Beam Search for Solving QP

1
Scene Labeling Using Beam Search Under Mutex
Constraints Anirban Roy and Sinisa Todorovic

• Beam Search for Solving QP
• Results

• CRF Inference as QP
• Specifying Mutex Constraints

• Problem and Motivation
• Approach
• Extracting superpixels
• Incorporating mutex in the standard CRF
  formulation
• Formulating CRF inference as QP
• Beam search for solving QP
• Learning piecewise
• How to Specify CRF Energy?

State label assignment
Heuristic function
Semantic segmentation without Mutex
Semantic segmentation with Mutex
Input Image
Assignment vector
MUTual EXclusion (object, object, relationship)
next state
previous state
Superpixel
Class label
Score
maximum score
Matrix of CRF potentials
Pixelwise accuracy()
Method MSRC Test time
Galleguillos et al. CVPR 10 70.4 N/A
Gould et al. ICCV 09 76.4 N/A
Payet et al. PAMI 12 82.9 30-32s
Krahenbuhl et al. NIPS 12 86.0 0.2s
Yao et al. CVPR 12 86.5 N/A
Zhang et al. CVPR 12 87.0 N/A
Ours 91.5 0.8s
Method Stanford Background
Gould et al. ICCV 09 76.4
Munoz et al. ECCV 10 76.9
Singh et al. CVPR 13 74.1
Ours 81.1
8
must be
Mutex violations
Appearance features of the superpixels
Smoothness and Context
can be arbitrary
Matrix of mutex constraints
NSF RI 1302700