Package edu.gsu.cs.dmlab.tracking.appearance

Class SparseGenLikeliModel

  • All Implemented Interfaces:
    ISTSparseCandidateModel
    public class SparseGenLikeliModel
extends Object
implements ISTSparseCandidateModel
    This class is used to calculate the generative likelihood of the candidate object given a learned dictionary on a training object. The input is the error of reconstruction of the candidate using the dictionary. It assumes that the minimum size of an object is 5x5 as this is what objects are automatically sized to if they are smaller than this when using the database connection class in this library. It also assumes that patches are extracted by starting in the upper left corner of the object, then moving down by one step to extract the next patch, until the bottom of the object is reached. Then it moves to the right by one and starts at the top again. This is what is done by the Image Patch Vectorizer in this library when step size is set to one. The weights of each element are determined by how close center of the patch being processed is to the center of the object, using an isotropic gaussian kernel. From Kempton et. al. 2015 and 2018.
    Author:
    Dustin Kempton, Data Mining Lab, Georgia State University

    • Constructor Detail

      • SparseGenLikeliModel

        public SparseGenLikeliModel​(int patchSize,
                            int paramDownSample)
        Constructor
        Parameters:
        patchSize - The size of the patches used in the construction of the input matrix when creating the sparse representation matrix.
        paramDownSample - The down sampling value for the objects that are passed in. The objects are generally in full resolution size but the image parameters are a down sampling of the original image, so we need to shrink the objects using this value.

    • Method Detail

      • getCandidateProb

        public double getCandidateProb​(smile.math.matrix.DenseMatrix errorMatCandidate,
                               org.locationtech.jts.geom.Envelope bbox,
                               org.locationtech.jts.geom.Envelope bbox2)
        Description copied from interface: ISTSparseCandidateModel
        Gets the likelihood value of the target candidate being a correct match based upon the error of the recreation matrix.
        Specified by:
        getCandidateProb in interface ISTSparseCandidateModel
        Parameters:
        errorMatCandidate - The error of the recreation matrix of the target candidate used to calculate the likelihood
        bbox - The original area used to determine how much the candidate needs to scale to match
        bbox2 - The target candidate area we are interested in.
        Returns:
        The likelihood value of the target candidate being correct, based on the recreation error.