You are here

Visual Attention and Language Constructs: analysis of bottom-up saliency in image captions by human and machine

Friday, 22. December 2017 - 11:00
Dr. Hamed Tavakoli
IRISA Rennes, Salle Lipari (Orange)


This talk is about how well human and machine agree on image captions with respect to bottom-up visual attention and saliency. Two questions will be discussed, (1) How well do image descriptions, by humans or models, on a scene agree with saliency?, (2) Can saliency benefit image captioning by machine? On the first part, we will analyse the data with respect to attention. On the second part, we will go through the steps for building a model to validate saliency contribution. The emphasis of the talk would be on bottom-up visual attention.


Dr. Hamed Tavakoli is  postdoctoral Researcher at The Finnish Center of Excellence in Computational Inference Research, Aalto University. His Google scholar account is: