1 / 22

Object class recognition using unsupervised scale-invariant learning

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 Learning Recognition.

tmccool
Download Presentation

Object class recognition using unsupervised scale-invariant learning

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

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

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

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

  4. Model: constellation of Parts • 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 Fischler & Elschlager, 1973

  5. Generative probabilistic model 0.8 0.75 0.9 Foreground model p(A)=PpartsG(Apart|cpart,Vpart) Gaussian shape pdf Gaussian part appearance pdf Gaussian relative scale pdf Log(scale) Prob. of detection Clutter model Clutter pdf Uniform relative scale pdf Gaussian appearance pdf Uniform shape pdf Ppoisson(N|l) Log(scale) p(x)=A-1(uniform) Poission pdf on # detections

  6. Interest Operator • Kadir and Brady's interest operator. • Finds maxima in entropy over scale and location

  7. Representation of appearance c1 c2 Projection onto PCA basis 11x11 patch Normalize c15

  8. Experimental procedure Two series of experiments: Scale variant (using pre-scaled images) Scale invariant • 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 • Testing • 50% images • Simple object present/absent test • ROC equal error rate computed, using background set of images

  9. Motorbikes

  10. Motorbikes

  11. Motorbikes

  12. Motorbikes Equal error rate: 7.5%

  13. Background images

  14. Frontal faces Equal error rate: 4.6%

  15. Equal error rate: 9.8% Airplanes

  16. Scale-invariant Spotted cats Equal error rate: 10.0%

  17. Scale-invariant cars Equal error rate: 9.7%

  18. Robustness of algorithm

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

  20. Scale-invariant cars

  21. Discussion Questions • What prior work does this paper build on (besides Kadir’s) ? • What was significant about the training for object vs. background? • What are the main contributions of the Fergus work? • What are the drawbacks to this approach?

More Related