Object class recognition using unsupervised scale-invariant learning Goal Some object categories Representation

Object class recognition using unsupervised scale-invariant learning Goal Some object categories Representation www.phwiki.com

Object class recognition using unsupervised scale-invariant learning Goal Some object categories Representation

Brock, Elizabeth, Interim Station Manager has reference to this Academic Journal, PHwiki organized this Journal Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Ox as long as d University Cali as long as nia Institute of Technology Goal Recognition of object categories Unassisted learning Some object categories Learn from examples Difficulties: Size variation Background clutter Occlusion Intra-class variation

Debre Birhan University ET www.phwiki.com

This Particular University is Related to this Particular Journal

Representation Learning Recognition Main issues Representation Learning Recognition Model: Constellation of Parts Fischler & Elschlager 1973 Yuille ‘91 Brunelli & Poggio ‘93 Lades, v.d. Malsburg et al. ‘93 Cootes, Lanitis, Taylor et al. ‘95 Amit & Geman ‘95, ‘99 Perona et al. ‘95, ‘96, ’98, ’00 Agarwal & Roth ‘02 Detection & Representation of regions Appearance Location Scale (x,y) coords. of region center Diameter of region (pixels) Gives representation of appearance in low-dimensional vector space Find regions within image Use Kadir in addition to Brady’s salient region operator [IJCV ’01]

Foreground model Gaussian shape pdf Poission pdf on detections Uni as long as m shape pdf Gaussian part appearance pdf Generative probabilistic model Clutter model Gaussian background appearance pdf based on Burl, Weber et al. [ECCV ’98, ’00] Recognition Motorbikes Samples from appearance model

Learning Task: Estimation of model parameters Learning Let the assignments be a hidden variable in addition to use EM algorithm to learn them in addition to the model parameters Chicken in addition to Egg type problem, since we initially know neither: Model parameters – Assignment of regions to as long as eground / background Learning procedure E-step: Compute assignments as long as which regions are as long as eground / background M-step: Update model parameters Find regions & their location, scale & appearance Initialize model parameters Use EM in addition to iterate to convergence: Trying to maximize likelihood – consistency in shape & appearance

Experiments Experimental procedure Two series of experiments: Fixed-scale model – Objects the same size (manual normalization) Scale-invariant model – Objects between 100 in addition to 550 pixels in width Datasets Training 50% images No identifcation of object within image Testing 50% images Simple object present/absent test Motorbikes

Background images evaluated with motorbike model Frontal faces Airplanes

Spotted cats Summary of results % equal error rate Note: Within each series, same settings used as long as all datasets Comparison to other methods % equal error rate

Robustness of Algorithm Summary Limitations future work Comprehensive probabilistic model as long as object classes Learn appearance, shape, relative scale, occlusion etc. simultaneously in scale in addition to translation invariant manner Same algorithm gives <= 10% error across 5 diverse datasets with identical settings Datasets available from: http://www.robots.ox.ac.uk/~vgg/data Very reliant on region detector Different part types (e.g. edgel curves) Only learns a single viewpoint Use mixture models Need lots of images to learn Bayesian learning - fewer images [ICCV ’03 (Fei Fei, Fergus, Perona)] Need more through testing Looking towards testing 100’s of datasets Ease of training Categories (log2) Overview of approaches to category recognition Unsupervised 0 Labelled Normalized & labelled Normalized & labelled & segmented 1 2 3 8 Schneiderman & Kanade [CVPR ’00] Weber et al. [ECCV ’00] Fergus et al. [CVPR ’03] 14 Humans Viola & Jones [CVPR ’01] Brock, Elizabeth WAPR-FM Interim Station Manager www.phwiki.com

Cars from rear ROC equal error rates Pre-scaled data (identical settings): Scale-invariant learning in addition to recognition: Image size histograms

Sampling from models Faces Motorbikes Overview of Object Representation It h in addition to les: Appearance Shape Relative scale Occlusion Background statistics all in a probabilistic manner Objects represented as a constellation of parts The constellation of parts uses a generative parametric probabilistic model, which is an extension of Weber et. al. [ECCV ’00] Model uses salient regions found by Kadir in addition to Brady’s interest operator PCA is used to give a compact representation of appearance as long as each region Learning Task: Estimation of model parameters Done using EM algorithm in an maximum-likelihood manner: E-step: Compute soft assignments as long as which P regions are as long as eground / background M-step: Update model parameters 200-400 images required to avoid overfitting 6 part model with 30 regions/image in addition to 400 training images takes around 24 hrs to run (~100 EM iterations) Foreground/background decision is automated Characteristics of region detector are learnt Generative model, so no background set used

Salient regions Find regions within image Use Kadir in addition to Brady’s salient region operator [IJCV ’01] Uses gray-scale input Finds maxima in entropy over scale in addition to location Motorbikes

Brock, Elizabeth Interim Station Manager

Brock, Elizabeth is from United States and they belong to WAPR-FM and they are from  Tuscaloosa, United States got related to this Particular Journal. and Brock, Elizabeth deal with the subjects like Music Programming; News Programming

Journal Ratings by Debre Birhan University

This Particular Journal got reviewed and rated by Debre Birhan University and short form of this particular Institution is ET and gave this Journal an Excellent Rating.