Deep Learning for Automatic Image Captioning in Poor Training Conditions

Abstract

Recent advancements in Deep Learning have proved that an architecture that combines Convolutional Neural Networks and Recurrent Neural Networks enables the definition of very effective methods for the automatic captioning of images. The disadvantage that comes with this straightforward result is that this approach requires the existence of large-scale corpora, which are not available for many languages.This paper introduces a simple methodology to automatically acquire a large-scale corpus of 600 thousand image/sentences pairs in Italian. At the best of our knowledge, this corpus has been used to train one of the first neural captioning systems for the same language. The experimental evaluation over a subset of validated image/captions pairs suggests that the achieved results are comparable with the English counterpart, despite a reduced amount of training examples.