void convert_to_polar_data_transformer::transform(
const void * data,
void * data_transformed,
neuron_data_type::input_type type,
const layer_configuration_specific& original_config,
unsigned int sample_id)
{
if (type != neuron_data_type::type_byte)
throw neural_network_exception("convert_to_polar_data_transformer is implemented for data stored as bytes only");
if (original_config.dimension_sizes.size() != 2)
throw neural_network_exception((boost::format("convert_to_polar_data_transformer is processing 2D data only, data is passed with number of dimensions %1%") % original_config.dimension_sizes.size()).str());
if (original_config.dimension_sizes != input_window_sizes)
throw neural_network_exception("convert_to_polar_data_transformer: input window size mismatch between creation and actual transform");
unsigned int original_neuron_count_per_feature_map = original_config.get_neuron_count_per_feature_map();
unsigned int transformed_neuron_count_per_feature_map = get_transformed_configuration(original_config).get_neuron_count_per_feature_map();
for(unsigned int feature_map_id = 0; feature_map_id < original_config.feature_map_count; ++feature_map_id)
{
cv::Mat1b original_image(static_cast<int>(original_config.dimension_sizes[1]), static_cast<int>(original_config.dimension_sizes[0]), const_cast<unsigned char *>(static_cast<const unsigned char *>(data)) + (original_neuron_count_per_feature_map * feature_map_id));
cv::Mat1b dest_image(static_cast<int>(output_window_sizes[1]), static_cast<int>(output_window_sizes[0]), static_cast<unsigned char *>(data_transformed) + (transformed_neuron_count_per_feature_map * feature_map_id));
// Should try INTER_CUBIC and INTER_LANCZOS4 as well
cv::remap(original_image, dest_image, map_x, map_y, cv::INTER_LINEAR, cv::BORDER_CONSTANT, border_value);
}
}
void extract_2d_data_transformer::transform(
const void * input_data,
void * output_data,
neuron_data_type::input_type type,
const layer_configuration_specific& original_config)
{
if (type != neuron_data_type::type_byte)
throw neural_network_exception("extract_2d_data_transformer is implemented for data stored as bytes only");
if (original_config.dimension_sizes.size() != 2)
throw neural_network_exception((boost::format("extract_2d_data_transformer is processing 2d data only, data is passed with number of dimensions %1%") % original_config.dimension_sizes.size()).str());
if (original_config.feature_map_count != 1)
throw neural_network_exception("extract_2d_data_transformer is implemented for 1 feature map data only");
cv::Mat1b original_image(static_cast<int>(original_config.dimension_sizes[1]), static_cast<int>(original_config.dimension_sizes[0]), const_cast<unsigned char *>(static_cast<const unsigned char *>(input_data)));
int window_top_left_x = (original_config.dimension_sizes[0] - input_window_width) / 2;
int window_bottom_right_x = window_top_left_x + input_window_width;
int window_top_left_y = (original_config.dimension_sizes[1] - input_window_height) / 2;
int window_bottom_right_y = window_top_left_y + input_window_height;
cv::Mat1b cropped_image = original_image.rowRange(window_top_left_y, window_bottom_right_y).colRange(window_top_left_x, window_bottom_right_x);
cv::Mat1b dest_image(static_cast<int>(output_window_height), static_cast<int>(output_window_width), static_cast<unsigned char *>(output_data));
cv::resize(cropped_image, dest_image, dest_image.size());
}
int main(const int argc, const char *argv[]) {
if (argc != 4) {
std::cout << "ERROR!!! Specify all arguments\n";
std::cout << "Usage:\n";
std::cout << "\tVITDatasetCreator DIR_WITH_ETALONS ALPHABET_FILE OUTPUT_DIR\n";
return 1;
}
const char *kInputDirectory = argv[1];
const char *kAlphabetFile = argv[2];
const char *kOutputDir = argv[3];
auto alphabet = ReadAlphabetFromFile(kAlphabetFile);
for (auto elem : alphabet.GetMap()) {
fs::path directory_path = fs::path(kOutputDir)
/ fs::path(elem.second.name);
if (!fs::is_directory(fs::status(directory_path))
&& !fs::create_directory(directory_path)) {
std::cout << "Can't create directory: " << directory_path << std::endl;
return 1;
}
}
std::string process_dir = kInputDirectory;
auto img_paths = DirFiles(process_dir, ".jpg");
std::cout << "There are " << img_paths.size() << " images\n";
int processed = 0;
for (auto impath : img_paths) {
if (processed % 100 == 0) {
std::cout << "Processed " << processed << " files of "
<< img_paths.size() << std::endl;
}
auto etalon_path = fs::path(impath).replace_extension("txt");
DataInfo dinfo;
try {
dinfo = ReadEtalonsFromFile(etalon_path.string());
if (0 == dinfo.pinfo.size()) continue;
}
catch(std::exception &e) {
std::cout << "Skip " << impath << ", reason: " << e.what() << std::endl;
continue;
}
//Export file to filesystem
cv::Mat original_image = cv::imread(impath.string(), CV_LOAD_IMAGE_GRAYSCALE);
for (auto p : dinfo.pinfo) {
//std::cout << p.to_string() << std::endl;
for (int i = 0; i < p.symbols.size(); i++) {
std::string dir_name = alphabet.GetElemName(p.symbols[i].text);
if (!dir_name.length()) {
std::cout << "Unknown symbol: " << p.symbols[i].text << std::endl;
continue;
}
fs::path output_dir = fs::path(kOutputDir) / fs::path(dir_name);
auto save_image = [&output_dir](const cv::Mat& image) {
std::string image_name = GetGUID() + ".bmp";
cv::Mat output_image;
cv::resize(image, output_image, cv::Size(10, 12), 0, 0, CV_INTER_LANCZOS4);
cv::imwrite((output_dir / fs::path(image_name)).c_str(), output_image);
};
auto equalize = [](const cv::Mat& image) {
cv::Mat result;
cv::equalizeHist(image, result);
return result;
};
auto get_symbol_by_roi = [&original_image](cv::Rect r,
int left_pad, int right_pad, int top_pad, int bottom_pad)
{
cv::Rect new_roi = r;
int width = r.width;
int height = r.height;
new_roi.x = std::max(0, new_roi.x - width * left_pad / 100);
new_roi.y = std::max(0, new_roi.y - height * top_pad / 100);
int new_x2 = std::min(original_image.cols, r.x + width * (1 + right_pad / 100));
int new_y2 = std::min(original_image.rows, r.y + height * (1 + bottom_pad / 100));
new_roi.width = new_x2 - new_roi.x;
new_roi.height = new_y2 - new_roi.y;
return original_image(new_roi);
};
cv::Rect symbol_roi = p.symbols[i].rect;
//todo: rewrite using cycles
save_image(equalize(get_symbol_by_roi(symbol_roi, 0, 0, 0, 0)));
save_image(equalize(get_symbol_by_roi(symbol_roi, 20, 0, 0, 0)));
save_image(equalize(get_symbol_by_roi(symbol_roi, 0, 20, 0, 0)));
save_image(equalize(get_symbol_by_roi(symbol_roi, 0, 0, 20, 0)));
save_image(equalize(get_symbol_by_roi(symbol_roi, 0, 0, 0, 20)));
save_image(equalize(get_symbol_by_roi(symbol_roi, 20, 20, 20, 20)));
save_image(get_symbol_by_roi(symbol_roi, 0, 0, 0, 0));
save_image(get_symbol_by_roi(symbol_roi, 20, 0, 0, 0));
save_image(get_symbol_by_roi(symbol_roi, 0, 20, 0, 0));
save_image(get_symbol_by_roi(symbol_roi, 0, 0, 20, 0));
save_image(get_symbol_by_roi(symbol_roi, 0, 0, 0, 20));
save_image(get_symbol_by_roi(symbol_roi, 20, 20, 20, 20));
}
}
}
exit(1);
}
void extract_data_transformer::transform(
const float * data,
float * data_transformed,
const layer_configuration_specific& original_config,
unsigned int sample_id)
{
if (input_window_sizes == output_window_sizes)
{
const std::vector<unsigned int>& dimension_sizes = original_config.dimension_sizes;
if (dimension_sizes.size() != input_window_sizes.size())
throw neural_network_exception((boost::format("extract_data_transformer is created with %1%-dimensions, data has %2% dimensions") % input_window_sizes.size() % dimension_sizes.size()).str());
std::vector<unsigned int> src_offset_list;
for(unsigned int i = 0; i < dimension_sizes.size(); ++i)
{
if (dimension_sizes[i] < output_window_sizes[i])
throw neural_network_exception((boost::format("Dimension %1% of original config has %2% size while minimum is %3%") % i % dimension_sizes[i] % output_window_sizes[i]).str());
src_offset_list.push_back((dimension_sizes[i] - output_window_sizes[i]) / 2);
}
std::vector<unsigned int> dst_pos_list(dimension_sizes.size(), 0);
const float * src_begin = data;
float * dst = data_transformed;
for(unsigned int feature_map_id = 0; feature_map_id < original_config.feature_map_count; ++feature_map_id)
{
while (true)
{
unsigned int offset = dst_pos_list.back() + src_offset_list.back();
for(int i = static_cast<int>(dimension_sizes.size()) - 2; i >= 0; --i)
offset = offset * dimension_sizes[i] + dst_pos_list[i] + src_offset_list[i];
memcpy(dst, src_begin + offset, output_window_sizes[0] * sizeof(float));
dst += output_window_sizes[0];
bool inc = false;
for(int i = 1; i < output_window_sizes.size(); ++i)
{
dst_pos_list[i]++;
if (dst_pos_list[i] < output_window_sizes[i])
{
inc = true;
break;
}
else
dst_pos_list[i] = 0;
}
if (!inc)
break;
}
src_begin += original_config.get_neuron_count_per_feature_map();
}
}
else
{
if (original_config.dimension_sizes.size() != 2)
throw neural_network_exception((boost::format("Resizing extract_data_transformer is processing 2D data only, data is passed with number of dimensions %1%") % original_config.dimension_sizes.size()).str());
int window_top_left_x = (original_config.dimension_sizes[0] - input_window_sizes[0]) / 2;
int window_bottom_right_x = window_top_left_x + input_window_sizes[0];
int window_top_left_y = (original_config.dimension_sizes[1] - input_window_sizes[1]) / 2;
int window_bottom_right_y = window_top_left_y + input_window_sizes[1];
unsigned int original_neuron_count_per_feature_map = original_config.get_neuron_count_per_feature_map();
unsigned int transformed_neuron_count_per_feature_map = get_transformed_configuration(original_config).get_neuron_count_per_feature_map();
for(unsigned int feature_map_id = 0; feature_map_id < original_config.feature_map_count; ++feature_map_id)
{
cv::Mat1f original_image(static_cast<int>(original_config.dimension_sizes[1]), static_cast<int>(original_config.dimension_sizes[0]), const_cast<float *>(data) + (original_neuron_count_per_feature_map * feature_map_id));
cv::Mat1f cropped_image = original_image.rowRange(window_top_left_y, window_bottom_right_y).colRange(window_top_left_x, window_bottom_right_x);
cv::Mat1f dest_image(static_cast<int>(output_window_sizes[1]), static_cast<int>(output_window_sizes[0]), data_transformed + (transformed_neuron_count_per_feature_map * feature_map_id));
cv::resize(cropped_image, dest_image, dest_image.size());
}
}
}
int main( int argc, char * argv[] )
{
const char * WINDOW_NAME = "Original Image vs. Box Filter vs. Gaussian";
const int QUIT_KEY_CODE = 113;
int box_filter_width = 3;
float sigma = 1.0;
std::string filename = "cameraman.tif";
ImageRAII original_image( filename );
CvSize image_dimensions = { original_image.image->width, original_image.image->height };
ImageRAII box_filter_image( cvCreateImage( image_dimensions, original_image.image->depth, 3 ) );
ImageRAII gaussian_image( cvCreateImage( image_dimensions, original_image.image->depth, 3 ) );
ImageRAII combined_image( cvCreateImage( cvSize( original_image.image->width * 3, original_image.image->height ), original_image.image->depth, 3 ) );
MatrixRAII box_filter = makeBoxFilter( box_filter_width );
MatrixRAII gaussian_filter_x = make1DGaussianFilter( sigma );
MatrixRAII gaussian_filter_y = cvCreateMat( sigma * 5, 1, CV_64FC1 );
cvTranspose( gaussian_filter_x.matrix, gaussian_filter_y.matrix );
std::vector<ImageRAII> original_image_channels( 3 );
std::vector<ImageRAII> box_filter_channels( 3 );
std::vector<ImageRAII> gaussian_filter_channels( 3 );
std::vector<ImageRAII> gaussian_filter_2_channels( 3 );
// initialize image channel vectors
for( int i = 0; i < original_image.image->nChannels; i++ )
{
original_image_channels[i].image = cvCreateImage( image_dimensions, original_image.image->depth, 1 );
box_filter_channels[i].image = cvCreateImage( image_dimensions, original_image.image->depth, 1 );
gaussian_filter_channels[i].image = cvCreateImage( image_dimensions, original_image.image->depth, 1 );
gaussian_filter_2_channels[i].image = cvCreateImage( image_dimensions, original_image.image->depth, 1 );
}
// split image channels
cvSplit( original_image.image, original_image_channels[0].image, original_image_channels[1].image, original_image_channels[2].image, NULL );
// apply filters
for( int i = 0; i < original_image.image->nChannels; i++ )
{
cvFilter2D( original_image_channels[i].image, box_filter_channels[i].image, box_filter.matrix );
cvFilter2D( original_image_channels[i].image, gaussian_filter_channels[i].image, gaussian_filter_x.matrix );
cvFilter2D( gaussian_filter_channels[i].image, gaussian_filter_2_channels[i].image, gaussian_filter_y.matrix );
}
// Merge channels back
cvMerge( box_filter_channels[0].image, box_filter_channels[1].image, box_filter_channels[2].image, NULL, box_filter_image.image );
cvMerge( gaussian_filter_2_channels[0].image, gaussian_filter_2_channels[1].image, gaussian_filter_2_channels[2].image, NULL, gaussian_image.image );
// Combine images side by side
int step = original_image.image->widthStep;
int step_destination = combined_image.image->widthStep;
int nChan = original_image.image->nChannels;
char *buf = combined_image.image->imageData;
char *original_buf = original_image.image->imageData;
char *box_filter_buf = box_filter_image.image->imageData;
char *gaussian_filter_buf = gaussian_image.image->imageData;
for( int row = 0; row < original_image.image->width; row++ )
{
for( int col = 0; col < original_image.image->height; col++ )
{
int width_adjust = 0;
//original image
// blue
*( buf + row * step_destination + nChan * col + width_adjust ) = *( original_buf + row * step + nChan * col );
// green
*( buf + row * step_destination + nChan * col + 1 + width_adjust ) = *( original_buf + row * step + nChan * col );
// red
*( buf + row * step_destination + nChan * col + 2 + width_adjust ) = *( original_buf + row * step + nChan * col );
// box filter
width_adjust = original_image.image->height * nChan;
*( buf + row * step_destination + nChan * col + width_adjust ) = *( box_filter_buf + row * step + nChan * col );
*( buf + row * step_destination + nChan * col + 1 + width_adjust ) = *( box_filter_buf + row * step + nChan * col );
*( buf + row * step_destination + nChan * col + 2 + width_adjust ) = *( box_filter_buf + row * step + nChan * col );
// gaussian filter
width_adjust = original_image.image->height * 2 * nChan;
*( buf + row * step_destination + nChan * col + width_adjust ) = *( gaussian_filter_buf + row * step + nChan * col );
*( buf + row * step_destination + nChan * col + 1 + width_adjust ) = *( gaussian_filter_buf + row * step + nChan * col );
*( buf + row * step_destination + nChan * col + 2 + width_adjust ) = *( gaussian_filter_buf + row * step + nChan * col );
}
}
// create windows
cvNamedWindow( WINDOW_NAME, CV_WINDOW_AUTOSIZE );
cvShowImage( WINDOW_NAME, combined_image.image );
// wait for keyboard input
int key_code = 0;
while( key_code != QUIT_KEY_CODE )
{
key_code = cvWaitKey( 0 );
}
return 0;
}