SaltiNet: Scan-path Prediction on 360 Degree Images using Saliency Volumes

被引：90

作者：

Assens, Marc ^{[1
,3
]}

Giro-i-Nieto, Xavier ^{[1
]}

McGuinness, Kevin ^{[2
]}

O'Connor, Noel E. ^{[2
]}

机构：

[1] UPC, Image Proc Grp, Barcelona, Catalonia, Spain

[2] Dublin City Univ, Insight Ctr Data Analyt, Dublin, Ireland

[3] Insight Ctr Data Analyt, Dublin, Ireland

来源：

2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION WORKSHOPS (ICCVW 2017) | 2017年

基金：

爱尔兰科学基金会;

关键词：

ATTENTION; MODEL;

D O I：

10.1109/ICCVW.2017.275

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

We introduce SaltiNet, a deep neural network for scan-path prediction trained on 360-degree images. The model is based on a temporal-aware novel representation of saliency information named the saliency volume. The first part of the network consists of a model trained to generate saliency volumes, whose parameters are fit by back-propagation computed from a binary cross entropy (BCE) loss over downsampled versions of the saliency volumes. Sampling strategies over these volumes are used to generate scan-paths over the 360-degree images. Our experiments show the advantages of using saliency volumes, and how they can be used for related tasks. Our source code and trained models available at https://github.com/massens/saliency-360salient-2017.

引用

页码：2331 / 2338

页数：8