load from config xml and fix integral representation

apps/sft/octave.cpp

@@ -92,6 +92,8 @@ bool sft::Octave::train( const cv::Mat& trainData, const cv::Mat& _responses, co
         _params.weak_count           = 1;
     }
 
+    std::cout << "WARNING: " << sampleIdx << std::endl;
+
     bool update = false;
     return cv::Boost::train(trainData, CV_COL_SAMPLE, _responses, varIdx, sampleIdx, varType, missingDataMask, _params,
     update);
@@ -104,7 +106,7 @@ class Preprocessor
 public:
     Preprocessor(int shr) : shrinkage(shr) {}
 
-    void apply(const Mat& frame, Mat integrals)
+    void apply(const Mat& frame, Mat& integrals)
     {
         CV_Assert(frame.type() == CV_8UC3);
 
@@ -178,7 +180,7 @@ void sft::Octave::processPositives(const Dataset& dataset, const FeaturePool& po
         dprintf("Process candidate positive image %s\n", curr.c_str());
 
         cv::Mat sample   = cv::imread(curr);
-        cv::Mat channels = integrals.col(total).reshape(0, h + 1);
+        cv::Mat channels = integrals.row(total).reshape(0, h + 1);
         prepocessor.apply(sample, channels);
 
         responses.ptr<float>(total)[0] = 1.f;
@@ -198,6 +200,9 @@ void sft::Octave::generateNegatives(const Dataset& dataset)
     sft::Random::engine eng;
     sft::Random::engine idxEng;
 
+    int w =  64 * pow(2, logScale) /shrinkage;
+    int h = 128 * pow(2, logScale) /shrinkage * 10;
+
     Preprocessor prepocessor(shrinkage);
 
     int nimages = (int)dataset.neg.size();
@@ -215,24 +220,33 @@ void sft::Octave::generateNegatives(const Dataset& dataset)
         Mat frame = cv::imread(dataset.neg[curr]);
         prepocessor.apply(frame, sum);
 
-        int maxW = frame.cols - 2 * boundingBox.x - boundingBox.width;
+        std::cout << "WARNING: " << frame.cols << " " << frame.rows << std::endl;
-        int maxH = frame.rows - 2 * boundingBox.y - boundingBox.height;
+        std::cout << "WARNING: " << frame.cols / shrinkage << " " << frame.rows / shrinkage << std::endl;
 
-        sft::Random::uniform wRand(0, maxW);
+        int maxW = frame.cols / shrinkage - 2 * boundingBox.x - boundingBox.width;
-        sft::Random::uniform hRand(0, maxH);
+        int maxH = frame.rows / shrinkage - 2 * boundingBox.y - boundingBox.height;
+
+        std::cout << "WARNING: " << maxW << " " << maxH << std::endl;
+
+        sft::Random::uniform wRand(0, maxW -1);
+        sft::Random::uniform hRand(0, maxH -1);
 
         int dx = wRand(eng);
         int dy = hRand(eng);
 
-        sum = sum(cv::Rect(dx, dy, boundingBox.width, boundingBox.height));
+        std::cout << "WARNING: " << dx << " " << dy << std::endl;
+        std::cout << "WARNING: " << dx + boundingBox.width + 1 << " " << dy + boundingBox.height + 1 << std::endl;
+        std::cout << "WARNING: " << sum.cols << " " << sum.rows << std::endl;
+
+        sum = sum(cv::Rect(dx, dy, boundingBox.width + 1, boundingBox.height * 10 + 1));
 
         dprintf("generated %d %d\n", dx, dy);
 
-        if (predict(sum))
+        // if (predict(sum))
         {
             responses.ptr<float>(i)[0] = 0.f;
-            sum = sum.reshape(0, 1);
+            // sum = sum.reshape(0, 1);
-            sum.copyTo(integrals.col(i));
+            sum.copyTo(integrals.row(i).reshape(0, h + 1));
             ++i;
         }
     }
@@ -257,7 +271,7 @@ bool sft::Octave::train(const Dataset& dataset, const FeaturePool& pool)
     // 3. only sumple case (all samples used)
     int nsamples = npositives + nnegatives;
     cv::Mat sampleIdx(1, nsamples, CV_32SC1);
-    ptr = varIdx.ptr<int>(0);
+    ptr = sampleIdx.ptr<int>(0);
 
     for (int x = 0; x < nsamples; ++x)
         ptr[x] = x;
@@ -281,7 +295,10 @@ bool sft::Octave::train(const Dataset& dataset, const FeaturePool& pool)
 
     cv::Mat missingMask;
 
-    return train(trainData, responses, varIdx, sampleIdx, varType, missingMask);
+    bool ok = train(trainData, responses, varIdx, sampleIdx, varType, missingMask);
+    if (!ok)
+        std::cout << "ERROR:tree couldnot be trained" << std::endl;
+    return ok;
 
 }
 

apps/sft/sft.cpp

@@ -44,85 +44,119 @@
 
 #include <sft/common.hpp>
 #include <sft/octave.hpp>
+#include <sft/config.hpp>
 
 int main(int argc, char** argv)
 {
-// hard coded now
+    using namespace sft;
-    int nfeatures  = 50;
-    int npositives = 10;
-    int nnegatives = 10;
 
-    int shrinkage  = 4;
+    const string keys =
-    int octave = 0;
+        "{help h usage ? |      | print this message              }"
+        "{config c       |      | path to configuration xml       }"
+    ;
 
-    int nsamples = npositives + nnegatives;
+    cv::CommandLineParser parser(argc, argv, keys);
-    cv::Size model(64, 128);
+    parser.about("Soft cascade training application.");
-    std::string path = "/home/kellan/cuda-dev/opencv_extra/testdata/sctrain/rescaled-train-2012-10-27-19-02-52";
 
-    cv::Rect boundingBox(5, 5 ,16, 32);
+    if (parser.has("help"))
-    sft::Octave boost(boundingBox, npositives, nnegatives, octave, shrinkage);
-
-    sft::FeaturePool pool(model, nfeatures);
-    sft::Dataset dataset(path, boost.logScale);
-
-    boost.train(dataset, pool);
-
-    cv::Mat train_data(nfeatures, nsamples, CV_32FC1);
-    cv::RNG rng;
-
-    for (int y = 0; y < nfeatures; ++y)
-        for (int x = 0; x < nsamples; ++x)
-            train_data.at<float>(y, x) = rng.uniform(0.f, 1.f);
-// +
-    int tflag = CV_COL_SAMPLE;
-    cv::Mat responses(nsamples, 1, CV_32FC1);
-    for (int y = 0; y < nsamples; ++y)
-        responses.at<float>(y, 0) = (y < npositives) ? 1.f : 0.f;
-
-
-    cv::Mat var_idx(1, nfeatures, CV_32SC1);
-    for (int x = 0; x < nfeatures; ++x)
-        var_idx.at<int>(0, x) = x;
-
-    // Mat sample_idx;
-    cv::Mat sample_idx(1, nsamples, CV_32SC1);
-    for (int x = 0; x < nsamples; ++x)
-        sample_idx.at<int>(0, x) = x;
-
-    cv::Mat var_type(1, nfeatures + 1, CV_8UC1);
-    for (int x = 0; x < nfeatures; ++x)
-        var_type.at<uchar>(0, x) = CV_VAR_ORDERED;
-
-    var_type.at<uchar>(0, nfeatures) = CV_VAR_CATEGORICAL;
-
-    cv::Mat missing_mask;
-
-    CvBoostParams params;
     {
-        params.max_categories       = 10;
+        parser.printMessage();
-        params.max_depth            = 2;
+        return 0;
-        params.min_sample_count     = 2;
-        params.cv_folds             = 0;
-        params.truncate_pruned_tree = false;
-
-        /// ??????????????????
-        params.regression_accuracy = 0.01;
-        params.use_surrogates      = false;
-        params.use_1se_rule        = false;
-
-        ///////// boost params
-        params.boost_type       = CvBoost::GENTLE;
-        params.weak_count       = 1;
-        params.split_criteria   = CvBoost::SQERR;
-        params.weight_trim_rate = 0.95;
     }
 
-    bool update = false;
+    if (!parser.check())
+    {
+        parser.printErrors();
+        return 1;
+    }
 
-    // boost.train(train_data, responses, var_idx, sample_idx, var_type, missing_mask);
+    string configPath = parser.get<string>("config");
+    if (configPath.empty())
+    {
+        std::cout << "Configuration file is missing or empty. Could not start training." << std::endl << std::flush;
+        return 0;
+    }
 
-    // CvFileStorage* fs = cvOpenFileStorage( "/home/kellan/train_res.xml", 0, CV_STORAGE_WRITE );
+    std::cout << "Read configuration from file " << configPath << std::endl;
-    // boost.write(fs, "test_res");
+    cv::FileStorage fs(configPath, cv::FileStorage::READ);
+    if(!fs.isOpened())
+    {
+        std::cout << "Configuration file " << configPath << " can't be opened." << std::endl << std::flush;
+        return 1;
+    }
 
-    // cvReleaseFileStorage( &fs );
+    // 1. load config
+    sft::Config cfg;
+    fs["config"] >> cfg;
+    std::cout << std::endl << "Training will be executed for configuration:" << std::endl << cfg << std::endl;
+
+    // 2. check and open output file
+    cv::FileStorage fso(cfg.outXmlPath, cv::FileStorage::WRITE);
+    if(!fs.isOpened())
+    {
+        std::cout << "Training stopped. Output classifier Xml file " << cfg.outXmlPath << " can't be opened." << std::endl << std::flush;
+        return 1;
+    }
+
+    // ovector strong;
+    // strong.reserve(cfg.octaves.size());
+
+    // fso << "softcascade" << "{" << "octaves" << "[";
+
+    // 3. Train all octaves
+    for (ivector::const_iterator it = cfg.octaves.begin(); it != cfg.octaves.end(); ++it)
+    {
+        int nfeatures  = cfg.poolSize;
+        int npositives = cfg.positives;
+        int nnegatives = cfg.negatives;
+
+        int shrinkage  = cfg.shrinkage;
+        int octave = *it;
+
+        cv::Size model = cfg.modelWinSize;
+        std::string path = cfg.trainPath;
+
+        cv::Rect boundingBox(cfg.offset.x / cfg.shrinkage, cfg.offset.y / cfg.shrinkage,
+            cfg.modelWinSize.width / cfg.shrinkage, cfg.modelWinSize.height / cfg.shrinkage);
+
+        sft::Octave boost(boundingBox, npositives, nnegatives, octave, shrinkage);
+
+        sft::FeaturePool pool(model, nfeatures);
+        sft::Dataset dataset(path, boost.logScale);
+
+        if (boost.train(dataset, pool))
+        {
+        }
+        std::cout << "Octave " << octave << " was successfully trained..." << std::endl;
+    //     // d. crain octave
+    //     if (octave.train(pool, cfg.positives, cfg.negatives, cfg.weaks))
+    //     {
+    //         strong.push_back(octave);
+    //     }
+    }
+
+    // fso << "]" << "}";
+
+//     // 3. create Soft Cascade
+//     // sft::SCascade cascade(cfg.modelWinSize, cfg.octs, cfg.shrinkage);
+
+//     // // 4. Generate feature pool
+//     // std::vector<sft::ICF> pool;
+//     // sft::fillPool(pool, cfg.poolSize, cfg.modelWinSize / cfg.shrinkage, cfg.seed);
+
+//     // // 5. Train all octaves
+//     // cascade.train(cfg.trainPath);
+
+//     // // 6. Set thresolds
+//     // cascade.prune();
+
+//     // // 7. Postprocess
+//     // cascade.normolize();
+
+//     // // 8. Write result xml
+//     // cv::FileStorage ofs(cfg.outXmlPath, cv::FileStorage::WRITE);
+//     // ofs << cfg.cascadeName << cascade;
+
+    std::cout << "Training complete..." << std::endl;
+    return 0;
 }