1. First Law Of Geography:
"Everything is related to everything else, but near
things are more related to each other.”
He’s at UCSB
-Waldo Tobler (1970)

2. Pyramids of Features For Categorization
Presented by Greg Griffin
Project Partner Will Coulter

3. Pyramids of Features For Categorization
Presented by Greg Griffin
Project Partner Will Coulter

4. Buckets of Features For Categorization
Presented by Greg Griffin
Project Partner Will Coulter

5. This talk is mostly about this paper:
With a little bit about benchmarking:
CALTECH 256

6. Images
Features
A Number
What is this number?

7. Images Features A Number “How well do they match”
Now get specific about features
“How well do
they match”

8. “Weak Features”
Show you an implm of their wf

9. “Weak Features” (I think?)
“…points whose gradient magnitude in a given
direction exceeds a minimum threshold.”
This is just their toy example
They use SIFT descriptors as “Strong Features”.
But you could use any features you want!

10. Images
Features
A Number
Number that indicates matching

12. Images
Features
X1 X5
Y1 Y5

13. Images
Features
X1 X5
The point of this exercise
Y1 Y5

14. Features
X1 X5
Y1 Y5

15. Features Start By Matching Reds
Progressively smaller cells, count matches

16. Features
Start By Matching Reds 1
l is… D is… I is…

17. Features
Start By Matching Reds 10 1 10 10

18. Features
Start By Matching Reds 1 10 4

19. Features
Start By Matching Reds 7 1 1 10 4 2 6 1 1 2

20. Features
Start By Matching Reds 7 1 1 10 4 8 2 6 1 1 2

21. Features
Start By Matching Reds 1 10 4 8 2 16

22. Features 5 2 1 2 1 3 2 1 1 1 1 Start By Matching Reds 1 10 4 8 2 16 61 10 4 8 2 16 6 2 1
Have table, throw out im,features 3 2 1 1 1 1

23. Start By Matching Reds 1 10 4 8 2 16 6

24. A compact set of matches is preferable to widely dispersed matches
Start By Matching Reds 1 10 4 8 2 16 6
A compact set of matches
is preferable to widely
dispersed matches

25. Start By Matching Reds 1 10 4 8 2 16 6

27. Start By Matching Reds 1 10 4 8 2 16 6
For noise…

28. Features X vs. Purely Isotropic Y1 10 4
5.5 2 16
1.9
A Sanity Check:
Features X vs. Purely Isotropic Y

29. Start By Matching Reds, Then The Blues, Then…1 10 4 8 2 16 6

30.
M = 8

31. Features 1 10

32. Features 1 10 4 8

33. Features 1 10 4 8 2 16 6

34. Foreach feature M=1…m
Foreach level = 0…L
Foreach cell i=1…D

35. Training Set Test Set SL(X,Y) 3.4 15 Categories 5.6 7.8 office 1.5
store
Disjoint!
5.4
coast
street
100 Images
per Category
Images
per Category
suburb

36. Confusion Matrix
Train on 100
Test on
(per category)

37. Scene Database
Caltech 101

38. Hypothesis Pyramid Matching works well when:
Objects are aligned and localized
ie. certain Caltech 101 categories
biased by different values of Ntest?
A few common features that define the category get randomly permuted through many positions, thanks to a large dataset
ie. scene database
now and then pyramid matching gets lucky
example: Library books

39. Test How well will Pyramid matching work?
Objects are not well aligned, or cluttered
Caltech 256 is more challenging in this respect
example: Library books

40. Scene Database
Caltech 101

41. Cluster Matching
example: Library books
k-means of SIFT positions

42. More Flexible Than A Grid?
k-means of SIFT positions + color

43. Position Invariance & Clutter
Grid can cut across ducks’ chest etc.
Alignment
Any cluster can match any cluster
Clusters respect duck / water boundaries
(sort of)

44. Tackles Alignment Problem
But… how to match efficiently?
How many clusters? How big?

45. Summary Spatial Pyramid Matching is efficient and
handles a range of scales, but seems to be
sensitive to translation and clutter.
Cluster Matching has the potential to
improve translational invariance and tolerance
of clutter. But inefficient. Less principled: how
many clusters are optimal? How big should they be?
No scale invariance.
Can we have the best of both worlds?

46. Try Sliding Spatial Pyramids?
Slide puzzle photo from: