I was looking at an example of the CvNormalBayesClassifier::train in which the input/output matrix is to be a 1D vector.
The example I was looking at achieved this by creating a cv::Mat matrix with 0 rows and 1000 columns using this line:
Mat trainingData(0, 1000, CV_32FC1);
Reading the basic data types in opencv documentation this is what I found for Mat:
There are many different ways to create Mat object. Here are the some popular ones:
using create(nrows, ncols, type) method or the similar constructor Mat(nrows, ncols, type[, fill_value]) constructor.
In any way the first parameter is the rows. The way I look at it is even if we do create a 1000 column matrix it will atleast have 1 row. How can it have 0 rows?
Sorry if this is a very basic question.
update: upon request, here is the complete code.
#include <vector>
#include <boost/filesystem.hpp>
#include <opencv2/opencv.hpp>
using namespace std;
using namespace boost::filesystem;
using namespace cv;
//location of the training data
#define TRAINING_DATA_DIR "data/train/"
//location of the evaluation data
#define EVAL_DATA_DIR "data/eval/"
//See article on BoW model for details
Ptr<DescriptorMatcher> matcher = DescriptorMatcher::create("FlannBased");
Ptr<DescriptorExtractor> extractor = DescriptorExtractor::create("SURF");
Ptr<FeatureDetector> detector = FeatureDetector::create("SURF");
//See article on BoW model for details
int dictionarySize = 1000;
TermCriteria tc(CV_TERMCRIT_ITER, 10, 0.001);
int retries = 1;
int flags = KMEANS_PP_CENTERS;
//See article on BoW model for details
BOWKMeansTrainer bowTrainer(dictionarySize, tc, retries, flags);
//See article on BoW model for details
BOWImgDescriptorExtractor bowDE(extractor, matcher);
/**
* \brief Recursively traverses a folder hierarchy. Extracts features from the training images and adds them to the bowTrainer.
*/
void extractTrainingVocabulary(const path& basepath) {
for (directory_iterator iter = directory_iterator(basepath); iter
!= directory_iterator(); iter++) {
directory_entry entry = *iter;
if (is_directory(entry.path())) {
cout << "Processing directory " << entry.path().string() << endl;
extractTrainingVocabulary(entry.path());
} else {
path entryPath = entry.path();
if (entryPath.extension() == ".jpg") {
cout << "Processing file " << entryPath.string() << endl;
Mat img = imread(entryPath.string());
if (!img.empty()) {
vector<KeyPoint> keypoints;
detector->detect(img, keypoints);
if (keypoints.empty()) {
cerr << "Warning: Could not find key points in image: "
<< entryPath.string() << endl;
} else {
Mat features;
extractor->compute(img, keypoints, features);
bowTrainer.add(features);
}
} else {
cerr << "Warning: Could not read image: "
<< entryPath.string() << endl;
}
}
}
}
}
/**
* \brief Recursively traverses a folder hierarchy. Creates a BoW descriptor for each image encountered.
*/
void extractBOWDescriptor(const path& basepath, Mat& descriptors, Mat& labels) {
for (directory_iterator iter = directory_iterator(basepath); iter
!= directory_iterator(); iter++) {
directory_entry entry = *iter;
if (is_directory(entry.path())) {
cout << "Processing directory " << entry.path().string() << endl;
extractBOWDescriptor(entry.path(), descriptors, labels);
} else {
path entryPath = entry.path();
if (entryPath.extension() == ".jpg") {
cout << "Processing file " << entryPath.string() << endl;
Mat img = imread(entryPath.string());
if (!img.empty()) {
vector<KeyPoint> keypoints;
detector->detect(img, keypoints);
if (keypoints.empty()) {
cerr << "Warning: Could not find key points in image: "
<< entryPath.string() << endl;
} else {
Mat bowDescriptor;
bowDE.compute(img, keypoints, bowDescriptor);
descriptors.push_back(bowDescriptor);
float label=atof(entryPath.filename().c_str());
labels.push_back(label);
}
} else {
cerr << "Warning: Could not read image: "
<< entryPath.string() << endl;
}
}
}
}
}
int main(int argc, char ** argv) {
cout<<"Creating dictionary..."<<endl;
extractTrainingVocabulary(path(TRAINING_DATA_DIR));
vector<Mat> descriptors = bowTrainer.getDescriptors(); //descriptors from training images
int count=0;
for(vector<Mat>::iterator iter=descriptors.begin();iter!=descriptors.end();iter++)
{
count+=iter->rows;
}
cout<<"Clustering "<<count<<" features"<<endl;
Mat dictionary = bowTrainer.cluster();
bowDE.setVocabulary(dictionary);
cout<<"Processing training data..."<<endl;
Mat trainingData(0, dictionarySize, CV_32FC1);
Mat labels(0, 1, CV_32FC1);
extractBOWDescriptor(path(TRAINING_DATA_DIR), trainingData, labels);
NormalBayesClassifier classifier;
cout<<"Training classifier..."<<endl;
classifier.train(trainingData, labels);
cout<<"Processing evaluation data..."<<endl;
Mat evalData(0, dictionarySize, CV_32FC1);
Mat groundTruth(0, 1, CV_32FC1);
extractBOWDescriptor(path(EVAL_DATA_DIR), evalData, groundTruth);
cout<<"Evaluating classifier..."<<endl;
Mat results;
classifier.predict(evalData, &results);
double errorRate = (double) countNonZero(groundTruth - results) / evalData.rows;
;
cout << "Error rate: " << errorRate << endl;
}
