Journal: Behavior Research Methods

Article Title: Scanpath modeling and classification with hidden Markov models

doi: 10.3758/s13428-017-0876-8

Figure Lengend Snippet: SMAC with HMM toolbox plot Three-state HMM modeling 19 scanpaths on an image from Koehler’s dataset. Scanpaths: fixation points of the same color belong to the same observer. Emissions: three states have been identified. Emission counts: number of fixations associated with each state. Posterior probabilities: temporal evolution of the probability of being in each state. Shaded error bars represent standard error from the mean. Transition matrix: probability of going from state (or region of interest) i to j, with ( i , j ) ∈ [1..3] 2 . Priors: initial state of the model

Article Snippet: The aim of this paper is to provide a ready-made solution for gaze modeling and classification, as well as an associated Matlab toolbox: SMAC with HMM ( S canpath M odeling A nd C lassification with H idden M arkov M odels).

Techniques:

Journal: Behavior Research Methods

Article Title: Scanpath modeling and classification with hidden Markov models

doi: 10.3758/s13428-017-0876-8

Figure Lengend Snippet: SMAC with HMM toolbox plot Three-state HMM modeling 19 scanpaths recorded on a video from Coutrot’s dataset. Scanpaths: eye positions of the same color belong to the same observer. Emissions: Three states have been identified. Emission counts: number of eye positions associated with each state. Posterior probabilities: temporal evolution of the probability of being in each state. Shaded error bars represent standard error from the mean. Transition matrix: probability of going from state (or region of interest) i to j, with ( i , j ) ∈ [1..3] 2 . Priors: initial state of the model

Article Snippet: The aim of this paper is to provide a ready-made solution for gaze modeling and classification, as well as an associated Matlab toolbox: SMAC with HMM ( S canpath M odeling A nd C lassification with H idden M arkov M odels).

Techniques:

Journal: Behavior Research Methods

Article Title: Scanpath modeling and classification with hidden Markov models

doi: 10.3758/s13428-017-0876-8

Figure Lengend Snippet: SMAC with HMM toolbox plot One HMM for each of nine scanpaths recorded on a video from Coutrot’s dataset. Maximum state number \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$K^{\max }=3$\end{document} K max = 3 . Small white circles represent observer’s eye positions, red , green , and blue distributions represent HMM states. Covariance matrices have been tied to produce similar circular distributions

Article Snippet: The aim of this paper is to provide a ready-made solution for gaze modeling and classification, as well as an associated Matlab toolbox: SMAC with HMM ( S canpath M odeling A nd C lassification with H idden M arkov M odels).

Techniques:

Journal: Behavior Research Methods

Article Title: Scanpath modeling and classification with hidden Markov models

doi: 10.3758/s13428-017-0876-8

Figure Lengend Snippet: Hidden Markov models for four images and three tasks. For each image and each task, we train one HMM with the eye data of one observer. Small white circles represent the fixations of all observers following the same task. HMMs are made of states represented by Gaussian pdf ( red , green , and blue ), a transition matrix and priors. The optimal number of state has been determined by Bayesian variational approach

Article Snippet: The aim of this paper is to provide a ready-made solution for gaze modeling and classification, as well as an associated Matlab toolbox: SMAC with HMM ( S canpath M odeling A nd C lassification with H idden M arkov M odels).

Techniques:

Journal: Behavior Research Methods

Article Title: Scanpath modeling and classification with hidden Markov models

doi: 10.3758/s13428-017-0876-8

Figure Lengend Snippet: Hidden Markov models for two videos and two auditory conditions. For each video, we train one HMM with the eye data of one observer ( small white circles ) in each auditory condition (with or without the original soundtrack). HMMs are made of states represented by Gaussian pdf ( red , green , and blue ), a transition matrix and priors. The optimal number of states has been determined by Bayesian variational approach. The covariance of the HMM states on the first row is data-driven, while the one of the second rows has been tied to a circular distribution

Article Snippet: The aim of this paper is to provide a ready-made solution for gaze modeling and classification, as well as an associated Matlab toolbox: SMAC with HMM ( S canpath M odeling A nd C lassification with H idden M arkov M odels).

Techniques: