Temporal Saliency Adaptation in Egocentric Videos.

This work adapts a deep neural model for image saliencyprediction to the temporal domain of egocentric video. We compute thesaliency map for each video frame, firstly with an off-the-shelf modeltrained from static images, secondly by adding a a convolutional orconv-LSTM layers trained with a dataset for video saliency prediction.We study each configuration on EgoMon, a new dataset made of sevenegocentric videos recorded by three subjects in both free-viewing andtask-driven set ups. Our results indicate that the temporal adaptation isbeneficial when the viewer is not moving and observing the scene froma narrow field of view. Encouraged by this observation, we compute andpublish the saliency maps for the EPIC Kitchens dataset, in which view-ers are cooking.