Video Saliency Detection using Deep Learning

University essay from KTH/Skolan för elektroteknik och datavetenskap (EECS)

Author: Jakob Wiesinger; [2019]

Keywords: ;

Abstract: A deep learning model for video saliency detection is proposed and trained. The neural network architecture combines recent innovations in the field: A twostream approach merges two separate input streams for appearance and motion aspects of saliency. Pre-trained convolutional features detect objectness. Attention modules are employed to efficiently reweight features. A ConvLSTM module ensures temporal consistency. Training data comprises both videos and images with corresponding gaze fixation locations from eye trackers. The model is evaluated and shown to perform on par with the state of the art.

  AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)