Li Fei-Fei, Princeton Rob Fergus, MIT Antonio Torralba, MIT CVPR 2007 Minneapolis, Short Course, June 17 Recognizing and L...
Agenda <ul><li>Introduction </li></ul><ul><li>Bag-of-words models </li></ul><ul><li>Part-based models </li></ul><ul><li>Di...
 
perceptible vision material thing
Plato said… <ul><li>Ordinary objects are classified together if they `participate' in the same abstract Form, such as the ...
Bruegel, 1564
How many object categories are there? ~10,000 to 30,000 Biederman 1987
So what does object recognition involve?
Verification: is that a lamp?
Detection: are there people?
Identification: is that Potala Palace?
Object categorization mountain building tree banner vendor people street lamp
Scene and context categorization <ul><li>outdoor </li></ul><ul><li>city </li></ul><ul><li>… </li></ul>
Computational photography
Assisted driving Lane detection Pedestrian and car detection <ul><li>Collision warning systems with adaptive cruise contro...
Improving online search Query: STREET Organizing photo collections
Challenges 1: view point variation Michelangelo 1475-1564
Challenges 2: illumination slide credit: S. Ullman
Challenges 3: occlusion Magritte, 1957
Challenges 4: scale
Challenges 5: deformation Xu, Beihong 1943
Challenges 6: background clutter Klimt, 1913
History: single object recognition
History: single object recognition <ul><li>Lowe, et al. 1999, 2003 </li></ul><ul><li>Mahamud and Herbert, 2000 </li></ul><...
Challenges 7: intra-class variation
History: early object categorization
<ul><li>Turk and Pentland, 1991 </li></ul><ul><li>Belhumeur, Hespanha, & Kriegman, 1997 </li></ul><ul><li>Schneiderman & K...
~10,000 to 30,000
Object categorization:  the statistical viewpoint vs. <ul><li>Bayes rule: </li></ul>posterior ratio likelihood ratio prior...
Object categorization:  the statistical viewpoint posterior ratio likelihood ratio prior ratio <ul><li>Discriminative meth...
Discriminative <ul><li>Direct modeling of  </li></ul>Zebra Non-zebra Decision boundary
Generative <ul><li>Model  and  </li></ul>Middle  Low High Middle Low
Three main issues <ul><li>Representation </li></ul><ul><ul><li>How to represent an object category </li></ul></ul><ul><li>...
Representation <ul><ul><li>Generative / discriminative / hybrid </li></ul></ul>
Representation <ul><ul><li>Generative / discriminative / hybrid </li></ul></ul><ul><ul><li>Appearance only or location and...
Representation <ul><ul><li>Generative / discriminative / hybrid </li></ul></ul><ul><ul><li>Appearance only or location and...
Representation <ul><ul><li>Generative / discriminative / hybrid </li></ul></ul><ul><ul><li>Appearance only or location and...
Representation <ul><ul><li>Generative / discriminative / hybrid </li></ul></ul><ul><ul><li>Appearance only or location and...
<ul><ul><li>Unclear how to model categories, so we learn what distinguishes them rather than manually specify the differen...
<ul><ul><li>Unclear how to model categories, so we learn what distinguishes them rather than manually specify the differen...
<ul><ul><li>Unclear how to model categories, so we learn what distinguishes them rather than manually specify the differen...
<ul><ul><li>Unclear how to model categories, so we learn what distinguishes them rather than manually specify the differen...
<ul><ul><li>Unclear how to model categories, so we learn what distinguishes them rather than manually specify the differen...
<ul><ul><li>Unclear how to model categories, so we learn what distinguishes them rather than manually specify the differen...
<ul><ul><li>Scale / orientation range to search over  </li></ul></ul><ul><ul><li>Speed </li></ul></ul><ul><ul><li>Context ...
Hoiem, Efros, Herbert, 2006
OBJECTS ANIMALS INANIMATE PLANTS MAN-MADE NATURAL VERTEBRATE … .. MAMMALS BIRDS GROUSE BOAR TAPIR CAMERA
Upcoming SlideShare
Loading in...5
×

Cvpr2007 object category recognition p0 - introduction

288

Published on

Published in: Education, Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
288
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
8
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Cvpr2007 object category recognition p0 - introduction

  1. 1. Li Fei-Fei, Princeton Rob Fergus, MIT Antonio Torralba, MIT CVPR 2007 Minneapolis, Short Course, June 17 Recognizing and Learning Object Categories: Year 2007
  2. 2. Agenda <ul><li>Introduction </li></ul><ul><li>Bag-of-words models </li></ul><ul><li>Part-based models </li></ul><ul><li>Discriminative methods </li></ul><ul><li>Segmentation and recognition </li></ul><ul><li>Datasets & Conclusions </li></ul>
  3. 4. perceptible vision material thing
  4. 5. Plato said… <ul><li>Ordinary objects are classified together if they `participate' in the same abstract Form, such as the Form of a Human or the Form of Quartz. </li></ul><ul><li>Forms are proper subjects of philosophical investigation, for they have the highest degree of reality. </li></ul><ul><li>Ordinary objects, such as humans, trees, and stones, have a lower degree of reality than the Forms. </li></ul><ul><li>Fictions, shadows, and the like have a still lower degree of reality than ordinary objects and so are not proper subjects of philosophical enquiry. </li></ul>
  5. 6. Bruegel, 1564
  6. 7. How many object categories are there? ~10,000 to 30,000 Biederman 1987
  7. 8. So what does object recognition involve?
  8. 9. Verification: is that a lamp?
  9. 10. Detection: are there people?
  10. 11. Identification: is that Potala Palace?
  11. 12. Object categorization mountain building tree banner vendor people street lamp
  12. 13. Scene and context categorization <ul><li>outdoor </li></ul><ul><li>city </li></ul><ul><li>… </li></ul>
  13. 14. Computational photography
  14. 15. Assisted driving Lane detection Pedestrian and car detection <ul><li>Collision warning systems with adaptive cruise control, </li></ul><ul><li>Lane departure warning systems, </li></ul><ul><li>Rear object detection systems, </li></ul>meters meters Ped Ped Car
  15. 16. Improving online search Query: STREET Organizing photo collections
  16. 17. Challenges 1: view point variation Michelangelo 1475-1564
  17. 18. Challenges 2: illumination slide credit: S. Ullman
  18. 19. Challenges 3: occlusion Magritte, 1957
  19. 20. Challenges 4: scale
  20. 21. Challenges 5: deformation Xu, Beihong 1943
  21. 22. Challenges 6: background clutter Klimt, 1913
  22. 23. History: single object recognition
  23. 24. History: single object recognition <ul><li>Lowe, et al. 1999, 2003 </li></ul><ul><li>Mahamud and Herbert, 2000 </li></ul><ul><li>Ferrari, Tuytelaars, and Van Gool, 2004 </li></ul><ul><li>Rothganger, Lazebnik, and Ponce, 2004 </li></ul><ul><li>Moreels and Perona, 2005 </li></ul><ul><li>… </li></ul>
  24. 25. Challenges 7: intra-class variation
  25. 26. History: early object categorization
  26. 27. <ul><li>Turk and Pentland, 1991 </li></ul><ul><li>Belhumeur, Hespanha, & Kriegman, 1997 </li></ul><ul><li>Schneiderman & Kanade 2004 </li></ul><ul><li>Viola and Jones, 2000 </li></ul><ul><li>Amit and Geman, 1999 </li></ul><ul><li>LeCun et al. 1998 </li></ul><ul><li>Belongie and Malik, 2002 </li></ul><ul><li>Schneiderman & Kanade, 2004 </li></ul><ul><li>Argawal and Roth, 2002 </li></ul><ul><li>Poggio et al. 1993 </li></ul>
  27. 28. ~10,000 to 30,000
  28. 29. Object categorization: the statistical viewpoint vs. <ul><li>Bayes rule: </li></ul>posterior ratio likelihood ratio prior ratio
  29. 30. Object categorization: the statistical viewpoint posterior ratio likelihood ratio prior ratio <ul><li>Discriminative methods model posterior </li></ul><ul><li>Generative methods model likelihood and prior </li></ul>
  30. 31. Discriminative <ul><li>Direct modeling of </li></ul>Zebra Non-zebra Decision boundary
  31. 32. Generative <ul><li>Model and </li></ul>Middle  Low High Middle Low
  32. 33. Three main issues <ul><li>Representation </li></ul><ul><ul><li>How to represent an object category </li></ul></ul><ul><li>Learning </li></ul><ul><ul><li>How to form the classifier, given training data </li></ul></ul><ul><li>Recognition </li></ul><ul><ul><li>How the classifier is to be used on novel data </li></ul></ul>
  33. 34. Representation <ul><ul><li>Generative / discriminative / hybrid </li></ul></ul>
  34. 35. Representation <ul><ul><li>Generative / discriminative / hybrid </li></ul></ul><ul><ul><li>Appearance only or location and appearance </li></ul></ul>
  35. 36. Representation <ul><ul><li>Generative / discriminative / hybrid </li></ul></ul><ul><ul><li>Appearance only or location and appearance </li></ul></ul><ul><ul><li>Invariances </li></ul></ul><ul><ul><ul><li>View point </li></ul></ul></ul><ul><ul><ul><li>Illumination </li></ul></ul></ul><ul><ul><ul><li>Occlusion </li></ul></ul></ul><ul><ul><ul><li>Scale </li></ul></ul></ul><ul><ul><ul><li>Deformation </li></ul></ul></ul><ul><ul><ul><li>Clutter </li></ul></ul></ul><ul><ul><ul><li>etc. </li></ul></ul></ul>
  36. 37. Representation <ul><ul><li>Generative / discriminative / hybrid </li></ul></ul><ul><ul><li>Appearance only or location and appearance </li></ul></ul><ul><ul><li>invariances </li></ul></ul><ul><ul><li>Part-based or global w/sub-window </li></ul></ul>
  37. 38. Representation <ul><ul><li>Generative / discriminative / hybrid </li></ul></ul><ul><ul><li>Appearance only or location and appearance </li></ul></ul><ul><ul><li>invariances </li></ul></ul><ul><ul><li>Parts or global w/sub-window </li></ul></ul><ul><ul><li>Use set of features or each pixel in image </li></ul></ul>
  38. 39. <ul><ul><li>Unclear how to model categories, so we learn what distinguishes them rather than manually specify the difference -- hence current interest in machine learning </li></ul></ul>Learning
  39. 40. <ul><ul><li>Unclear how to model categories, so we learn what distinguishes them rather than manually specify the difference -- hence current interest in machine learning) </li></ul></ul><ul><ul><li>Methods of training: generative vs. discriminative </li></ul></ul>Learning
  40. 41. <ul><ul><li>Unclear how to model categories, so we learn what distinguishes them rather than manually specify the difference -- hence current interest in machine learning) </li></ul></ul><ul><ul><li>What are you maximizing? Likelihood (Gen.) or performances on train/validation set (Disc.) </li></ul></ul><ul><ul><li>Level of supervision </li></ul></ul><ul><ul><ul><li>Manual segmentation; bounding box; image labels; noisy labels </li></ul></ul></ul>Learning Contains a motorbike
  41. 42. <ul><ul><li>Unclear how to model categories, so we learn what distinguishes them rather than manually specify the difference -- hence current interest in machine learning) </li></ul></ul><ul><ul><li>What are you maximizing? Likelihood (Gen.) or performances on train/validation set (Disc.) </li></ul></ul><ul><ul><li>Level of supervision </li></ul></ul><ul><ul><ul><li>Manual segmentation; bounding box; image labels; noisy labels </li></ul></ul></ul><ul><ul><li>Batch/incremental (on category and image level; user-feedback ) </li></ul></ul>Learning
  42. 43. <ul><ul><li>Unclear how to model categories, so we learn what distinguishes them rather than manually specify the difference -- hence current interest in machine learning) </li></ul></ul><ul><ul><li>What are you maximizing? Likelihood (Gen.) or performances on train/validation set (Disc.) </li></ul></ul><ul><ul><li>Level of supervision </li></ul></ul><ul><ul><ul><li>Manual segmentation; bounding box; image labels; noisy labels </li></ul></ul></ul><ul><ul><li>Batch/incremental (on category and image level; user-feedback ) </li></ul></ul><ul><ul><li>Training images: </li></ul></ul><ul><ul><ul><li>Issue of overfitting </li></ul></ul></ul><ul><ul><ul><li>Negative images for discriminative methods Priors </li></ul></ul></ul>Learning
  43. 44. <ul><ul><li>Unclear how to model categories, so we learn what distinguishes them rather than manually specify the difference -- hence current interest in machine learning) </li></ul></ul><ul><ul><li>What are you maximizing? Likelihood (Gen.) or performances on train/validation set (Disc.) </li></ul></ul><ul><ul><li>Level of supervision </li></ul></ul><ul><ul><ul><li>Manual segmentation; bounding box; image labels; noisy labels </li></ul></ul></ul><ul><ul><li>Batch/incremental (on category and image level; user-feedback ) </li></ul></ul><ul><ul><li>Training images: </li></ul></ul><ul><ul><ul><li>Issue of overfitting </li></ul></ul></ul><ul><ul><ul><li>Negative images for discriminative methods </li></ul></ul></ul><ul><ul><li>Priors </li></ul></ul>Learning
  44. 45. <ul><ul><li>Scale / orientation range to search over </li></ul></ul><ul><ul><li>Speed </li></ul></ul><ul><ul><li>Context </li></ul></ul>Recognition
  45. 46. Hoiem, Efros, Herbert, 2006
  46. 47. OBJECTS ANIMALS INANIMATE PLANTS MAN-MADE NATURAL VERTEBRATE … .. MAMMALS BIRDS GROUSE BOAR TAPIR CAMERA
  1. A particular slide catching your eye?

    Clipping is a handy way to collect important slides you want to go back to later.

×