Predicting Human Gaze Beyond Pixels

Matlab tools for "Predicting human gaze beyond pixels," Journal of Vision, 2014
Juan Xu, Ming Jiang, Shuo Wang, Mohan Kankanhalli, Qi Zhao

Copyright (c) 2014 NUS VIP - Visual Information Processing Lab

Distributed under the MIT License. See LICENSE file in the distribution folder.

Contents

Source Code

• demo.m: demonstrates the usage of this package.
• src/common/config.m defines the configuration parameters.
• src/common/normalise.m normalises a saliency map.
• src/dataset/computeFixationMaps.m generates the human fixation maps.
• src/dataset/showEyeData.m visualises the scanpaths of a given subject.
• src/metric/computeInterSubjectAUC.m computes the ideal (inter-subject) AUC scores.
• src/metric/normalizedAUC.m computes the normalized AUC scores.
• src/model/computeIttiMaps.m computes the pixel-level feature maps (Itti & Koch model).
• src/model/extractObjectFeatures.m computes the object level feature values.
• src/model/computeObjectMaps.m computes the object-level feature maps.
• src/model/computeSemanticMaps.m computes the semantic-level feature maps.
• src/model/splitData.m splits the data into training and testing sets.
• src/model/trainModel.m trains the saliency model.
• src/model/computeSaliencyMaps.m computes the predicted saliency maps.

Data

• data/stimuli/*.jpg stimuli files
• data/eye/fixations.mat eye-tracking data (fixation points and durations)
• data/attrs.mat manually labelled object masks and attributes

Dependencies

• lib/gbvs Graph-Based Visual Saliency http://www.vision.caltech.edu/~harel/share/gbvs.php
• lib/liblinear LIBLINEAR http://www.csie.ntu.edu.tw/~cjlin/liblinear/

Getting Started

Open Matlab and run demo.m to compute the fixation maps, the feature maps, to learn and evaluate the saliency model.

Contacts

Send feedback, suggestions and questions to:
Juan Xu at jxu@nus.edu.sg
Ming Jiang at mjiang@nus.edu.sg