|
@@ -89,10 +89,13 @@ namespace igl
|
|
|
typedef typename DerivedF::Index Index;
|
|
|
Index count;
|
|
|
typedef std::vector<CGAL::Object> ObjectList;
|
|
|
+ // Using a vector here makes this **not** output sensitive
|
|
|
std::vector<ObjectList > F_objects;
|
|
|
Triangles T;
|
|
|
typedef std::vector<Index> IndexList;
|
|
|
IndexList lIF;
|
|
|
+ // #F-long list of bools revealing whether face is participating in any
|
|
|
+ // intersections
|
|
|
std::vector<bool> offensive;
|
|
|
std::vector<Index> offending_index;
|
|
|
std::vector<Index> offending;
|
|
@@ -209,7 +212,7 @@ namespace igl
|
|
|
// A_objects_3 updated list of intersection objects for A
|
|
|
// Outputs:
|
|
|
// cdt Contrained delaunay triangulation in projected 2D plane
|
|
|
- inline void projected_delaunay(
|
|
|
+ static inline void projected_delaunay(
|
|
|
const Triangle_3 & A,
|
|
|
const ObjectList & A_objects_3,
|
|
|
CDT_plus_2 & cdt);
|
|
@@ -414,241 +417,260 @@ inline igl::cgal::SelfIntersectMesh<
|
|
|
return;
|
|
|
}
|
|
|
|
|
|
- int NF_count = 0;
|
|
|
- // list of new faces, we'll fix F later
|
|
|
- vector<
|
|
|
- typename Eigen::Matrix<typename DerivedFF::Scalar,Dynamic,Dynamic>
|
|
|
- > NF(offending.size());
|
|
|
- // list of new vertices
|
|
|
- typedef vector<Point_3> Point_3List;
|
|
|
- Point_3List NV;
|
|
|
- Index NV_count = 0;
|
|
|
- vector<CDT_plus_2> cdt(offending.size());
|
|
|
- vector<Plane_3> P(offending.size());
|
|
|
- // Use map for *all* faces
|
|
|
- map<typename CDT_plus_2::Vertex_handle,Index> v2i;
|
|
|
- // Loop over offending triangles
|
|
|
- const size_t noff = offending.size();
|
|
|
+ const auto & remesh = [](
|
|
|
+ const Eigen::PlainObjectBase<DerivedV> & V,
|
|
|
+ const Eigen::PlainObjectBase<DerivedF> & F,
|
|
|
+ const std::vector<ObjectList > & F_objects,
|
|
|
+ const Triangles & T,
|
|
|
+ const std::vector<bool> & offensive,
|
|
|
+ const std::vector<Index> & offending_index,
|
|
|
+ const std::vector<Index> & offending,
|
|
|
+ const EdgeMap & edge2faces,
|
|
|
+ Eigen::PlainObjectBase<DerivedVV> & VV,
|
|
|
+ Eigen::PlainObjectBase<DerivedFF> & FF,
|
|
|
+ Eigen::PlainObjectBase<DerivedJ> & J,
|
|
|
+ Eigen::PlainObjectBase<DerivedIM> & IM
|
|
|
+ )
|
|
|
+ {
|
|
|
+ int NF_count = 0;
|
|
|
+ // list of new faces, we'll fix F later
|
|
|
+ vector<
|
|
|
+ typename Eigen::Matrix<typename DerivedFF::Scalar,Dynamic,Dynamic>
|
|
|
+ > NF(offending.size());
|
|
|
+ // list of new vertices
|
|
|
+ typedef vector<Point_3> Point_3List;
|
|
|
+ Point_3List NV;
|
|
|
+ Index NV_count = 0;
|
|
|
+ vector<CDT_plus_2> cdt(offending.size());
|
|
|
+ vector<Plane_3> P(offending.size());
|
|
|
+ // Use map for *all* faces
|
|
|
+ map<typename CDT_plus_2::Vertex_handle,Index> v2i;
|
|
|
+ // Loop over offending triangles
|
|
|
+ const size_t noff = offending.size();
|
|
|
#ifdef IGL_SELFINTERSECTMESH_DEBUG
|
|
|
- double t_proj_del = 0;
|
|
|
+ double t_proj_del = 0;
|
|
|
#endif
|
|
|
- // Unfortunately it looks like CGAL has trouble allocating memory when
|
|
|
- // multiple openmp threads are running. Crashes durring CDT...
|
|
|
-//# pragma omp parallel for if (noff>1000)
|
|
|
- for(Index o = 0;o<(Index)noff;o++)
|
|
|
- {
|
|
|
- // index in F
|
|
|
- const Index f = offending[o];
|
|
|
+ // Unfortunately it looks like CGAL has trouble allocating memory when
|
|
|
+ // multiple openmp threads are running. Crashes durring CDT...
|
|
|
+ //# pragma omp parallel for if (noff>1000)
|
|
|
+ for(Index o = 0;o<(Index)noff;o++)
|
|
|
{
|
|
|
+ // index in F
|
|
|
+ const Index f = offending[o];
|
|
|
+ {
|
|
|
#ifdef IGL_SELFINTERSECTMESH_DEBUG
|
|
|
- const double t_before = get_seconds();
|
|
|
+ const double t_before = get_seconds();
|
|
|
#endif
|
|
|
- projected_delaunay(T[f],F_objects[f],cdt[o]);
|
|
|
+ projected_delaunay(T[f],F_objects[f],cdt[o]);
|
|
|
#ifdef IGL_SELFINTERSECTMESH_DEBUG
|
|
|
- t_proj_del += (get_seconds()-t_before);
|
|
|
+ t_proj_del += (get_seconds()-t_before);
|
|
|
#endif
|
|
|
- }
|
|
|
- // Q: Is this also delaunay in 3D?
|
|
|
- // A: No, because the projection is affine and delaunay is not affine
|
|
|
- // invariant.
|
|
|
- // Q: Then, can't we first get the 2D delaunay triangulation, then lift it
|
|
|
- // to 3D and flip any offending edges?
|
|
|
- // Plane of projection (also used by projected_delaunay)
|
|
|
- P[o] = Plane_3(T[f].vertex(0),T[f].vertex(1),T[f].vertex(2));
|
|
|
- // Build index map
|
|
|
- {
|
|
|
- Index i=0;
|
|
|
- for(
|
|
|
- typename CDT_plus_2::Finite_vertices_iterator vit = cdt[o].finite_vertices_begin();
|
|
|
- vit != cdt[o].finite_vertices_end();
|
|
|
- ++vit)
|
|
|
+ }
|
|
|
+ // Q: Is this also delaunay in 3D?
|
|
|
+ // A: No, because the projection is affine and delaunay is not affine
|
|
|
+ // invariant.
|
|
|
+ // Q: Then, can't we first get the 2D delaunay triangulation, then lift it
|
|
|
+ // to 3D and flip any offending edges?
|
|
|
+ // Plane of projection (also used by projected_delaunay)
|
|
|
+ P[o] = Plane_3(T[f].vertex(0),T[f].vertex(1),T[f].vertex(2));
|
|
|
+ // Build index map
|
|
|
{
|
|
|
- if(i<3)
|
|
|
+ Index i=0;
|
|
|
+ for(
|
|
|
+ typename CDT_plus_2::Finite_vertices_iterator vit = cdt[o].finite_vertices_begin();
|
|
|
+ vit != cdt[o].finite_vertices_end();
|
|
|
+ ++vit)
|
|
|
{
|
|
|
- //cout<<T[f].vertex(i)<<
|
|
|
- // (T[f].vertex(i) == P[o].to_3d(vit->point())?" == ":" != ")<<
|
|
|
- // P[o].to_3d(vit->point())<<endl;
|
|
|
+ if(i<3)
|
|
|
+ {
|
|
|
+ //cout<<T[f].vertex(i)<<
|
|
|
+ // (T[f].vertex(i) == P[o].to_3d(vit->point())?" == ":" != ")<<
|
|
|
+ // P[o].to_3d(vit->point())<<endl;
|
|
|
#ifndef NDEBUG
|
|
|
- // I want to be sure that the original corners really show up as the
|
|
|
- // original corners of the CDT. I.e. I don't trust CGAL to maintain
|
|
|
- // the order
|
|
|
- assert(T[f].vertex(i) == P[o].to_3d(vit->point()));
|
|
|
+ // I want to be sure that the original corners really show up as the
|
|
|
+ // original corners of the CDT. I.e. I don't trust CGAL to maintain
|
|
|
+ // the order
|
|
|
+ assert(T[f].vertex(i) == P[o].to_3d(vit->point()));
|
|
|
#endif
|
|
|
- // For first three, use original index in F
|
|
|
-//# pragma omp critical
|
|
|
- v2i[vit] = F(f,i);
|
|
|
- }else
|
|
|
- {
|
|
|
- const Point_3 vit_point_3 = P[o].to_3d(vit->point());
|
|
|
- // First look up each edge's neighbors to see if exact point has
|
|
|
- // already been added (This makes everything a bit quadratic)
|
|
|
- bool found = false;
|
|
|
- for(int e = 0; e<3 && !found;e++)
|
|
|
+ // For first three, use original index in F
|
|
|
+ //# pragma omp critical
|
|
|
+ v2i[vit] = F(f,i);
|
|
|
+ }else
|
|
|
{
|
|
|
- // Index of F's eth edge in V
|
|
|
- Index i = F(f,(e+1)%3);
|
|
|
- Index j = F(f,(e+2)%3);
|
|
|
- // Be sure that i<j
|
|
|
- if(i>j)
|
|
|
+ const Point_3 vit_point_3 = P[o].to_3d(vit->point());
|
|
|
+ // First look up each edge's neighbors to see if exact point has
|
|
|
+ // already been added (This makes everything a bit quadratic)
|
|
|
+ bool found = false;
|
|
|
+ for(int e = 0; e<3 && !found;e++)
|
|
|
{
|
|
|
- swap(i,j);
|
|
|
- }
|
|
|
- assert(edge2faces.count(EMK(i,j))==1);
|
|
|
- // loop over neighbors
|
|
|
- for(
|
|
|
- typename IndexList::const_iterator nit = edge2faces[EMK(i,j)].begin();
|
|
|
- nit != edge2faces[EMK(i,j)].end() && !found;
|
|
|
- nit++)
|
|
|
- {
|
|
|
- // don't consider self
|
|
|
- if(*nit == f)
|
|
|
+ // Index of F's eth edge in V
|
|
|
+ Index i = F(f,(e+1)%3);
|
|
|
+ Index j = F(f,(e+2)%3);
|
|
|
+ // Be sure that i<j
|
|
|
+ if(i>j)
|
|
|
{
|
|
|
- continue;
|
|
|
+ swap(i,j);
|
|
|
}
|
|
|
- // index of neighbor in offending (to find its cdt)
|
|
|
- Index no = offending_index[*nit];
|
|
|
- // Loop over vertices of that neighbor's cdt (might not have been
|
|
|
- // processed yet, but then it's OK because it'll just be empty)
|
|
|
+ assert(edge2faces.count(EMK(i,j))==1);
|
|
|
+ const EMV & facesij = edge2faces.find(EMK(i,j))->second;
|
|
|
+ // loop over neighbors
|
|
|
for(
|
|
|
- typename CDT_plus_2::Finite_vertices_iterator uit = cdt[no].finite_vertices_begin();
|
|
|
- uit != cdt[no].finite_vertices_end() && !found;
|
|
|
- ++uit)
|
|
|
+ typename IndexList::const_iterator nit = facesij.begin();
|
|
|
+ nit != facesij.end() && !found;
|
|
|
+ nit++)
|
|
|
{
|
|
|
- if(vit_point_3 == P[no].to_3d(uit->point()))
|
|
|
+ // don't consider self
|
|
|
+ if(*nit == f)
|
|
|
+ {
|
|
|
+ continue;
|
|
|
+ }
|
|
|
+ // index of neighbor in offending (to find its cdt)
|
|
|
+ Index no = offending_index[*nit];
|
|
|
+ // Loop over vertices of that neighbor's cdt (might not have been
|
|
|
+ // processed yet, but then it's OK because it'll just be empty)
|
|
|
+ for(
|
|
|
+ typename CDT_plus_2::Finite_vertices_iterator uit = cdt[no].finite_vertices_begin();
|
|
|
+ uit != cdt[no].finite_vertices_end() && !found;
|
|
|
+ ++uit)
|
|
|
{
|
|
|
- assert(v2i.count(uit) == 1);
|
|
|
-//# pragma omp critical
|
|
|
- v2i[vit] = v2i[uit];
|
|
|
- found = true;
|
|
|
+ if(vit_point_3 == P[no].to_3d(uit->point()))
|
|
|
+ {
|
|
|
+ assert(v2i.count(uit) == 1);
|
|
|
+ //# pragma omp critical
|
|
|
+ v2i[vit] = v2i[uit];
|
|
|
+ found = true;
|
|
|
+ }
|
|
|
}
|
|
|
}
|
|
|
}
|
|
|
- }
|
|
|
- if(!found)
|
|
|
- {
|
|
|
-//# pragma omp critical
|
|
|
+ if(!found)
|
|
|
{
|
|
|
- v2i[vit] = V.rows()+NV_count;
|
|
|
- NV.push_back(vit_point_3);
|
|
|
- NV_count++;
|
|
|
+ //# pragma omp critical
|
|
|
+ {
|
|
|
+ v2i[vit] = V.rows()+NV_count;
|
|
|
+ NV.push_back(vit_point_3);
|
|
|
+ NV_count++;
|
|
|
+ }
|
|
|
}
|
|
|
}
|
|
|
+ i++;
|
|
|
}
|
|
|
- i++;
|
|
|
}
|
|
|
- }
|
|
|
- {
|
|
|
- Index i = 0;
|
|
|
- // Resize to fit new number of triangles
|
|
|
- NF[o].resize(cdt[o].number_of_faces(),3);
|
|
|
-//# pragma omp atomic
|
|
|
- NF_count+=NF[o].rows();
|
|
|
- // Append new faces to NF
|
|
|
- for(
|
|
|
- typename CDT_plus_2::Finite_faces_iterator fit = cdt[o].finite_faces_begin();
|
|
|
- fit != cdt[o].finite_faces_end();
|
|
|
- ++fit)
|
|
|
{
|
|
|
- NF[o](i,0) = v2i[fit->vertex(0)];
|
|
|
- NF[o](i,1) = v2i[fit->vertex(1)];
|
|
|
- NF[o](i,2) = v2i[fit->vertex(2)];
|
|
|
- i++;
|
|
|
+ Index i = 0;
|
|
|
+ // Resize to fit new number of triangles
|
|
|
+ NF[o].resize(cdt[o].number_of_faces(),3);
|
|
|
+ //# pragma omp atomic
|
|
|
+ NF_count+=NF[o].rows();
|
|
|
+ // Append new faces to NF
|
|
|
+ for(
|
|
|
+ typename CDT_plus_2::Finite_faces_iterator fit = cdt[o].finite_faces_begin();
|
|
|
+ fit != cdt[o].finite_faces_end();
|
|
|
+ ++fit)
|
|
|
+ {
|
|
|
+ NF[o](i,0) = v2i[fit->vertex(0)];
|
|
|
+ NF[o](i,1) = v2i[fit->vertex(1)];
|
|
|
+ NF[o](i,2) = v2i[fit->vertex(2)];
|
|
|
+ i++;
|
|
|
+ }
|
|
|
}
|
|
|
}
|
|
|
- }
|
|
|
#ifdef IGL_SELFINTERSECTMESH_DEBUG
|
|
|
- cout<<"CDT: "<<tictoc()<<" "<<t_proj_del<<endl;
|
|
|
+ cout<<"CDT: "<<tictoc()<<" "<<t_proj_del<<endl;
|
|
|
#endif
|
|
|
|
|
|
- assert(NV_count == (Index)NV.size());
|
|
|
- // Build output
|
|
|
+ assert(NV_count == (Index)NV.size());
|
|
|
+ // Build output
|
|
|
#ifndef NDEBUG
|
|
|
- {
|
|
|
- Index off_count = 0;
|
|
|
- for(Index f = 0;f<F.rows();f++)
|
|
|
{
|
|
|
- off_count+= (offensive[f]?1:0);
|
|
|
+ Index off_count = 0;
|
|
|
+ for(Index f = 0;f<F.rows();f++)
|
|
|
+ {
|
|
|
+ off_count+= (offensive[f]?1:0);
|
|
|
+ }
|
|
|
+ assert(off_count==(Index)offending.size());
|
|
|
}
|
|
|
- assert(off_count==(Index)offending.size());
|
|
|
- }
|
|
|
#endif
|
|
|
- // Append faces
|
|
|
- FF.resize(F.rows()-offending.size()+NF_count,3);
|
|
|
- J.resize(FF.rows());
|
|
|
- // First append non-offending original faces
|
|
|
- // There's an Eigen way to do this in one line but I forget
|
|
|
- Index off = 0;
|
|
|
- for(Index f = 0;f<F.rows();f++)
|
|
|
- {
|
|
|
- if(!offensive[f])
|
|
|
- {
|
|
|
- FF.row(off) = F.row(f);
|
|
|
- J(off) = f;
|
|
|
- off++;
|
|
|
- }
|
|
|
- }
|
|
|
- assert(off == (Index)(F.rows()-offending.size()));
|
|
|
- // Now append replacement faces for offending faces
|
|
|
- for(Index o = 0;o<(Index)offending.size();o++)
|
|
|
- {
|
|
|
- FF.block(off,0,NF[o].rows(),3) = NF[o];
|
|
|
- J.block(off,0,NF[o].rows(),1).setConstant(offending[o]);
|
|
|
- off += NF[o].rows();
|
|
|
- }
|
|
|
- // Append vertices
|
|
|
- VV.resize(V.rows()+NV_count,3);
|
|
|
- VV.block(0,0,V.rows(),3) = V.template cast<typename DerivedVV::Scalar>();
|
|
|
- {
|
|
|
- Index i = 0;
|
|
|
- for(
|
|
|
- typename Point_3List::const_iterator nvit = NV.begin();
|
|
|
- nvit != NV.end();
|
|
|
- nvit++)
|
|
|
+ // Append faces
|
|
|
+ FF.resize(F.rows()-offending.size()+NF_count,3);
|
|
|
+ J.resize(FF.rows());
|
|
|
+ // First append non-offending original faces
|
|
|
+ // There's an Eigen way to do this in one line but I forget
|
|
|
+ Index off = 0;
|
|
|
+ for(Index f = 0;f<F.rows();f++)
|
|
|
{
|
|
|
- for(Index d = 0;d<3;d++)
|
|
|
+ if(!offensive[f])
|
|
|
{
|
|
|
- const Point_3 & p = *nvit;
|
|
|
- // Don't convert via double if output type is same as Kernel
|
|
|
- to_output_type(p[d], VV(V.rows()+i,d));
|
|
|
+ FF.row(off) = F.row(f);
|
|
|
+ J(off) = f;
|
|
|
+ off++;
|
|
|
}
|
|
|
- i++;
|
|
|
}
|
|
|
- }
|
|
|
- IM.resize(VV.rows(),1);
|
|
|
- map<Point_3,Index> vv2i;
|
|
|
- // Safe to check for duplicates using double for original vertices: if
|
|
|
- // incoming reps are different then the points are unique.
|
|
|
- for(Index v = 0;v<V.rows();v++)
|
|
|
- {
|
|
|
- const Point_3 p(V(v,0),V(v,1),V(v,2));
|
|
|
- if(vv2i.count(p)==0)
|
|
|
+ assert(off == (Index)(F.rows()-offending.size()));
|
|
|
+ // Now append replacement faces for offending faces
|
|
|
+ for(Index o = 0;o<(Index)offending.size();o++)
|
|
|
{
|
|
|
- vv2i[p] = v;
|
|
|
+ FF.block(off,0,NF[o].rows(),3) = NF[o];
|
|
|
+ J.block(off,0,NF[o].rows(),1).setConstant(offending[o]);
|
|
|
+ off += NF[o].rows();
|
|
|
}
|
|
|
- assert(vv2i.count(p) == 1);
|
|
|
- IM(v) = vv2i[p];
|
|
|
- }
|
|
|
- // Must check for duplicates of new vertices using exact.
|
|
|
- {
|
|
|
- Index v = V.rows();
|
|
|
- for(
|
|
|
- typename Point_3List::const_iterator nvit = NV.begin();
|
|
|
- nvit != NV.end();
|
|
|
- nvit++)
|
|
|
+ // Append vertices
|
|
|
+ VV.resize(V.rows()+NV_count,3);
|
|
|
+ VV.block(0,0,V.rows(),3) = V.template cast<typename DerivedVV::Scalar>();
|
|
|
{
|
|
|
- const Point_3 & p = *nvit;
|
|
|
+ Index i = 0;
|
|
|
+ for(
|
|
|
+ typename Point_3List::const_iterator nvit = NV.begin();
|
|
|
+ nvit != NV.end();
|
|
|
+ nvit++)
|
|
|
+ {
|
|
|
+ for(Index d = 0;d<3;d++)
|
|
|
+ {
|
|
|
+ const Point_3 & p = *nvit;
|
|
|
+ // Don't convert via double if output type is same as Kernel
|
|
|
+ to_output_type(p[d], VV(V.rows()+i,d));
|
|
|
+ }
|
|
|
+ i++;
|
|
|
+ }
|
|
|
+ }
|
|
|
+ IM.resize(VV.rows(),1);
|
|
|
+ map<Point_3,Index> vv2i;
|
|
|
+ // Safe to check for duplicates using double for original vertices: if
|
|
|
+ // incoming reps are different then the points are unique.
|
|
|
+ for(Index v = 0;v<V.rows();v++)
|
|
|
+ {
|
|
|
+ const Point_3 p(V(v,0),V(v,1),V(v,2));
|
|
|
if(vv2i.count(p)==0)
|
|
|
{
|
|
|
vv2i[p] = v;
|
|
|
}
|
|
|
assert(vv2i.count(p) == 1);
|
|
|
IM(v) = vv2i[p];
|
|
|
- v++;
|
|
|
}
|
|
|
- }
|
|
|
+ // Must check for duplicates of new vertices using exact.
|
|
|
+ {
|
|
|
+ Index v = V.rows();
|
|
|
+ for(
|
|
|
+ typename Point_3List::const_iterator nvit = NV.begin();
|
|
|
+ nvit != NV.end();
|
|
|
+ nvit++)
|
|
|
+ {
|
|
|
+ const Point_3 & p = *nvit;
|
|
|
+ if(vv2i.count(p)==0)
|
|
|
+ {
|
|
|
+ vv2i[p] = v;
|
|
|
+ }
|
|
|
+ assert(vv2i.count(p) == 1);
|
|
|
+ IM(v) = vv2i[p];
|
|
|
+ v++;
|
|
|
+ }
|
|
|
+ }
|
|
|
#ifdef IGL_SELFINTERSECTMESH_DEBUG
|
|
|
- cout<<"Output + dupes: "<<tictoc()<<endl;
|
|
|
+ cout<<"Output + dupes: "<<tictoc()<<endl;
|
|
|
#endif
|
|
|
+ };
|
|
|
+ remesh(
|
|
|
+ V,F,F_objects,T,offensive,offending_index,offending,edge2faces,VV,FF,J,IM);
|
|
|
|
|
|
// Q: Does this give the same result as TETGEN?
|
|
|
// A: For the cow and beast, yes.
|
|
@@ -660,6 +682,7 @@ inline igl::cgal::SelfIntersectMesh<
|
|
|
// CGAL implementation on the beast takes 98 seconds but tetgen detection
|
|
|
// takes 14 seconds
|
|
|
|
|
|
+
|
|
|
}
|
|
|
|
|
|
|