零行 1000 列的矩阵？

Question

我正在查看CvNormalBayesClassifier::train 的示例，其中输入/输出矩阵将是一维向量。

我正在查看的示例通过使用此行创建一个 0 行和 1000 列的 cv::Mat 矩阵来实现这一点：

Mat trainingData(0, 1000, CV_32FC1);

阅读 opencv 文档中的基本数据类型，这是我为 Mat 找到的：

有许多不同的方法来创建 Mat 对象。这里有一些 热门：

using create(nrows, ncols, type) method or

    the similar constructor

Mat(nrows, ncols, type[, fill_value]) constructor.

无论如何，第一个参数是行。我看它的方式是，即使我们确实创建了一个 1000 列的矩阵，它也至少有 1 行。怎么会有0行？

抱歉，这是一个非常基本的问题。

更新：根据要求，这里是完整的代码。

    #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;

}

Answer 1

现在您已经发布了代码，这很有意义。这个 0 行向量被初始化为有 0 行，但它是增量创建的。

0 行矩阵被传递给extractBOWDescriptor()，它自己计算几个描述符并使用cv::Mat.push_back() 向矩阵添加行。

它从 0 行开始，因为一开始我们没有描述符来填充矩阵。