Thomas Serre

Home

Research

Publications

Resources

Curriculum vitae

Collaborators

SUnS symposium

Teaching

In the news

 

Last update: Wed 06-May-2009

Resources
Source code
Model ventral stream

[Biologically motivated framework for object recognition] We have developped a computational theory of object recognition in the cortex. We have extensively compared the tuning properties of the units in the model to those of cortical cells. The model is qualitatively and quantitatively consistent with several properties of subpopulations of cells in V1, V4, IT, and PFC as well as fMRI and psychophysical data.
The model has evolved over the past few years and we have made several software implementations available.
>> Download web page.

[Biologically motivated framework for action recognition] We have recently extended our work on object recognition to the recognition of actions with a model of the dorsal based on a dictionary of position and scale invariant spatio-temporal motion features.
>> Source code.

Supplementary web material
Serre et al, PNAS 2007

[T. Serre, A. Oliva and T. Poggio. A feedforward architecture accounts for rapid categorization. Proceedings of the National Academy of Science, 104(15), pp. 6424-6429, April 2007 open access ] The supplementary web material accompanies the study and includes, in particular, a basic software implementation of the computational model and the animal vs. non-animal stimulus database (along with the performance of benchmark systems). It also includes a summary of various performance measures for both human observers and the model (including ROC analysis, error and hit rates) as well as reaction times for human observers.

Street-scene

[T. Serre, L. Wolf, S. Bileschi, M. Riesenhuber and T. Poggio. Object recognition with ortex-like mechanisms. In: IEEE Transactions on Pattern Analysis and Machine Intelligence, 29 (3), pp. 411-426 , 2007] Here is a link to the software implementation to the model of the ventral stream of the visual cortex used in this study. We used the Street-scene database collected by Stan Bileschi (CBCL, MIT) to benchmark the system. We also used some the CalTech-6 and CalTech-101 datasets.
In this study we used osusvm, the matlab interface for libsvm. Here is a list of SVM software implementations and general machine learning toolboxes that are publicly available.

SVM software implementations and general machine learning toolboxes

[General machine learning toolboxes] [matlab] Spider <Max Planck Institute for Biol. Cybernetics> / STPRtool toolbox <Czech Technical University in Prague> / Regularization tools <Technical University of Denmark>

[SVM software][C] libsvm / svmlight / svmFu (no longer supported)

[SVM software][matlab interfaces] osusv (libsvm) / Tom Briggs (svmlight) / Anton Schwaighofer (svmlight)

Image datasets

CBCL street scene database (created by Stan Bileschi)

Other CBCL image datasets

Misc image databases: