0 - Computer Vision Group

advertisement
CUMULATIVE ATTRIBUTE SPACE
FOR AGE AND CROWD DENSITY
ESTIMATION
K E C H E N 1, S H A O G A N G G O N G 1, T A O X I A N G 1, C H E N C H A N G E L O Y 2
1. QUEEN MARY, UNIVERSITY OF LONDON
2. THE CHINESE UNIVERSITY OF HONG KONG
CVPR 2013, Portland, Oregon
PROBLEMS
How old are they?
How many persons are in the scene?
What is the head pose (viewing angles) of this person?
A REGRESSION FORMULATION
Original images/frames
Feature space
Facial images
Label space
AAM feature
Feature extraction
Learning the mapping
Segment feature
Regression
Crowd frames
Edge feature
Texture feature
Labels
CHALLENGE – FEATURE VARIATION
The same age
Feature
 Extrinsic conditions: Lighting conditions;
Viewing angles
 Intrinsic conditions: aging process of different people
glasses, hairstyle, gender, ethnicity
CHALLENGE – FEATURE VARIATION
The same person count
Feature
 Extrinsic conditions: Lighting conditions;
Viewing angles
 Intrinsic conditions: occlusion, density distribution in the scene
CHALLENGE – SPARSE AND
IMBALANCED DATA
Data distribution of FG-NET Dataset
Max number of samples for each age group is 46
CHALLENGE – SPARSE AND
IMBALANCED DATA
Data distribution of UCSD Dataset
RELATED WORKS
• Most focused on feature variation challenge
1. Improve feature robustness
[Guo et al, CVPR, 2009; Guo et al, TIP, 2012; Ryan et al, DICTA, 2009; Zhang et al, IEEE T
ITS, 2011].
2. Improve regressor
[Guo et al, TIP 2008; Chang et al, CVPR 2011; Chao et al, PR 2013; Chan et al, CVPR
2008; Chen et al, BMVC 2012]
• Few focused on sparse and imbalanced data challenge
• Two challenges are related
OUR APPROACH
Solution:
• Attribute Learning can address data sparsity problem - Exploits the shared characteristics between classes
 Has sematic meaning
 Discriminative
Problems:
• Applied successfully in classification but not in regression
• How to exploit cumulative dependent nature of labels in
regression?
……
……
Age 20
Age 21
……
Age 60
CUMULATIVE ATTRIBUTE
Cumulative attribute
(dependent)
0
1
…
20
1
…
Age 20
1
0
Vs.
20th
1
0
0
0
0
0
…
…
the rest
Non-cumulative
attribute (independent)
0
LIMITATION OF NON-CUMULATIVE
ATTRIBUTE
00
1
00
0
1
0
0
1
0
0
…
0
0
21st
……
0
…
…
…
0
…
Age 20
00
…
20th
0
00
Age 60
21
60th
ADVANTAGES OF CUMULATIVE
ATTRIBUTE
1
1
1
1
0
the rest
0
…
0
1
0
1
0
0
…
…
0
1
…
…
Age 20
attributes
1 40
attribute
changes
change
…
…
20
1
0
21
60
60
Age 21
OUR FRAMEWORK
1
1
2
1
yi
1
…
yi+1
0
…
N
0
Cumulative
Attributes ai
Multi-output
Regression Learning
Facial images
Crowd frames
Feature Extraction
Imagery
Features xi
Regression Mapping
Labels yi
Regression Learning
Conventional frameworks
JOINT ATTRIBUTE LEARNING
• Joint Attribute Learning
1
2
𝑗
min
𝐰
+𝐶
2
2
N
𝑙𝑜𝑠𝑠(𝑎𝑖 𝑗 , 𝑓 𝑗 (𝐱 𝑖 )))
𝑖=1
with quadratic loss function
1
2
min
𝐖
+𝐶
2
𝐹
N
𝑖=1
𝐚𝑖 𝑇 − (𝐱 𝑖 𝑇 𝐖 + 𝐛)
2
𝐹
• Regression Learning
 with attribute representation as input
 is not limited to a specific regression model
COMPARATIVE EVALUATION
Age Estimation
CA-SVR: our method; AGES: Geng et al, TPAMI, 2007; RUN: Yan et al, ICCV, 2007; Ranking: Yan et
al, ICME, 2007; RED-SVM: Chang et al, ICPR, 2010; LARR: Guo et al, TIP, 2008; MTWGP: Zhang et al,
CVPR, 2010; OHRank: Chang et al, CVPR, 2011; SVR: Guo et al, TIP, 2008;
COMPARATIVE EVALUATION
Crowd Counting
CA-RR: our method; LSSVR: Suykens et al, IJCNN, 2001; KRR: An et al, CVPR, 2007; RFR: Liaw et al,
R News, 2002; GPR: Chan et al, CVPR, 2008; RR: Chen et al, BMVC, 2012;
CUMULATIVE (CA) VS. NONCUMULATIVE (NCA)
Age Estimation
Crowd Counting
ROBUSTNESS AGAINST SPARSE AND
IMBALANCED DATA
Age Estimation
Crowd Counting
FEATURE SELECTION BY ATTRIBUTES
Shape plays a more
important role than texture
when one is younger.
CONCLUSION
• A novel attribute framework for regression
• Exploits cumulative dependent nature of label space
• Effectively addresses sparse and imbalanced data
problem
Thanks a lot for your attention! Any questions?
Welcome to our poster 3A-2 for more details.
Ke Chen
Ph.D student
Shaogang Gong
Professor
Tao Xiang
Associate Professor
Chen Change Loy
Assistant Professor
Download