This page is under construction. In the meantime you can:
Download the subspace clustering / segmentation MATLAB code that is presented in the paper
Vasileios Zografos, Liam Ellis, Rudolf Mester Discriminative Subspace Clustering. CVPR 2013
Get the code here
You will also need PRtools for the quadratic classifier (will remove this dependency in the near future).
Get a local copy of PRtools here
Instructions:
Unzip "DiSC.zip" a single folder. Unzip; "prtools.zip" anywhere you like and add a path to it in Matlab.
Then, given a data matrix W=[D,N] where D=number of ambient dimensions and N=number of points, simply run
[I,I_ALL,ProjectionMatrix]=DiSC(W,dim,final_clusters,'quadrc',lambda,Ensembles);
where:
Input
- dim = max dimension of the subspaces
- final_clusters = the number of clusters (i.e. subspaces)
- lambda = regularisation parameter based on noise. Try 0.01
- Ensembles = the number of classifiers for the ensemble. Try 50
Output
- I = the label vector of the clustering (that is the final answer)
- I_ALL= the label vector for each of the classifiers in the ensemble)
- ProjectionMatrix = the random projection matrix used for each ensemble classifier
Some results:
The MNIST dataset
The YaleB dataset (including background)
Synthetic subspace dataset (To be released soon)
Back to homepage