Author: Stanislaw Kajetan Morawski
Supervisors: Dimosthenis Karatzas & Ali Biten
Presentation time: 9:15
Virtual Room: 5.2 | Live presentation URL
Most of the deep learning attempts at captioning result in factual, neutral and boring captions. This may be valuable in certain scenarios (like historical archives), but in the world of entertainment is a big limitation. In this work we have explored image captioning in poetic and dignified English, mimicking the written language of William Shakespeare. Using SemStyle as a baseline we have tried to improve upon the framework using a different style. Also we have trained the model to mimic the modern equivalent of Shakespearean style. Furthermore we have examined the differences in performance of our model when built on top of different CNN feature extractors.