/reimaginet

Visually grounded word and sentence representations

Primary LanguageJupyter NotebookMIT LicenseMIT

reimaginet

Visually grounded word and sentence representations

Reimaginet is the new version of Imaginet, which was originally introduced in Chrupała et al 2015.

In this repository we are trying new models and features.

Pre-requisites

Reimaginet relies on Theano and funktional. For training models on the MS-COCO dataset we currently rely on data from Neuraltalk.