Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew...
-
date post
21-Dec-2015 -
Category
Documents
-
view
219 -
download
1
Transcript of Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew...
![Page 1: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/1.jpg)
Object class recognition using unsupervised scale-invariant
learning
Rob FergusPietro Perona
Andrew Zisserman
Oxford UniversityCalifornia Institute of Technology
![Page 2: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/2.jpg)
1 Representation 2 Learning3 Recognition
OverviewTask: Recognition of object categories
3 main issues:
![Page 3: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/3.jpg)
Some object categories
Learn from just examples
Difficulties:
Size variation Background clutter Occlusion Intra-class variation
![Page 4: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/4.jpg)
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
![Page 5: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/5.jpg)
Foreground model
Gaussian shape pdf
Poission pdf on # detections
Uniform shape pdf
Prob. of detection
Gaussian part appearance pdf
Generative probabilistic model
Clutter model
Gaussian relative scale pdf
Log(scale)
Gaussian appearance pdf
0.8 0.75 0.9
Uniformrelative scale pdf
Log(scale)
![Page 6: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/6.jpg)
Interest OperatorKadir and Brady's interest operator. Finds maxima in entropy over scale and location
![Page 7: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/7.jpg)
Representation of appearance
11x11 patch
c1
c2
Normalize
Projection ontoPCA basis
c15
![Page 8: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/8.jpg)
Experimental procedureTwo series of experiments:
Datasets: Motorbikes, Faces, Spotted cats, Airplanes, Cars from behind and side Between 200 and 800 images in size
Training 50% images No identifcation of object within image
1 Scale variant (using pre-scaled images)2 Scale invariant
Testing 50% images Simple object present/absent test ROC equal error rate computed, using background set of images
![Page 9: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/9.jpg)
Motorbikes
![Page 10: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/10.jpg)
Motorbikes
![Page 11: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/11.jpg)
Motorbikes
![Page 12: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/12.jpg)
MotorbikesEqual error rate: 7.5%
![Page 13: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/13.jpg)
Background images
![Page 14: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/14.jpg)
Frontal facesEqual error rate: 4.6%
![Page 15: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/15.jpg)
AirplanesEqual error rate: 9.8%
![Page 16: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/16.jpg)
Scale-invariant Spotted catsEqual error rate: 10.0%
![Page 17: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/17.jpg)
Scale-invariant carsEqual error rate: 9.7%
![Page 18: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/18.jpg)
Robustness of algorithm
![Page 19: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/19.jpg)
ROC equal error rates
Pre-scaled data (identical settings):
Scale-invariant learning and recognition:
![Page 20: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/20.jpg)
![Page 21: Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of.](https://reader036.fdocuments.in/reader036/viewer/2022062320/56649d565503460f94a3482d/html5/thumbnails/21.jpg)
Scale-invariant cars