Mean-Field Theory and Its Applications In Computer Vision6 1
Mar 28, 2015
Mean-Field Theory and Its Applications In Computer Vision6
1
Inference In Product Label Space
• Many problem requires jointly estimating labels in product label space
2
Black Box
SolverLeft Camera Image
Right Camera Image
Object Class Segmentation
Dense Stereo Reconstruction
Joint Object-Stereo Labelling• Computation complexity very high
• Graph-cuts based method takes almost 50 secs for 320x200 image size
• We propose mean-field based inference method
• Our method takes 2 secs for the same task
3
Joint stereo-object inference• Introduce two different set of variables
4
disparity variable
object variable
Messages exchanged between the variables
Joint stereo-object formulation
5
Unary Potential
• Weighted sum of object class, depth and joint potential
• Joint unary potential based on histograms of height
Joint stereo-object formulation
6
Pairwise Potential
• Object class and depth edges correlated
• We disregard the joint pairwise terms though
• Dense pairwise connection at both disparity variable and object variables
Joint stereo-object formulation
7
Higher Order Potential
• Use higher order terms only for object variables
Joint stereo-object updation
8
For object variables
Message from disparity variables to object variables
Joint stereo-object updation
9
For object variables
Filtering is done using permutohedral lattice based filtering strategy
Joint stereo-object updation
10
For disparity variables
Message from object variables to disparity variables
Joint stereo-object updation
11
For disparity variables
Filtering is done using domain transform based filtering strategy
Leuven dataset
12
Some of qualitative results
Leuven dataset
13Some of qualitative results
Algorithm Time (s) Object (% correct) Stereo (% correct)
GC + Range (1) 24.6 95.94 76.97
GC + Range (2) 49.9 95.94 77.31
GC + Range (3) 74.4 95.94 77.46
Extended CostVol 4.2 95.20 77.18
Dense + HO(PLBF)
3.1 95.24 78.89
Dense + HO(DTBF)
2.1 95.06 78.21
Dense + HO + CostVol + DTBF
6.3 94.98 79.00