A compact deep architecture for real-time saliency prediction

作者：

Highlights：

• Here we propose a compact yet fast model for real-time saliency prediction.

• We employ a modified U-net architecture to reduce the number of model parameters.

• We employ a location-dependent layer to capture location-dependent information.

• We compare the structural capabilities of saliency models under unified assumptions.

摘要

•Here we propose a compact yet fast model for real-time saliency prediction.•We employ a modified U-net architecture to reduce the number of model parameters.•We employ a location-dependent layer to capture location-dependent information.•We compare the structural capabilities of saliency models under unified assumptions.

论文关键词：Fast saliency prediction,Deep convolutional neural network,Transfer learning,Compact architecture,Real-time application

论文评审过程：Received 30 April 2021, Revised 25 January 2022, Accepted 14 February 2022, Available online 22 February 2022, Version of Record 5 March 2022.

论文官网地址：https://doi.org/10.1016/j.image.2022.116671