Fix T53581: remesh modifier artifacts in sharp mode

Replace relative threshold for pseudo inverse in sharp remeshing modifier with
0.1 as proposed in the original paper.

Also change pseudo-inverse implementation that works with dynamic heap-allocated
matrix to static 3x3 version, for performance

Differential Revision: https://developer.blender.org/D5078
This commit is contained in:
Dan Koschier 2019-06-17 18:35:18 +02:00 committed by Brecht Van Lommel
parent 538f2aeaef
commit 4cc98af3a4

@ -2180,18 +2180,12 @@ void Octree::countIntersection(Node *node, int height, int &nedge, int &ncell, i
}
/* from http://eigen.tuxfamily.org/bz/show_bug.cgi?id=257 */
template<typename _Matrix_Type_>
void pseudoInverse(const _Matrix_Type_ &a,
_Matrix_Type_ &result,
double epsilon = std::numeric_limits<typename _Matrix_Type_::Scalar>::epsilon())
static void pseudoInverse(const Eigen::Matrix3f &a, Eigen::Matrix3f &result, float tolerance)
{
Eigen::JacobiSVD<_Matrix_Type_> svd = a.jacobiSvd(Eigen::ComputeFullU | Eigen::ComputeFullV);
typename _Matrix_Type_::Scalar tolerance = epsilon * std::max(a.cols(), a.rows()) *
svd.singularValues().array().abs().maxCoeff();
Eigen::JacobiSVD<Eigen::Matrix3f> svd = a.jacobiSvd(Eigen::ComputeFullU | Eigen::ComputeFullV);
result = svd.matrixV() *
_Matrix_Type_((svd.singularValues().array().abs() > tolerance)
Eigen::Vector3f((svd.singularValues().array().abs() > tolerance)
.select(svd.singularValues().array().inverse(), 0))
.asDiagonal() *
svd.matrixU().adjoint();
@ -2203,9 +2197,9 @@ static void solve_least_squares(const float halfA[],
float rvalue[])
{
/* calculate pseudo-inverse */
Eigen::MatrixXf A(3, 3), pinv(3, 3);
Eigen::Matrix3f A, pinv;
A << halfA[0], halfA[1], halfA[2], halfA[1], halfA[3], halfA[4], halfA[2], halfA[4], halfA[5];
pseudoInverse(A, pinv);
pseudoInverse(A, pinv, 0.1f);
Eigen::Vector3f b2(b), mp(midpoint), result;
b2 = b2 + A * -mp;