本文整理汇总了C++中Mat_::colRange方法的典型用法代码示例。如果您正苦于以下问题：C++ Mat_::colRange方法的具体用法？C++ Mat_::colRange怎么用？C++ Mat_::colRange使用的例子？那么, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类Mat_的用法示例。
在下文中一共展示了Mat_::colRange方法的1个代码示例，这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞，您的评价将有助于系统推荐出更棒的C++代码示例。
示例1: locallyLinearEmbeddings
void locallyLinearEmbeddings(const Mat_<float> &samples, int outDim, Mat_<float> &embeddings, int k) {
assert(outDim < samples.cols);
assert(k >= 1);
Mat_<int> nearestNeighbors(samples.rows, k);
// determining k nearest neighbors for each sample
flann::Index flannIndex(samples, flann::LinearIndexParams());
for (int i = 0; i < samples.rows; i++) {
Mat_<int> nearest;
Mat dists;
flannIndex.knnSearch(samples.row(i), nearest, dists, k + 1);
nearest.colRange(1, nearest.cols).copyTo(nearestNeighbors.row(i));
}
// determining weights for each sample
vector<Triplet<double> > tripletList;
tripletList.reserve(samples.rows * k);
for (int i = 0; i < samples.rows; i++) {
Mat_<double> C(k,k);
for (int u = 0; u < k; u++) {
for (int v = u; v < k; v++) {
C(u,v) = (samples.row(i) - samples.row(nearestNeighbors(i,u))).dot(samples.row(i) - samples.row(nearestNeighbors(i,v)));
C(v,u) = C(u,v);
}
}
// regularize when the number of neighbors is greater than the input
// dimension
if (k > samples.cols) {
C = C + Mat_<double>::eye(k,k) * 10E-3 * trace(C)[0];
}
Map<MatrixXd,RowMajor> eigC((double*)C.data, C.rows, C.cols);
LDLT<MatrixXd> solver(eigC);
Mat_<double> weights(k,1);
Map<MatrixXd,RowMajor> eigWeights((double*)weights.data, weights.rows, weights.cols);
eigWeights = solver.solve(MatrixXd::Ones(k,1));
Mat_<double> normWeights;
double weightsSum = sum(weights)[0];
if (weightsSum == 0) {
cout<<"error: cannot reconstruct point "<<i<<" from its neighbors"<<endl;
exit(EXIT_FAILURE);
}
normWeights = weights / weightsSum;
for (int j = 0; j < k; j++) {
tripletList.push_back(Triplet<double>(nearestNeighbors(i,j), i, normWeights(j)));
}
}
SparseMatrix<double> W(samples.rows, samples.rows);
W.setFromTriplets(tripletList.begin(), tripletList.end());
// constructing vectors in lower dimensional space from the weights
VectorXd eigenvalues;
MatrixXd eigenvectors;
LLEMult mult(&W);
symmetricSparseEigenSolver(samples.rows, "SM", outDim + 1, samples.rows, eigenvalues, eigenvectors, mult);
embeddings = Mat_<double>(samples.rows, outDim);
if (DEBUG_LLE) {
cout<<"actual eigenvalues : "<<eigenvalues<<endl;
cout<<"actual : "<<endl<<eigenvectors<<endl;
MatrixXd denseW(W);
MatrixXd tmp = MatrixXd::Identity(W.rows(), W.cols()) - denseW;
MatrixXd M = tmp.transpose() * tmp;
SelfAdjointEigenSolver<MatrixXd> eigenSolver(M);
MatrixXd expectedEigenvectors = eigenSolver.eigenvectors();
cout<<"expected eigenvalues : "<<eigenSolver.eigenvalues()<<endl;
cout<<"expected : "<<endl<<expectedEigenvectors<<endl;
}
for (int i = 0; i < samples.rows; i++) {
for (int j = 0; j < outDim; j++) {
embeddings(i,j) = eigenvectors(i, j + 1);
}
}
}