generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

upgraded to FLANN 1.6. Added miniflann interface, which is now used in the rest of OpenCV. Added Python bindings for FLANN.

Browse Source
This commit is contained in:
Vadim Pisarevsky
2011-07-13 23:04:39 +00:00
parent 4e42bf6308
commit 562914e33b
48 changed files with 8503 additions and 3606 deletions
Download Patch File Download Diff File Expand all files Collapse all files

422
modules/flann/include/opencv2/flann/flann_base.hpp
View File

@@ -28,232 +28,264 @@
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*************************************************************************/
#ifndef _OPENCV_FLANN_BASE_HPP_
#define _OPENCV_FLANN_BASE_HPP_
#ifndef FLANN_BASE_HPP_
#define FLANN_BASE_HPP_
#include <vector>
#include <string>
#include <cassert>
#include <cstdio>
#include "opencv2/flann/general.h"
#include "opencv2/flann/matrix.h"
#include "opencv2/flann/result_set.h"
#include "opencv2/flann/index_testing.h"
#include "opencv2/flann/object_factory.h"
#include "opencv2/flann/saving.h"
#include "general.h"
#include "matrix.h"
#include "params.h"
#include "saving.h"
#include "opencv2/flann/all_indices.h"
#include "all_indices.h"
namespace cvflann
{
/**
Sets the log level used for all flann functions
Params:
level = verbosity level
*/
CV_EXPORTS void log_verbosity(int level);
/**
* Sets the distance type to use throughout FLANN.
* If distance type specified is MINKOWSKI, the second argument
* specifies which order the minkowski distance should have.
* Sets the log level used for all flann functions
* @param level Verbosity level
*/
CV_EXPORTS void set_distance_type(flann_distance_t distance_type, int order);
inline void log_verbosity(int level)
{
if (level >= 0) {
Logger::setLevel(level);
}
}
struct CV_EXPORTS SavedIndexParams : public IndexParams {
SavedIndexParams(std::string filename_) : IndexParams(FLANN_INDEX_SAVED), filename(filename_) {}
std::string filename; // filename of the stored index
void print() const
{
logger().info("Index type: %d\n",(int)algorithm);
logger().info("Filename: %s\n", filename.c_str());
}
/**
* (Deprecated) Index parameters for creating a saved index.
*/
struct SavedIndexParams : public IndexParams
{
SavedIndexParams(std::string filename)
{
(* this)["algorithm"] = FLANN_INDEX_SAVED;
(*this)["filename"] = filename;
}
};
template<typename T>
class CV_EXPORTS Index {
NNIndex<T>* nnIndex;
bool built;
template<typename Distance>
NNIndex<Distance>* load_saved_index(const Matrix<typename Distance::ElementType>& dataset, const std::string& filename, Distance distance)
{
typedef typename Distance::ElementType ElementType;
FILE* fin = fopen(filename.c_str(), "rb");
if (fin == NULL) {
return NULL;
}
IndexHeader header = load_header(fin);
if (header.data_type != Datatype<ElementType>::type()) {
throw FLANNException("Datatype of saved index is different than of the one to be created.");
}
if ((size_t(header.rows) != dataset.rows)||(size_t(header.cols) != dataset.cols)) {
throw FLANNException("The index saved belongs to a different dataset");
}
IndexParams params;
params["algorithm"] = header.index_type;
NNIndex<Distance>* nnIndex = create_index_by_type<Distance>(dataset, params, distance);
nnIndex->loadIndex(fin);
fclose(fin);
return nnIndex;
}
template<typename Distance>
class Index : public NNIndex<Distance>
{
public:
Index(const Matrix<T>& features, const IndexParams& params);
typedef typename Distance::ElementType ElementType;
typedef typename Distance::ResultType DistanceType;
~Index();
Index(const Matrix<ElementType>& features, const IndexParams& params, Distance distance = Distance() )
: index_params_(params)
{
flann_algorithm_t index_type = get_param<flann_algorithm_t>(params,"algorithm");
loaded_ = false;
void buildIndex();
if (index_type == FLANN_INDEX_SAVED) {
nnIndex_ = load_saved_index<Distance>(features, get_param<std::string>(params,"filename"), distance);
loaded_ = true;
}
else {
nnIndex_ = create_index_by_type<Distance>(features, params, distance);
}
}
void knnSearch(const Matrix<T>& queries, Matrix<int>& indices, Matrix<float>& dists, int knn, const SearchParams& params);
~Index()
{
delete nnIndex_;
}
int radiusSearch(const Matrix<T>& query, Matrix<int>& indices, Matrix<float>& dists, float radius, const SearchParams& params);
/**
* Builds the index.
*/
void buildIndex()
{
if (!loaded_) {
nnIndex_->buildIndex();
}
}
void save(std::string filename);
void save(std::string filename)
{
FILE* fout = fopen(filename.c_str(), "wb");
if (fout == NULL) {
throw FLANNException("Cannot open file");
}
save_header(fout, *nnIndex_);
saveIndex(fout);
fclose(fout);
}
int veclen() const;
/**
* \brief Saves the index to a stream
* \param stream The stream to save the index to
*/
virtual void saveIndex(FILE* stream)
{
nnIndex_->saveIndex(stream);
}
int size() const;
/**
* \brief Loads the index from a stream
* \param stream The stream from which the index is loaded
*/
virtual void loadIndex(FILE* stream)
{
nnIndex_->loadIndex(stream);
}
NNIndex<T>* getIndex() { return nnIndex; }
/**
* \returns number of features in this index.
*/
size_t veclen() const
{
return nnIndex_->veclen();
}
const IndexParams* getIndexParameters() { return nnIndex->getParameters(); }
/**
* \returns The dimensionality of the features in this index.
*/
size_t size() const
{
return nnIndex_->size();
}
/**
* \returns The index type (kdtree, kmeans,...)
*/
flann_algorithm_t getType() const
{
return nnIndex_->getType();
}
/**
* \returns The amount of memory (in bytes) used by the index.
*/
virtual int usedMemory() const
{
return nnIndex_->usedMemory();
}
/**
* \returns The index parameters
*/
IndexParams getParameters() const
{
return nnIndex_->getParameters();
}
/**
* \brief Perform k-nearest neighbor search
* \param[in] queries The query points for which to find the nearest neighbors
* \param[out] indices The indices of the nearest neighbors found
* \param[out] dists Distances to the nearest neighbors found
* \param[in] knn Number of nearest neighbors to return
* \param[in] params Search parameters
*/
void knnSearch(const Matrix<ElementType>& queries, Matrix<int>& indices, Matrix<DistanceType>& dists, int knn, const SearchParams& params)
{
nnIndex_->knnSearch(queries, indices, dists, knn, params);
}
/**
* \brief Perform radius search
* \param[in] query The query point
* \param[out] indices The indinces of the neighbors found within the given radius
* \param[out] dists The distances to the nearest neighbors found
* \param[in] radius The radius used for search
* \param[in] params Search parameters
* \returns Number of neighbors found
*/
int radiusSearch(const Matrix<ElementType>& query, Matrix<int>& indices, Matrix<DistanceType>& dists, float radius, const SearchParams& params)
{
return nnIndex_->radiusSearch(query, indices, dists, radius, params);
}
/**
* \brief Method that searches for nearest-neighbours
*/
void findNeighbors(ResultSet<DistanceType>& result, const ElementType* vec, const SearchParams& searchParams)
{
nnIndex_->findNeighbors(result, vec, searchParams);
}
/**
* \brief Returns actual index
*/
FLANN_DEPRECATED NNIndex<Distance>* getIndex()
{
return nnIndex_;
}
/**
* \brief Returns index parameters.
* \deprecated use getParameters() instead.
*/
FLANN_DEPRECATED const IndexParams* getIndexParameters()
{
return &index_params_;
}
private:
/** Pointer to actual index class */
NNIndex<Distance>* nnIndex_;
/** Indices if the index was loaded from a file */
bool loaded_;
/** Parameters passed to the index */
IndexParams index_params_;
};
template<typename T>
NNIndex<T>* load_saved_index(const Matrix<T>& dataset, const std::string& filename)
/**
* Performs a hierarchical clustering of the points passed as argument and then takes a cut in the
* the clustering tree to return a flat clustering.
* @param[in] points Points to be clustered
* @param centers The computed cluster centres. Matrix should be preallocated and centers.rows is the
* number of clusters requested.
* @param params Clustering parameters (The same as for cvflann::KMeansIndex)
* @param d Distance to be used for clustering (eg: cvflann::L2)
* @return number of clusters computed (can be different than clusters.rows and is the highest number
* of the form (branching-1)*K+1 smaller than clusters.rows).
*/
template <typename Distance>
int hierarchicalClustering(const Matrix<typename Distance::ElementType>& points, Matrix<typename Distance::ResultType>& centers,
const KMeansIndexParams& params, Distance d = Distance())
{
FILE* fin = fopen(filename.c_str(), "rb");
if (fin==NULL) {
return NULL;
}
IndexHeader header = load_header(fin);
if (header.data_type!=Datatype<T>::type()) {
throw FLANNException("Datatype of saved index is different than of the one to be created.");
}
if (size_t(header.rows)!=dataset.rows || size_t(header.cols)!=dataset.cols) {
throw FLANNException("The index saved belongs to a different dataset");
}
IndexParams* params = ParamsFactory_instance().create(header.index_type);
NNIndex<T>* nnIndex = create_index_by_type(dataset, *params);
nnIndex->loadIndex(fin);
fclose(fin);
return nnIndex;
}
template<typename T>
Index<T>::Index(const Matrix<T>& dataset, const IndexParams& params)
{
flann_algorithm_t index_type = params.getIndexType();
built = false;
if (index_type==FLANN_INDEX_SAVED) {
nnIndex = load_saved_index(dataset, ((const SavedIndexParams&)params).filename);
built = true;
}
else {
nnIndex = create_index_by_type(dataset, params);
}
}
template<typename T>
Index<T>::~Index()
{
delete nnIndex;
}
template<typename T>
void Index<T>::buildIndex()
{
if (!built) {
nnIndex->buildIndex();
built = true;
}
}
template<typename T>
void Index<T>::knnSearch(const Matrix<T>& queries, Matrix<int>& indices, Matrix<float>& dists, int knn, const SearchParams& searchParams)
{
if (!built) {
throw FLANNException("You must build the index before searching.");
}
assert(queries.cols==nnIndex->veclen());
assert(indices.rows>=queries.rows);
assert(dists.rows>=queries.rows);
assert(int(indices.cols)>=knn);
assert(int(dists.cols)>=knn);
KNNResultSet<T> resultSet(knn);
for (size_t i = 0; i < queries.rows; i++) {
T* target = queries[i];
resultSet.init(target, (int)queries.cols);
nnIndex->findNeighbors(resultSet, target, searchParams);
int* neighbors = resultSet.getNeighbors();
float* distances = resultSet.getDistances();
memcpy(indices[i], neighbors, knn*sizeof(int));
memcpy(dists[i], distances, knn*sizeof(float));
}
}
template<typename T>
int Index<T>::radiusSearch(const Matrix<T>& query, Matrix<int>& indices, Matrix<float>& dists, float radius, const SearchParams& searchParams)
{
if (!built) {
throw FLANNException("You must build the index before searching.");
}
if (query.rows!=1) {
fprintf(stderr, "I can only search one feature at a time for range search\n");
return -1;
}
assert(query.cols==nnIndex->veclen());
RadiusResultSet<T> resultSet(radius);
resultSet.init(query.data, (int)query.cols);
nnIndex->findNeighbors(resultSet,query.data,searchParams);
// TODO: optimise here
int* neighbors = resultSet.getNeighbors();
float* distances = resultSet.getDistances();
size_t count_nn = std::min(resultSet.size(), indices.cols);
assert (dists.cols>=count_nn);
for (size_t i=0;i<count_nn;++i) {
indices[0][i] = neighbors[i];
dists[0][i] = distances[i];
}
return (int)count_nn;
}
template<typename T>
void Index<T>::save(std::string filename)
{
FILE* fout = fopen(filename.c_str(), "wb");
if (fout==NULL) {
throw FLANNException("Cannot open file");
}
save_header(fout, *nnIndex);
nnIndex->saveIndex(fout);
fclose(fout);
}
template<typename T>
int Index<T>::size() const
{
return nnIndex->size();
}
template<typename T>
int Index<T>::veclen() const
{
return nnIndex->veclen();
}
template <typename ELEM_TYPE, typename DIST_TYPE>
int hierarchicalClustering(const Matrix<ELEM_TYPE>& features, Matrix<DIST_TYPE>& centers, const KMeansIndexParams& params)
{
KMeansIndex<ELEM_TYPE, DIST_TYPE> kmeans(features, params);
kmeans.buildIndex();
KMeansIndex<Distance> kmeans(points, params, d);
kmeans.buildIndex();
int clusterNum = kmeans.getClusterCenters(centers);
return clusterNum;
return clusterNum;
}
} // namespace cvflann
#endif /* _OPENCV_FLANN_BASE_HPP_ */
}
#endif /* FLANN_BASE_HPP_ */