Chiebot-Mirror
/
opencv_contrib
mirror of https://github.com/opencv/opencv_contrib.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

text: improve DL-based samples

Browse Source
pull/1399/head
Vladislav Sovrasov 7 years ago
parent 27961cd8cc
commit fd2e37da56
4 changed files with 81 additions and 58 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 10
      modules/text/include/opencv2/text/textDetector.hpp
  2. 66
      modules/text/samples/text_recognition_cnn.cpp
  3. 39
      modules/text/samples/textbox_demo.cpp
  4. 24
      modules/text/src/text_detectorCNN.cpp

10
modules/text/include/opencv2/text/textDetector.hpp
Unescape View File

@ -54,9 +54,15 @@ public:
@param modelArchFilename the relative or absolute path to the prototxt file describing the classifiers architecture.
@param modelWeightsFilename the relative or absolute path to the file containing the pretrained weights of the model in caffe-binary form.
@param detectMultiscale if true, multiple scales of the input image will be used as network input
@param detectionSizes a list of sizes for multiscale detection. The values`[(300,300),(700,500),(700,300),(700,700),(1600,1600)]` are
recommended in @cite LiaoSBWL17 to achieve the best quality.
*/
CV_WRAP static Ptr<TextDetectorCNN> create(const String& modelArchFilename, const String& modelWeightsFilename, bool detectMultiscale = false);
static Ptr<TextDetectorCNN> create(const String& modelArchFilename, const String& modelWeightsFilename,
std::vector<Size> detectionSizes);
/**
@overload
*/
CV_WRAP static Ptr<TextDetectorCNN> create(const String& modelArchFilename, const String& modelWeightsFilename);
};
//! @}

66
modules/text/samples/text_recognition_cnn.cpp
Unescape View File

@ -1,6 +1,7 @@
#include <opencv2/text.hpp>
#include <opencv2/highgui.hpp>
#include <opencv2/imgproc.hpp>
#include <opencv2/dnn.hpp>
#include <iostream>
#include <fstream>
@ -29,22 +30,27 @@ bool fileExists (const string& filename)
return f.good();
}
void textbox_draw(Mat src, vector<Rect>& groups, vector<float>& probs, float thres)
void textbox_draw(Mat src, std::vector<Rect>& groups, std::vector<float>& probs, std::vector<int>& indexes)
{
for (size_t i = 0; i < groups.size(); i++)
for (size_t i = 0; i < indexes.size(); i++)
{
if(probs[i] > thres)
if (src.type() == CV_8UC3)
{
if (src.type() == CV_8UC3)
{
rectangle(src, groups[i], Scalar( 0, 255, 255 ), 2, LINE_AA);
String label = format("%.2f", probs[i]);
cout << "text box: " << groups[i] << " confidence: " << probs[i] << "\n";
putText(src, label, groups.at(i).tl(), FONT_HERSHEY_PLAIN, 1, Scalar( 0,0,255 ), 1, LINE_AA);
}
else
rectangle(src, groups[i], Scalar( 255 ), 3, 8 );
Rect currrentBox = groups[indexes[i]];
rectangle(src, currrentBox, Scalar( 0, 255, 255 ), 2, LINE_AA);
String label = format("%.2f", probs[indexes[i]]);
std::cout << "text box: " << currrentBox << " confidence: " << probs[indexes[i]] << "\n";
int baseLine = 0;
Size labelSize = getTextSize(label, FONT_HERSHEY_PLAIN, 1, 1, &baseLine);
int yLeftBottom = std::max(currrentBox.y, labelSize.height);
rectangle(src, Point(currrentBox.x, yLeftBottom - labelSize.height),
Point(currrentBox.x + labelSize.width, yLeftBottom + baseLine), Scalar( 255, 255, 255 ), FILLED);
putText(src, label, Point(currrentBox.x, yLeftBottom), FONT_HERSHEY_PLAIN, 1, Scalar( 0,0,0 ), 1, LINE_AA);
}
else
rectangle(src, groups[i], Scalar( 255 ), 3, 8 );
}
}
@ -73,33 +79,41 @@ int main(int argc, const char * argv[])
cout << "Starting Text Box Demo" << endl;
Ptr<text::TextDetectorCNN> textSpotter =
text::TextDetectorCNN::create(modelArch, moddelWeights, false);
text::TextDetectorCNN::create(modelArch, moddelWeights);
vector<Rect> bbox;
vector<float> outProbabillities;
textSpotter->detect(image, bbox, outProbabillities);
std::vector<int> indexes;
cv::dnn::NMSBoxes(bbox, outProbabillities, 0.4f, 0.5f, indexes);
float prob_threshold = 0.6f;
Mat image_copy = image.clone();
textbox_draw(image_copy, bbox, outProbabillities, prob_threshold);
textbox_draw(image_copy, bbox, outProbabillities, indexes);
imshow("Text detection", image_copy);
image_copy = image.clone();
Ptr<text::OCRHolisticWordRecognizer> wordSpotter =
text::OCRHolisticWordRecognizer::create("dictnet_vgg_deploy.prototxt", "dictnet_vgg.caffemodel", "dictnet_vgg_labels.txt");
for(size_t i = 0; i < bbox.size(); i++)
for(size_t i = 0; i < indexes.size(); i++)
{
if(outProbabillities[i] > prob_threshold)
{
Mat wordImg;
cvtColor(image(bbox[i]), wordImg, COLOR_BGR2GRAY);
string word;
vector<float> confs;
wordSpotter->run(wordImg, word, NULL, NULL, &confs);
rectangle(image_copy, bbox[i], Scalar(0, 255, 255), 1, LINE_AA);
putText(image_copy, word, bbox[i].tl(), FONT_HERSHEY_PLAIN, 1, Scalar(0, 0, 255), 1, LINE_AA);
}
Mat wordImg;
cvtColor(image(bbox[indexes[i]]), wordImg, COLOR_BGR2GRAY);
string word;
vector<float> confs;
wordSpotter->run(wordImg, word, NULL, NULL, &confs);
Rect currrentBox = bbox[indexes[i]];
rectangle(image_copy, currrentBox, Scalar( 0, 255, 255 ), 2, LINE_AA);
int baseLine = 0;
Size labelSize = getTextSize(word, FONT_HERSHEY_PLAIN, 1, 1, &baseLine);
int yLeftBottom = std::max(currrentBox.y, labelSize.height);
rectangle(image_copy, Point(currrentBox.x, yLeftBottom - labelSize.height),
Point(currrentBox.x + labelSize.width, yLeftBottom + baseLine), Scalar( 255, 255, 255 ), FILLED);
putText(image_copy, word, Point(currrentBox.x, yLeftBottom), FONT_HERSHEY_PLAIN, 1, Scalar( 0,0,0 ), 1, LINE_AA);
}
imshow("Text recognition", image_copy);
cout << "Recognition finished. Press any key to exit.\n";

39
modules/text/samples/textbox_demo.cpp
Unescape View File

@ -1,6 +1,7 @@
#include <opencv2/text.hpp>
#include <opencv2/highgui.hpp>
#include <opencv2/imgproc.hpp>
#include <opencv2/dnn.hpp>
#include <sstream>
#include <iostream>
@ -27,22 +28,27 @@ bool fileExists (const std::string& filename)
return f.good();
}
void textbox_draw(Mat src, std::vector<Rect>& groups, std::vector<float>& probs, float thres)
void textbox_draw(Mat src, std::vector<Rect>& groups, std::vector<float>& probs, std::vector<int>& indexes)
{
for (size_t i = 0; i < groups.size(); i++)
for (size_t i = 0; i < indexes.size(); i++)
{
if(probs[i] > thres)
if (src.type() == CV_8UC3)
{
if (src.type() == CV_8UC3)
{
rectangle(src, groups[i], Scalar( 0, 255, 255 ), 2, LINE_AA);
String label = format("%.2f", probs[i]);
std::cout << "text box: " << groups[i] << " confidence: " << probs[i] << "\n";
putText(src, label, groups.at(i).tl(), FONT_HERSHEY_PLAIN, 1, Scalar( 0,0,255 ), 1, LINE_AA);
}
else
rectangle(src, groups[i], Scalar( 255 ), 3, 8 );
Rect currrentBox = groups[indexes[i]];
rectangle(src, currrentBox, Scalar( 0, 255, 255 ), 2, LINE_AA);
String label = format("%.2f", probs[indexes[i]]);
std::cout << "text box: " << currrentBox << " confidence: " << probs[indexes[i]] << "\n";
int baseLine = 0;
Size labelSize = getTextSize(label, FONT_HERSHEY_PLAIN, 1, 1, &baseLine);
int yLeftBottom = std::max(currrentBox.y, labelSize.height);
rectangle(src, Point(currrentBox.x, yLeftBottom - labelSize.height),
Point(currrentBox.x + labelSize.width, yLeftBottom + baseLine), Scalar( 255, 255, 255 ), FILLED);
putText(src, label, Point(currrentBox.x, yLeftBottom), FONT_HERSHEY_PLAIN, 1, Scalar( 0,0,0 ), 1, LINE_AA);
}
else
rectangle(src, groups[i], Scalar( 255 ), 3, 8 );
}
}
@ -62,7 +68,7 @@ int main(int argc, const char * argv[])
if (!fileExists(modelArch) || !fileExists(moddelWeights))
{
std::cout<<getHelpStr(argv[0]);
std::cout << getHelpStr(argv[0]);
std::cout << "Model files not found in the current directory. Aborting!" << std::endl;
exit(1);
}
@ -71,13 +77,16 @@ int main(int argc, const char * argv[])
std::cout << "Starting Text Box Demo" << std::endl;
Ptr<text::TextDetectorCNN> textSpotter =
text::TextDetectorCNN::create(modelArch, moddelWeights, false);
text::TextDetectorCNN::create(modelArch, moddelWeights);
std::vector<Rect> bbox;
std::vector<float> outProbabillities;
textSpotter->detect(image, bbox, outProbabillities);
textbox_draw(image, bbox, outProbabillities, 0.5f);
std::vector<int> indexes;
cv::dnn::NMSBoxes(bbox, outProbabillities, 0.3f, 0.4f, indexes);
textbox_draw(image, bbox, outProbabillities, indexes);
imshow("TextBox Demo",image);
std::cout << "Done!" << std::endl << std::endl;

24
modules/text/src/text_detectorCNN.cpp
Unescape View File

@ -23,8 +23,6 @@ protected:
Net net_;
std::vector<Size> sizes_;
int inputChannelCount_;
bool detectMultiscale_;
void getOutputs(const float* buffer,int nbrTextBoxes,int nCol,
std::vector<Rect>& Bbox, std::vector<float>& confidence, Size inputShape)
@ -54,21 +52,12 @@ protected:
}
public:
TextDetectorCNNImpl(const String& modelArchFilename, const String& modelWeightsFilename, bool detectMultiscale) :
detectMultiscale_(detectMultiscale)
TextDetectorCNNImpl(const String& modelArchFilename, const String& modelWeightsFilename, std::vector<Size> detectionSizes) :
sizes_(detectionSizes)
{
net_ = readNetFromCaffe(modelArchFilename, modelWeightsFilename);
CV_Assert(!net_.empty());
inputChannelCount_ = 3;
sizes_.push_back(Size(700, 700));
if(detectMultiscale_)
{
sizes_.push_back(Size(300, 300));
sizes_.push_back(Size(700,500));
sizes_.push_back(Size(700,300));
sizes_.push_back(Size(1600,1600));
}
}
void detect(InputArray inputImage_, std::vector<Rect>& Bbox, std::vector<float>& confidence)
@ -92,9 +81,14 @@ public:
}
};
Ptr<TextDetectorCNN> TextDetectorCNN::create(const String &modelArchFilename, const String &modelWeightsFilename, bool detectMultiscale)
Ptr<TextDetectorCNN> TextDetectorCNN::create(const String &modelArchFilename, const String &modelWeightsFilename, std::vector<Size> detectionSizes)
{
return makePtr<TextDetectorCNNImpl>(modelArchFilename, modelWeightsFilename, detectionSizes);
}
Ptr<TextDetectorCNN> TextDetectorCNN::create(const String &modelArchFilename, const String &modelWeightsFilename)
{
return makePtr<TextDetectorCNNImpl>(modelArchFilename, modelWeightsFilename, detectMultiscale);
return create(modelArchFilename, modelWeightsFilename, std::vector<Size>(1, Size(300, 300)));
}
} //namespace text
} //namespace cv

Write Preview
Loading…
Cancel
Save