Object class recognition using unsupervised scaleinvariant learning Rob

Object class recognition using unsupervised scale-invariant learning Rob Fergus Pietro Perona Andrew Zisserman Oxford University California Institute of Technology

Overview Task: Recognition of object categories 3 main issues: Representation 2 Learning 3 Recognition 1

Some object categories Learn from just examples Difficulties: f f Size variation Background clutter Occlusion Intra-class variation

Model: constellation of Parts Fischler & Elschlager, 1973

Generative probabilistic model Foreground model Gaussian shape pdf Gaussian part appearance pdf Gaussian relative scale pdf Log(scale) Prob. of detection Clutter model Uniform shape pdf Gaussian appearance pdf Uniform relative scale pdf Log(scale) Poission pdf on # detections

Interest Operator �adir and Brady's interest operator. K �inds maxima in entropy over scale and location F

Representation of appearance 11 x 11 patch Normalize Projection onto PCA basis c 15

Experimental procedure Two series of experiments: Scale variant (using pre-scaled images) 2 Scale invariant 1 Datasets: � Motorbikes, Faces, Spotted cats, Airplanes, Cars from behind and side � Between 200 and 800 images in size Training Testing � 50% images � No identifcation of object within image � 50% images � Simple object present/absent test � ROC equal error rate computed, using background set of images

Motorbikes

Motorbikes

Motorbikes

Equal error rate: 7. 5% Motorbikes

Background images

Equal error rate: 4. 6% Frontal faces

Equal error rate: 9. 8% Airplanes

Scale-invariant Spotted cats Equal error rate: 10. 0%

Scale-invariant cars Equal error rate: 9. 7%

Robustness of algorithm

ROC equal error rates Pre-scaled data (identical settings): Scale-invariant learning and recognition:

Scale-invariant cars