kmeans_iou: K-means clustering of 2-d data with IoU distance

In dselivanov/kmeansiou: k-means clustering of with Intersection-over-Union distance

Description

Allows to cluster 2-dimensional with k-means using intersection over union distance. This is useful for anchors initialization in the YOLO-family of models for object detection. See Redmon et al. (2016) <arXiv:1612.08242>, Redmon et al. (2016) <arXiv:1804.02767>.

Usage

kmeans_iou(x, k, iter = 10, init = x[sample(nrow(x), size = k), , drop
  = FALSE], early_stop = 3L, verbose = FALSE)

Arguments

x	input 2-d matrix: first column is wisth, second column is height of the bounding box
k	number of clusters
iter	number of k-means iterations
init	2-d matrix of initial centroids. Should have k rows
early_stop	integer - stop k-means if there is no improvement during last early_stop iterations
verbose	logical, whether to print information during fitting the model

Value

list of 2 values. First is a matrix of cluster centers, second is cluster membership indices.

dselivanov/kmeansiou documentation built on Feb. 13, 2020, 10:46 a.m.