This is the code for the paper:
"Normalized Cut Meets MRF"
Meng Tang, Dmitrii Marin, Ismail Ben Ayed, Yuri Boykov
In European Conference on Computer Vision (ECCV), Amsterdam, the Netherlands, October, 2016
The CORE of our algorithm is linearization or unary bound for Normalized Cut (NC).
Simple implementation of such linearization is given in a FEW lines in "matlab/KernelBound.m" and "matlab/SpectralBound.m".
[ unaries ] = KernelBound( A, K, current_clustering)
% KERNELBOUND simply takes affinity, cluster number and current clustering and gives unary terms.
[ unaries ] = SpectralBound( A, K, dim, current_clustering)
% SPECTRALBOUND requires an extra argument dim, the dimensionality of spectral embedding
% When dim is the number of data points, SpectralBound is algorithmically EQUIVALENT to KernelCut.Example of optimizing NC or AA (avearge association) ONLY is in "matlab/syntheticclustering.m". Below is sample result with KernelCut for NC:
Multilabel example: Input image frames: directory "motionsegmentation/ducks01/images"
Initial Strokes for the first frame: directory "motionsegmentation/ducks01/seedsmulti"
Binary example: Input image frames: directory "motionsegmentation/horses01/images"
Initial Strokes for the first frame: directory "motionsegmentation/horses01/seeds"
Build dependency libraries (maxflow and easybmp)
$ cd libs
$ make all
Build main program
$ cd ../kernelcut
$ make main
$ cd ../
Download executable for optical flow
$ wget http://lmb.informatik.uni-freiburg.de/resources/binaries/pami2010Linux64.zip
$ unzip pami2010Linux64.zip -d libs/LDOF
Compute optical flow, flow and its visualization saved into opticalflow directory
$ chmod +x libs/LDOF/ldof
$ matlab -nojvm -nosplash -nodisplay -r "cd motionsegmentation/ducks01; computeopticalflow; exit()"
Compute KNN graph for joint LAB + XY + M space
$ matlab -nojvm -nosplash -nodisplay -r "cd motionsegmentation/ducks01; getsubpixelimages; exit();"
$ matlab -nojvm -nosplash -nodisplay -r "cd motionsegmentation/ducks01; computeknn; exit()"
(Visualization of KNN graph is by clicking on image pixel, simply run motionsegmentation/visualizeknnbyclick.m)
Go to motionsegmentation/motion.sh, change codepath, and run script
$ chmod +x ./motionsegmentation/motion_ducks01.sh
$ ./motionsegmentation/motion_ducks01.sh
Output segmentations are in the directory "motionsegmentation/ducks01/output".
(note that if initialized from seeds, the colors has to be of the following: {white,red,blue,green,black,navy})
Binary example "horses01" is similar to multilabel example "ducks01".
Python implementation of KernelCut for binary segmentation is provided. See directory 'PyKernelCut'. Example Usage:
$ cd PyKernelCut
$ python segmentation.py -h
$ python segmentation.py -i 124084.jpg -b 10 20 300 300 -k 100
$ python segmentation.py -i 0_5_5303.bmp -b 50 100 200 200 -s 0
$ python segmentation.py -i 0_5_5303.bmp -b 50 100 200 200
$ python segmentation.py -i 314016.jpg -b 10 80 300 300 --hard
Note that To use PyKernelCut several dependencies (skimage, scipy, PyMaxflow) have to be installed.
See LICENSE.