Skip to content. | Skip to navigation

Personal tools
You are here: Home Data Sets Tools mgsm-1 (Mixture of Gaussian Scale Mixture 1). About mgsm-1

About mgsm-1

Information about the software.


This package contains some MatLab tools for building Mixture of Gaussian Scale Mixture (MGSM) models, and perform inference and learning. MGSM models, and their application to the statistics of natural images and contextual phenomena in the visual cortex, are describe in the associated paper:

Coen-Cagli, Dayan, Schwartz. "Cortical Surround Interactions and Perceptual Salience via Natural Scene Statistics".
PLoS Comp Biol 2012. DOI: 10.1371/journal.pcbi.1002405

Contents of the software

Briefly, the tools include:

  • Scripts that allow one to implement 'spatial' MGSM's, i.e. models of the statistical dependencies between the outputs of visual filters applied to neighboring regions of images. In particular, full implementations are provided for two specific cases; other cases can be easily added.
  • Scripts that allow one to perform inference and learning, given a dataset. The training code is rather general, and directly applicable to any MGSM (ie beyond the two cases for which full implementation is provided). The only exception is that learning involves a constrained numerical optimization, and the nonlinear constraints need to be defined on a case by case basis.
  • Scripts to collect data, comprising the filters' outputs across many natural image patches.
  • Five example images from the 'standard' dataset.

How to download

The software may be downloaded from:
A account is required. See the download link for more instructions.  The software is in file "".

How to cite the software

Publications created through usage of the software should cite the publication referenced above and also cite this resource at in the following recommended format:

Ruben Coen-Cagli, Peter Dayan, Odelia Schwartz (2016). MatLab tools for building Mixture of Gaussian Scale Mixture (MGSM) models, and perform inference and learning.

The above citation uses a Digital Object Identifier (DOI) which is assigned to the data set.  The DOI was created using DataCite ( and the California Digital Library, "EZID" system (

Getting help.

If you have questions about the software, post them on the forum for using data sets.

Associated file

crcns_mgsm-1_descrption.pdf (Description of software).

Document Actions