Section: Software and Platforms

Large-scale image classification

Participants : Matthijs Douze [correspondant] , Zaid Harchaoui, Florent Perronnin [XRCE] , Cordelia Schmid.

JSGD is the implementation of a Stochastic Gradient Descent algorithm used to train linear multiclass classifiers. It is biased towards large classification problems (many classes, many examples, high-dimensional data). It can be used on the ImageNet large scale classification challenge. It uses several optimization techniques, both algorithmic (scale factors to spare vector multiplications, vector compression with product quantizers) and technical (vector operations, multithreading, improved cache locality). It has Python and Matlab interfaces. It is distributed under a Cecill licence. Project page: http://lear.inrialpes.fr/src/jsgd .