360RAT: A Tool for Annotating Regions of Interest in 360-degree Videos

Abstract

This paper introduces the software 360RAT as a tool for annotating regions of interest (RoIs) in 360-degree videos. These regions represent the portions of the video content that are important for telling a story throughout the video. We believe that this software is an invaluable tool for studying different aspects of 360-degree videos, including what viewers consider relevant and interesting to the experience. As part of this work, we conducted a subjective experiment in which 9 human observers used the proposed software to annotate 11 360-degree videos. As a result, we created a dataset containing a set of annotated 360-degree videos, i.e., videos with marked RoIs and their semantic classification. We present a simple analysis of the annotations gathered with the experiment for a subset of the videos. We noticed that there is a higher agreement of annotations among participants for videos with fewer objects. We also compared the RoI maps with saliency maps computed with the Cube Padding saliency model. We found a strong correlation between RoI maps and computed saliency models, indicating a link between the annotated RoI and the saliency properties of the content.