classification using intersection kernel svms is efficient joint work with subhransu maji and alex...
TRANSCRIPT
![Page 1: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/1.jpg)
Classification using intersection kernel SVMs is
efficient
Joint work with Subhransu Maji and Alex Berg
Jitendra Malik UC Berkeley
![Page 2: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/2.jpg)
Fast intersection kernel SVMs
and other generalizations of linear SVMs
- IKSVM is a (simple) generalization of a linear SVM- Can be evaluated very efficiently- Other kernels (including ) have a similar form- Novel features based on pyramid of oriented energy.- Methods applicable to current most successful object recognition/detection strategies.
![Page 3: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/3.jpg)
Detection: Is this an X?
Ask this question over and over again,varying position, scale, multiple categories…Speedups: hierarchical, early reject, feature sharing, cueingbut same underlying question!
![Page 4: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/4.jpg)
Detection: Is this an X?
Ask this question over and over again,varying position, scale, multiple categories…Speedups: hierarchical, early reject, feature sharing,but same underlying question!
![Page 5: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/5.jpg)
Detection: Is this an X?
Ask this question over and over again,varying position, scale, multiple categories…Speedups: hierarchical, early reject, feature sharing,but same underlying question!
Boosted dec. trees, cascades + Very fast evaluation - Slow training (esp. multi-class)Linear SVM + Fast evaluation + Fast training - Need to find good featuresNon-linear kernelized SVM + Better class. acc. than linear . Medium training - Slow evaluation
![Page 6: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/6.jpg)
Detection: Is this an X?
Ask this question over and over again,varying position, scale, multiple categories…Speedups: hierarchical, early reject, feature sharing,but same underlying question!
Boosted dec. trees, cascades + Very fast evaluation - Slow training (esp. multi-class)Linear SVM + Fast evaluation + Fast training - Need to find good featuresNon-linear kernelized SVM + Better class. acc. than linear . Medium training - Slow evaluation
This work
![Page 7: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/7.jpg)
Outline
What is Intersection Kernel SVM? Trick to make it fast (exact) Trick to make it very fast (approximate) Why use it? Multi-scale Features based on Oriented Energy
Generalization of linear classifiers Reinterpret the approximate IKSVM Fast training
Summary of where this matters
![Page 8: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/8.jpg)
Outline
What is Intersection Kernel SVM? Trick to make it fast (exact) Trick to make it very fast (approximate) Why use it? Multi-scale Features based on Oriented Energy
Generalization of linear classifiers Reinterpret the approximate IKSVM Fast training
Summary of where this matters
![Page 9: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/9.jpg)
Support Vector Machines
Linear Separators (aka. Perceptrons)
B2
![Page 10: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/10.jpg)
Support Vector Machines
Other possible solutions
B2
![Page 11: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/11.jpg)
Support Vector Machines
Which one is better? B1 or B2? How do you define better?
B1
B2
![Page 12: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/12.jpg)
Support Vector Machines
Find hyperplane maximizes the margin => B1 is better than B2
B1
B2
b11
b12
b21
b22
margin
![Page 13: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/13.jpg)
Kernel Support Vector Machines
Kernel :•Inner Product in Hilbert Space
•Can Learn Non Linear Boundaries
2
2( , ) exp( )
2
x zK x z
σ−
= −
( , ) ( ) ( )TK x z x z=Φ Φ
![Page 14: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/14.jpg)
Feature Representation
Discriminative Classifier
(+ examples)
(- examples)
Training Stage
![Page 15: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/15.jpg)
Our Multiscale HOG-like feature
Concatenate orientation histograms for each orange region.Differences from HOG: -- Hierarchy of regions -- Only performing L1 normalization once (at 16x16)
![Page 16: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/16.jpg)
Comparison to HOG (Dalal & Triggs)
![Page 17: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/17.jpg)
Smaller Dimensional (1360 vs. 3780)
Simple Implementation (Convolutions)
Faster to compute + No non-local Normalization+ No gaussian weighting+ No color normalization
Comparison to HOG (Dalal & Triggs)
![Page 18: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/18.jpg)
What is the Intersection Kernel?
Histogram Intersection kernel between histograms a, b
![Page 19: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/19.jpg)
What is the Intersection Kernel?
Histogram Intersection kernel between histograms a, b
K small -> a, b are differentK large -> a, b are similar
Intro. by Swain and Ballard 1991 to compare color histograms.Odone et al 2005 proved positive definiteness.Can be used directly as a kernel for an SVM.Compare to
![Page 20: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/20.jpg)
linear SVM, Kernelized SVM, IKSVM
Decision function is where:
Linear:
Non-linearUsingKernel
HistogramIntersectionKernel
![Page 21: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/21.jpg)
Kernelized SVMs slow to evaluate
Arbitrary Kernel
HistogramIntersectionKernel
Feature corresponding to a support vector l
Feature vector to evaluate
Kernel EvaluationSum over all support vectors
SVM with Kernel Cost: # Support Vectors x Cost of kernel comp.IKSVM Cost: # Support Vectors x # feature dimensions
Decision function is where:
![Page 22: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/22.jpg)
The TrickDecision function is where:
Just sort the support vectorvalues in each coordinate, andpre-compute
To evaluate, find position ofin the sorted support vectorvalues (cost: log #sv)look up values, multiply & add
![Page 23: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/23.jpg)
The TrickDecision function is where:
Just sort the support vectorvalues in each coordinate, andpre-compute
To evaluate, find position ofin the sorted support vectorvalues (cost: log #sv)look up values, multiply & add
#support vectors x #dimensions
log( #support vectors ) x #dimensions
![Page 24: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/24.jpg)
The Trick 2
For IK hi is piecewise linear, and quite smooth, blue plot. We can approximate with fewer uniformly spaced segments, red plot. Saves
time & space!
Decision function is where:
#support vectors x #dimensionslog( #support vectors ) x #dimensions
![Page 25: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/25.jpg)
The Trick 2Decision function is where:
#support vectors x #dimensionslog( #support vectors ) x #dimensions
constant x #dimensions
For IK hi is piecewise linear, and quite smooth, blue plot. We can approximate with fewer uniformly spaced segments, red plot. Saves
time & space!
![Page 26: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/26.jpg)
Timing Results
Time to evaluate 10,000 feature vectors
IKSVM with our multi-scale version of HOG featuresbeats Dalal & Triggs. Alsofor Daimler Chrysler data. Current Best on these datasets.
Linear SVM with our multi-scale Version of HOG featureshas worse classification perf.than Dalal & Triggs.
reduced memory!
![Page 27: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/27.jpg)
Distribution of support vector values and hi
Distribution of
![Page 28: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/28.jpg)
Best Performance on Pedestrian Detection,Improve on Linear for Many Tasks
INRIA PedestriansDaimler Chrysler Pedestrians
Caltech 101 with “simple features” Linear SVM 40% correct IKSVM 52% correct
![Page 29: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/29.jpg)
Classification Errors
![Page 30: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/30.jpg)
Results – ETHZ DatasetDataset: Ferrari et al., ECCV 2006 255 images, over 5 classes training = half of positive images for a class + same number from the other classes (1/4 from each) testing = all other images large scale changes; extensive clutter
![Page 31: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/31.jpg)
Method Applelogo
Bottle
Giraffe
Mug Swan Avg
PAS* 65.0 89.3
72.3 80.6 64.7 76.7
Our 86.1 81.0
62.1 78.0 100 81.4
Beats many current techniques without any changes to our features/classification framework.
Recall at 0.3 False Positive per Image Shape is an important cue (use Pb instead of OE)
Results – ETHZ Dataset
*Ferarri et.al, IEEE PAMI - 08
![Page 32: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/32.jpg)
Other kernels allow similar trick
Decision function is where:
IKSVM SVM
hi not piece-wise linear,but we can still use anapproximation for fastevaluation.
hi are piece-wise linear,uniformly spacedpiece-wise linear approx.is fast.
![Page 33: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/33.jpg)
Results outside computer vision
Accuracy of IK vs Linear on Text classification
Error rate of directly trained piecewise linear (blue) best kernel (green) and linear (red) on SVM benchmark datasets
![Page 34: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/34.jpg)
Results outside computer vision
Accuracy of IK vs Linear on Text classification
Error rate of directly trained piecewise linear (blue) best kernel (green) and linear (red) on SVM benchmark datasets
![Page 35: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/35.jpg)
Results outside computer vision
Accuracy of IK vs Linear on Text classification
Error rate of directly trained piecewise linear (blue) best kernel (green) and linear (red) on SVM benchmark datasets
Piecewise linear usually better,Depending on amount of data relative to the dimension and regularization
![Page 36: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/36.jpg)
Conclusions Exactly evaluate IKSVM in O(n log m) as opposed to O(nm) Makes SV cascade or other ordering schemes irrelevant for intersection kernel
Verified that IKSVM offers classification performance advantages over linear
Approximate decision functions that decompose to a sum of functions for each coordinate (including Chi squared)
Directly learn such classification functions (no SVM machinery)
Generalized linear svm beats linear SVM in some applications often as good as more expensive RBF kernels
Showed that relatively simple features with IKSVM beats Dalal & Triggs (linear SVM), leading to the state of the art in pedestrian detection.
Applies to best Caltech 256, Pascal VOC 2007 methods.
![Page 37: Classification using intersection kernel SVMs is efficient Joint work with Subhransu Maji and Alex Berg Jitendra Malik UC Berkeley](https://reader036.vdocument.in/reader036/viewer/2022062712/56649cae5503460f94972241/html5/thumbnails/37.jpg)
Classification Using Intersection Kernel Support Vector Machines is efficient.Subhransu Maji and Alexander C. Berg and Jitendra Malik.Proceedings of CVPR 2008, Anchorage, Alaska, June 2008.
Software and more results available at
http://www.cs.berkeley.edu/~smaji/projects/fiksvm/