GLOMO: MATLAB Toolbox for learning layers from video

GLOMO (Greedy Learning Of Multiple Objects) is a MATLAB Toolbox for learning moving layers (i.e. 2D appearance models of objects or object's parts) from a video sequence. The theoretical foundations of the implemented algorithm can be found in Titsias and Williams (2004), which builds upon the greedy algorithm of Williams and Titsias (2004). A complete description of the algorithm can be also found in my PhD thesis (Titsias, 2005).


The software consists of a set of MATLAB functions that can be downloaded as a compressed tar-file; click here. The software is accompanied with demos (see below) and documentation.


There are 4 demos included within the software involving three video sequences that you have also to download and place each of them in appropriate directories:

Acknowledgements: We thank Nebojsa Jojic and Brendan Frey for poviding the Fey-Jojic and panaroma sequences.


M. K. Titsias. Unsupervised Learning of Multiple Objects in Images,
Ph.D. Thesis, School of Informatics, University of Edinburgh, June 2005. [pdf]

M. K. Titsias and C. K. I. Williams. Fast Unsupervised Greedy Learning of Multiple Objects and Parts from Video.
Generative-Model Based Vision Workshop, July 2004. [pdf]

C. K.I. Williams and M. K. Titsias. Greedy Learning of Multiple Objects in Images using Robust Statistics and Factorial Learning.
Neural Computation, 16(5), 1039-1062, May 2004. [pdf]

This page is maintained by:
Michalis K. Titsias