

Efficient Method Felzenszwalb, P. and Huttenlocher Provides better results, converges quickly Prefers matches with low deviation from archetype, leading to low variance in next iteration. Had to jury-rig model variance to a constant because it converges to 0

Efficient Method

Degenerate Cases

Observations EM can reinforce degenerate cases. Requires some knowledge of the training data to find Can find largest cluster, while excluding outliers Helps most if initial input was bad, but correctable

Improvements Maintain multiple archetypes for each part ▫Would take advantage of iterative nature of EM to expand feature library Use better method than correlation for determining feature maps ▫HOG?

References Felzenszwalb, P. and Huttenlocher, D. "Pictorial Structures for Object Recognition." Intl. Journal of Computer Vision, 61(1), pp , January Fischler, M. and Elschlager, R. "The representation and matching of pictorial structures." IEEE Transactions on Computers, 22(1):67--92, 1973.

Thanks