Problem: Wide Variations In-plane rotationForeshortening
ScalingOut-of-plane rotation Intra-category variationAspect
ratio
Slide 10
Problem: Wide Variations In-plane rotationForeshortening
ScalingOut-of-plane rotation Intra-category variationAspect ratio
Nave brute-force evaluation is expensive
Slide 11
Our Method Mini-Parts Key idea: mini part model can approximate
deformations
Slide 12
Example: Arm Approximation
Slide 13
Example: Torso Approximation
Slide 14
Key Advantages Flexibility: General affine warps (orientation,
foreshortening, ) Speed: Mixtures of local templates + dynamic
programming
Slide 15
Pictorial Structure Model
Slide 16
Slide 17
Slide 18
Our Flexible Mixture Model
Slide 19
Slide 20
Slide 21
Slide 22
Co-occurrence Bias
Slide 23
Slide 24
Slide 25
Inference & Learning Inference
Slide 26
Inference & Learning Learning Inference
Slide 27
Benchmark Datasets PARSE Full-body
http://www.ics.uci.edu/~drama nan/papers/parse/index.html BUFFY
Upper-body http://www.robots.ox.ac.uk/~v
gg/data/stickmen/index.html
Slide 28
Solution: Cluster relative locations of joints w.r.t. parents
How to Get Part Mixtures?
Slide 29
Articulation
Slide 30
Slide 31
Slide 32
Qualitative Results
Slide 33
Quantitative Results on PARSE All previous work use explicitly
articulated models Image Parse Testset MethodTotal Ramanan 200727.2
Andrikluka 200955.2 Johnson 200956.4 Singh 201060.9 Johnson
201066.2 Our Model74.9 % of correctly localized limbs
Slide 34
Quantitative Results on PARSE 1 second per image Image Parse
Testset MethodHeadTorsoU. LegsL. LegsU. ArmsL. ArmsTotal Ramanan
200752.137.531.029.017.513.627.2 Andrikluka
200981.475.663.255.147.631.755.2 Johnson
200977.668.861.554.953.239.356.4 Singh
201091.276.671.564.950.034.260.9 Johnson
201085.476.173.465.464.746.966.2 Our
Model97.693.283.975.172.048.374.9 % of correctly localized
limbs
Slide 35
More Parts and Mixtures Help 14 parts (joints)27 parts (joints
+ midpoints)
Slide 36
Quantitative Results on BUFFY Our algorithm = 5 seconds -vs-
Next best = 5 minutes Subset of Buffy Testset MethodTotal Tran
201062.3 Andrikluka 200973.5 Eichner 200980.1 Sapp 2010a85.9 Sapp
2010b85.5 Our Model89.1 % of correctly localized limbs
Slide 37
Quantitative Results on BUFFY All previous work use explicitly
articulated models Subset of Buffy Testset MethodHeadTorsoU. ArmsL.
ArmsTotal Tran 2010 --- 62.3 Andrikluka 200990.795.579.341.273.5
Eichner 200998.797.982.859.880.1 Sapp 2010a100 91.165.785.9 Sapp
2010b10096.295.363.085.5 Our Model10099.696.670.989.1 % of
correctly localized limbs
Slide 38
Human Detection
Slide 39
Conclusion Model affine warps with a part-based model
Slide 40
Conclusion Model affine warps with a part-based model
Exponential set of pictorial structures
Slide 41
Conclusion Model affine warps with a part-based model
Exponential set of pictorial structures Flexible vs rigid
relations
Slide 42
Conclusion Model affine warps with a part-based model
Exponential set of pictorial structures Flexible vs rigid relations
Supervision helps