Deep learning-based image captioning

Abstract

A Common problem linking computer vision and natural language processing is the ability to generate an accurate caption for a given image. In this paper, various approaches of image captioning models towards achieving state of the art results are studied. After the various approaches are studied, the best approaches are then extracted and then recombined into a new single model in hopes to achieve a new state of the art model. A comparison of each model’s result will be used to determine the best performing model to be implemented. In this paper, we study the model of 2 different groups that created their image captioning model. They are namely the Google Brain team and the team that won the 2017 Visual Question Answering (VQA) Challenge in 2017.