Clustering and search

Functions
FASTCV_API int	fcvClusterEuclideanf32 (const float *__restrict points, int numPoints, int dim, int pointStride, const size_t *__restrict indices, int numIndices, int numClusters, float *__restrict clusterCenters, int clusterCenterStride, float *__restrict newClusterCenters, size_t *__restrict clusterMemberCounts, size_t *__restrict clusterBindings, float *sumOfClusterDistances)
	General function for computing cluster centers and cluster bindings for a set of points of dimension dim.
FASTCV_API int	fcvClusterEuclideanNormedf32 (const float *__restrict points, int numPoints, int dim, int pointStride, const size_t *__restrict indices, int numIndices, int numClusters, float *__restrict clusterCenters, int clusterCenterStride, float *__restrict newClusterCenters, size_t *__restrict clusterMemberCounts, size_t *__restrict clusterBindings, float *sumOfClusterDistances)
	Function for computing cluster centers and cluster bindings for a set of normalized points of dimension dim. Cluster centers are also normalized (see remark below)
FASTCV_API int	fcvClusterEuclideanNormed36f32 (const float *__restrict points, int numPoints, int pointStride, const size_t *__restrict indices, int numIndices, int numClusters, float *__restrict clusterCenters, int clusterCenterStride, float *__restrict newClusterCenters, size_t *__restrict clusterMemberCounts, size_t *__restrict clusterBindings, float *sumOfClusterDistances)
	Function for computing cluster centers and cluster bindings for a set of normalized points of dimension 36. Cluster centers are also normalized (see remark below)
FASTCV_API int	fcvClusterEuclideanu8 (const uint8_t *__restrict points, int32_t numPoints, int32_t dim, int32_t pointStride, int32_t numPointsUsed, int32_t numClusters, float32_t *__restrict clusterCenters, int32_t clusterCenterStride, float32_t *__restrict newClusterCenters, uint32_t *__restrict clusterMemberCounts, uint32_t *__restrict clusterBindings, float32_t *sumOfClusterDistances)
	General function for computing cluster centers and cluster bindings for a set of points of dimension dim.

---

Detailed Description

K clusters best fitting of a set of input points

---

Function Documentation

FASTCV_API int fcvClusterEuclideanf32	(	const float *__restrict	points,
		int	numPoints,
		int	dim,
		int	pointStride,
		const size_t *__restrict	indices,
		int	numIndices,
		int	numClusters,
		float *__restrict	clusterCenters,
		int	clusterCenterStride,
		float *__restrict	newClusterCenters,
		size_t *__restrict	clusterMemberCounts,
		size_t *__restrict	clusterBindings,
		float *	sumOfClusterDistances
	)

General function for computing cluster centers and cluster bindings for a set of points of dimension dim.

Parameters:

points	Array of all points. Array size must be greater than numPoints * dim.
numPoints	Number of points in points array.
dim	dimension, e.g. 36
pointStride	Byte distance between adjacent points in array
indices	Array of point indices in points array. Processing will only occur on points whose indices are in this array. Each index in array must be smaller numPoints.
numIndices	Length of indices array. numIndieces must be <= numPoints.
numClusters	Number of cluster centers
clusterCenters	current cluster centers; elements are distant by clusterCenterStride
clusterCenterStride	byte distance between adjacent cluster centers in array
newClusterCenters	array for new cluster centers; should be numClusterCenters long
clusterMemberCounts	Element counts for each cluster; should be numClusterCenters long
clusterBindings	Output indices of the clusters to which each vector belongs to, array must be numIndices long.
sumOfClusterDistances	Array for sum of distances of cluster elements to cluster centers; Must be numClusters long

Returns:

0 if successfully clustered, otherwise error code

Remarks:

This is general clusterer. There are no assumptions on points other than they belong to a vector space

FASTCV_API int fcvClusterEuclideanNormed36f32	(	const float *__restrict	points,
		int	numPoints,
		int	pointStride,
		const size_t *__restrict	indices,
		int	numIndices,
		int	numClusters,
		float *__restrict	clusterCenters,
		int	clusterCenterStride,
		float *__restrict	newClusterCenters,
		size_t *__restrict	clusterMemberCounts,
		size_t *__restrict	clusterBindings,
		float *	sumOfClusterDistances
	)

Function for computing cluster centers and cluster bindings for a set of normalized points of dimension 36. Cluster centers are also normalized (see remark below)

Parameters:

points	Array of all points. Array size must be greater than numPoints * 36.
numPoints	Number of points in points array.
pointStride	Byte distance between adjacent points in array
indices	Array of point indices in points array. Processing will only occur on points whose indices are in this array. Each index in array must be smaller numPoints.
numIndices	Length of indices array. numIndieces must be <= numPoints.
numClusters	Number of cluster centers
clusterCenters	current cluster centers; elements are distant by clusterCenterStride
clusterCenterStride	byte distance between adjacent cluster centers in array
newClusterCenters	array for new cluster centers; should be numClusterCenters long
clusterMemberCounts	Element counts for each cluster; should be numClusterCenters long
clusterBindings	Output indices of the clusters to which each vector belongs to, a rray must be numIndices long.
sumOfClusterDistances	Array for sum of distances of cluster elements to cluster centers; Must be numClusters long

Returns:

0 if successfully clustered, otherwise error code

Remarks:

this function assumes that points are normalized (e.g. NFT4 descriptors). Cluster centers are also normalized. Normalized points are on a surphace of unit sphere which is not a vector space but curved manifold of dimension (dim-1) embeded in Euclidean vector space of dimension dim

FASTCV_API int fcvClusterEuclideanNormedf32	(	const float *__restrict	points,
		int	numPoints,
		int	dim,
		int	pointStride,
		const size_t *__restrict	indices,
		int	numIndices,
		int	numClusters,
		float *__restrict	clusterCenters,
		int	clusterCenterStride,
		float *__restrict	newClusterCenters,
		size_t *__restrict	clusterMemberCounts,
		size_t *__restrict	clusterBindings,
		float *	sumOfClusterDistances
	)

Function for computing cluster centers and cluster bindings for a set of normalized points of dimension dim. Cluster centers are also normalized (see remark below)

Parameters:

points	Array of all points. Array size must be greater than numPoints * dim.
numPoints	Number of points in points array.
dim	dimension, e.g. 36
pointStride	Byte distance between adjacent points in array
indices	Array of point indices in points array. Processing will only occur on points whose indices are in this array. Each index in array must be smaller numPoints.
numIndices	Length of indices array. numIndieces must be <= numPoints.
numClusters	Number of cluster centers
clusterCenters	current cluster centers; elements are distant by clusterCenterStride
clusterCenterStride	byte distance between adjacent cluster centers in array
newClusterCenters	array for new cluster centers; should be numClusterCenters long
clusterMemberCounts	Element counts for each cluster; should be numClusterCenters long
clusterBindings	Output indices of the clusters to which each vector belongs to, a rray must be numIndices long.
sumOfClusterDistances	Array for sum of distances of cluster elements to cluster centers; Must be numClusters long

Returns:

0 if successfully clustered, otherwise error code

Remarks:

this function assumes that points are normalized (e.g. NFT4 descriptors). Cluster centers are also normalized. Normalized points are on a surface of unit sphere which is not a vector space but curved manifold of dimension (dim-1) embeded in Euclidean vector space of dimension dim

FASTCV_API int fcvClusterEuclideanu8	(	const uint8_t *__restrict	points,
		int32_t	numPoints,
		int32_t	dim,
		int32_t	pointStride,
		int32_t	numPointsUsed,
		int32_t	numClusters,
		float32_t *__restrict	clusterCenters,
		int32_t	clusterCenterStride,
		float32_t *__restrict	newClusterCenters,
		uint32_t *__restrict	clusterMemberCounts,
		uint32_t *__restrict	clusterBindings,
		float32_t *	sumOfClusterDistances
	)

General function for computing cluster centers and cluster bindings for a set of points of dimension dim.

Parameters:

points	Array of all points. Array size must be greater than numPoints * pointStride.
numPoints	Number of points in points array.
dim	dimension, e.g. 36
pointStride	Byte distance between adjacent points in array
numPointsUsed	Total number of points used for clustering, {0,1,...(numPointsUsed-1)}
numClusters	Number of clusters
clusterCenters	current cluster centers; elements are distant by clusterCenterStride
clusterCenterStride	byte distance between adjacent cluster centers in array
newClusterCenters	array for new cluster centers; should be numClusterCenters long
newClusterMemberCounts	Element counts for each cluster; should be numClusterCenters long
clusterBindings	Output indices of the clusters to which each vector belongs to, array must be numPointsUsed long.
sumOfClusterDistances	the sum of distances between each cluster center to its belonging points. The size should be numClusterCenters*sizeof(float_32)

Returns:

0 if successfully clustered, otherwise error code

Remarks:

This is general clusterer. There are no assumptions on points other than they belong to a vector space