Object Detection Using a Max-Margin Hough Trans as long as m Overview Our Approach: Hough Trans as long as m Generalized to object detection

Object Detection Using a Max-Margin Hough Trans as long as m Overview Our Approach: Hough Trans as long as m Generalized to object detection www.phwiki.com

Object Detection Using a Max-Margin Hough Trans as long as m Overview Our Approach: Hough Trans as long as m Generalized to object detection

Davis, George, News Director has reference to this Academic Journal, PHwiki organized this Journal CVPR 2009, Miami, Florida Subhransu Maji in addition to Jitendra Malik University of Cali as long as nia at Berkeley, Berkeley, CA-94720 Object Detection Using a Max-Margin Hough Trans as long as m Overview Overview of probabilistic Hough trans as long as m Learning framework Experiments Summary Our Approach: Hough Trans as long as m Popular as long as detecting parameterized shapes Hough’59, Duda&Hart’72, Ballard’81, Local parts vote as long as object pose Complexity : parts votes Can be significantly lower than brute as long as ce search over pose ( as long as example sliding window detectors)

Hampton University US www.phwiki.com

This Particular University is Related to this Particular Journal

Generalized to object detection Learning Learn appearance codebook Cluster over interest points on training images Use Hough space voting to find objects Lowe’99, Leibe et.al.’04,’08, Opelt&Pinz’08 Implicit Shape Model Leibe et.al.’04,’08 Learn spatial distributions Match codebook to training images Record matching positions on object Centroid is given Detection Pipeline B. Leibe, A. Leonardis, in addition to B. Schiele. Combined object categorization in addition to segmentation with an implicit shape model ‘ 2004 Probabilistic Hough Trans as long as m C – Codebook f – features, l – locations

Learning Feature Weights Given : Appearance Codebook, C Posterior distribution of object center as long as each codeword P(x ) To Do : Learn codebook weights such that the Hough trans as long as m detector works well (i.e. better detection rates) Contributions : Show that these weights can be learned optimally using a max-margin framework. Demonstrate that this leads to improved accuracy on various datasets Naïve Bayes weights: Encourages relatively rare parts However rare parts may not be good predictors of the object location Need to jointly consider both priors in addition to distribution of location centers. Learning Feature Weights : First Try Location invariance assumption Overall score is linear given the matched codebook entries Position Posterior Codeword Match Codeword likelihood Learning Feature Weights : Second Try

Max-Margin Training Training: Construct dictionary Record codeword distributions on training examples Compute “a” vectors on positive in addition to negative training examples Learn codebook weights using by max-margin training Experiment Datasets ETHZ Shape Dataset (Ferrari et al., ECCV 2006) 255 images, over 5 classes (Apple logo, Bottle, Giraffe, Mug, Swan) UIUC Single Scale Cars Dataset (Agarwal & Roth, ECCV 2002) 1050 training, 170 test images INRIA Horse Dataset (Jurie & Ferrari) 170 positive + 170 negative images (50 + 50 as long as training) Experimental Results Hough trans as long as m details Interest points : Geometric Blur descriptors at sparse sample of edges (Berg&Malik’01) Codebook constructed using k-means Voting over position in addition to aspect ratio Search over scales Correct detections (PASCAL criterion)

Learned Weights (ETHZ shape) Learned Weights (UIUC cars) blue (low) , dark red (high) Learned Weights (INRIA horses) blue (low) , dark red (high)

Detection Results (ETHZ dataset) Recall @ 1.0 False Positives Per Window Detection Results (INRIA Horses) Detection Results (UIUC Cars) INRIA horses

Hough Voting + Verification Classifier Recall @ 0.3 False Positives Per Image Hough Voting + Verification Classifier IKSVM was run on top 30 windows + local search Hough Voting + Verification Classifier UIUC Single Scale Car Dataset IKSVM was run on top 10 windows + local search

Summary Hough trans as long as m based detectors offer good detection per as long as mance in addition to speed. To get better per as long as mance one may learn Discriminative dictionaries (two talks ago, Gall et.al.’09) Weights on codewords (our work) Our approach directly optimizes detection per as long as mance using a max-margin as long as mulation Any weak predictor of object center can be used is this framework Eg. Regions (one talk ago, Gu et.al. CVPR’09) Work partially supported by: ARO MURI W911NF-06-1-0076 in addition to ONR MURI N00014-06-1-0734 Computer Vision Group @ UC Berkeley Acknowledgements Thank You Questions Backup Slide : Toy Example

Davis, George KAFF-AM News Director www.phwiki.com

Davis, George News Director

Davis, George is from United States and they belong to KAFF-AM and they are from  Flagstaff, United States got related to this Particular Journal. and Davis, George deal with the subjects like International News; Local News; National News; Regional News

Journal Ratings by Hampton University

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