/** Saves index on disk by using single or multiple files, having
proper extensions. */
int save_index (void *index, char *filename) {
char *basename = filename;
twcsa *wcsa=(twcsa *) index;
fprintf(stderr,"\nSaving structures to disk: %s.*\n",basename);
fflush(stderr);
// Saves the vocabulary of words
saveVocabulary (index, basename);
// Saves some configuration constants: maxNumOccs, sourceTextSize */
saveIndexConstants (index, filename);
#ifdef FREQ_VECTOR_AVAILABLE
// saves freqVector do disk
saveFreqVector (wcsa->freqs, wcsa->nwords, basename);
#endif
// Saving the Compressed Structure of posting lists (il)
int error = save_il(wcsa->ils, basename);
IFERRORIL(error);
fprintf(stderr,"\n \t**saved il_list");
fflush(stderr);
// Saving the Representation of the source text
save_representation(wcsa->ct, filename);
fprintf(stderr,"\n \t**saved compressed representation of the source text");
fflush(stderr);
return 0;
}
bool BowVocabulary::computeVocabulary(Mat& vocabularyOut, const string& vocabularyImgsList, bool outputAnalyzedImages, bool useOnlyTargetRegions) {
if (loadVocabulary(vocabularyOut)) {
return true;
}
_bowTrainer->clear();
ifstream imgsList(vocabularyImgsList);
if (imgsList.is_open()) {
vector<string> fileNames;
string filename;
while (getline(imgsList, filename)) {
fileNames.push_back(filename);
}
int numberOfFiles = fileNames.size();
cout << " -> Building vocabulary with " << numberOfFiles << " images..." << endl;
PerformanceTimer performanceTimer;
performanceTimer.start();
int descriptorsOriginalMatrixType = CV_32FC1;
//#pragma omp parallel for schedule(dynamic)
for (int i = 0; i < numberOfFiles; ++i) {
Mat imagePreprocessed;
string imageFilename = IMGS_DIRECTORY + fileNames[i] + IMAGE_TOKEN;
if (_imagePreprocessor->loadAndPreprocessImage(imageFilename, imagePreprocessed, CV_LOAD_IMAGE_GRAYSCALE, false)) {
Mat outputImage;
if (outputAnalyzedImages) {
outputImage = imagePreprocessed.clone();
}
if (useOnlyTargetRegions) {
vector<Mat> masks;
ImageUtils::retriveTargetsMasks(IMGS_DIRECTORY + fileNames[i], masks);
for (size_t maskIndex = 0; maskIndex < masks.size(); ++maskIndex) {
vector<KeyPoint> keypoints;
Mat targetMask = masks[maskIndex];
_featureDetector->detect(imagePreprocessed, keypoints, targetMask);
//_featureDetector->detect(imagePreprocessed, keypoints, masks[maskIndex]);
if (keypoints.size() > 3) {
Mat descriptors;
_descriptorExtractor->compute(imagePreprocessed, keypoints, descriptors);
descriptorsOriginalMatrixType = descriptors.type();
descriptors.convertTo(descriptors, CV_32FC1);
if (descriptors.rows > 0) {
//#pragma omp critical
_bowTrainer->add(descriptors);
}
if (outputAnalyzedImages) {
cv::drawKeypoints(outputImage, keypoints, outputImage);
}
}
}
} else {
vector<KeyPoint> keypoints;
_featureDetector->detect(imagePreprocessed, keypoints);
if (keypoints.size() > 3) {
Mat descriptors;
_descriptorExtractor->compute(imagePreprocessed, keypoints, descriptors);
descriptorsOriginalMatrixType = descriptors.type();
descriptors.convertTo(descriptors, CV_32FC1);
if (descriptors.rows > 0) {
//#pragma omp critical
_bowTrainer->add(descriptors);
}
if (outputAnalyzedImages) {
cv::drawKeypoints(outputImage, keypoints, outputImage);
}
}
}
if (outputAnalyzedImages) {
stringstream imageOutputFilename;
imageOutputFilename << VOCABULARY_BUILD_OUTPUT_DIRECTORY << fileNames[i] << FILENAME_SEPARATOR << _vocabularyFilename << IMAGE_OUTPUT_EXTENSION;
imwrite(imageOutputFilename.str(), outputImage);
}
}
}
vocabularyOut = _bowTrainer->cluster();
saveVocabulary(vocabularyOut);
vocabularyOut.convertTo(vocabularyOut, descriptorsOriginalMatrixType);
_bowImgDescriptorExtractor->setVocabulary(vocabularyOut);
cout << " -> Finished building vocabulary with " << vocabularyOut.rows << " word size in " << performanceTimer.getElapsedTimeFormated() << "\n" << endl;
_bowTrainer->clear();
return true;
}
return false;
}
