Temporal recurrences for video saliency prediction

  • 2019-07-11 16:03:44
  • Panagiotis Linardos, Eva Mohedano, Juan Jose Nieto, Noel E. O'Connor, Xavier Giro-i-Nieto, Kevin McGuinness
  • 0

Abstract

This paper investigates modifying an existing neural network architecture forstatic saliency prediction using two types of recurrences that integrateinformation from the temporal domain. The first modification is the addition ofa ConvLSTM within the architecture, while the second is a conceptually simpleexponential moving average of an internal convolutional state. We use weightspre-trained on the SALICON dataset and fine-tune our model on DHF1K. Ourresults show that both modifications achieve state-of-the-art results andproduce similar saliency maps. Source code is available athttps://git.io/fjPiB.

 

Quick Read (beta)

This feature is not avaialbe for this paper.