CollectNNTransform

Collect nearest neighbors and append them to metadata.

  • file: cluster/collectnn.cpp
  • inherits: UntrainableMetaTransform
  • author(s): Charles Otto

  • properties:

    Property Type Description
    keep int The maximum number of nearest neighbors to keep. Default is 20.

KMeansTransform

Wraps OpenCV kmeans and flann.

  • file: cluster/kmeans.cpp
  • inherits: Transform
  • author(s): Josh Klontz
  • see:

  • properties:

    Property Type Description
    kTrain int The number of random centroids to make at train time. Default is 256.
    kSearch int The number of nearest neighbors to search for at runtime. Default is 1.

KNNTransform

K nearest neighbors classifier.

LogNNTransform

Log nearest neighbors to specified file.

  • file: cluster/lognn.cpp
  • inherits: TimeVaryingTransform
  • author(s): Charles Otto

  • properties:

    Property Type Description
    fileName QString The name of the log file. An empty fileName won't be written to. Default is "".

MeanShiftClusteringTransform

A transform implementing the mean shift clustering algorithm.

  • file: cluster/meanshift.cpp
  • inherits: TimeVaryingTransform
  • author(s): [Jordan Cheney][JordanCheney]
  • see:

  • properties:

    Property Type Description
    distance br::Distance* The distance used to compute the distance between templates
    kernelBandwidth int The size of the kernel used to converge points to a cluster center
    shiftThreshold float The cutoff threshold distance for a shifted point. A value lower then this threshold indicates a point has finished shifting to a cluster center.
    distanceThreshold float The distance threshold for a point to join a cluster. A point must be at least this close to another point to be included in the same cluster as that point.

OnlineRODTransform

Constructors clusters based on the Rank-Order distance in an online, incremental manner

  • file: cluster/onlinerod.cpp
  • inherits: TimeVaryingTransform
  • author(s): Charles Otto, [Jordan Cheney][JordanCheney]

  • read:

    1. Zhu et al.
      "A Rank-Order Distance based Clustering Algorithm for Face Tagging" CVPR 2011

  • properties:

    Property Type Description
    distance br::Distance* Distance to compute the similarity score between templates. Default is L2.
    kNN int Maximum number of nearest neighbors to keep for each template. Default is 20.
    aggression float Clustering aggresiveness. A higher value will result in larger clusters. Default is 10.
    incremental bool If true, compute the clusters as each template is processed otherwise compute the templates at the end. Default is false.
    evalOutput QString Path to store cluster informtation. Optional. Default is an empty string.

RandomCentroidsTransform

Chooses k random points to be centroids.

  • file: cluster/randomcentroids.cpp
  • inherits: Transform
  • author(s): Austin Blanton
  • see:

  • properties:

    Property Type Description
    kTrain int The number of random centroids to make at train time. Default is 256.
    kSearch int The number of nearest neighbors to search for at runtime. Default is 1.