Retrieving facets and point from VTK file in python

Question

I have a vtk file containing a 3d model,

I would like to extract the point coordinates and the facets.

Here is a minimal working example:

import vtk
import numpy
from vtk.util.numpy_support import vtk_to_numpy

reader = vtk.vtkPolyDataReader()
reader.SetFileName('test.vtk')
reader.Update()

polydata = reader.GetOutput()

points = polydata.GetPoints()
array = points.GetData()
numpy_nodes = vtk_to_numpy(array)

This works as numpy_nodes contains the x,y,z coordinates of all points, but I am at loss to retrieve the list that relates the facets of this model to the corresponding points.

I tried:

facets= polydata.GetPolys()
array = facets.GetData()
numpy_nodes = vtk_to_numpy(array)

But then numpy_nodes is just a 1D array where I would expect a 2D array (size 3*number of facets) where the first dimension contains the number of the corresponding points to the facet (as in a .ply file).

Any advise on how to proceed would be welcome

Answer 1

You were almost there. To allow cells of different types (triangles, quads, etc.), the numpy array encodes the information with the following scheme:

numpyArray = [ n_0, id_0(0), id_0(1), ..., id_0(n0-1), 
               n_1, id_1(0), id_1(1), ..., id_1(n1-1), 
               ... 
               n_i, id_i(0), id_i(1), ..., id_1(n1-1), 
               ...
              ]

If all polys are of the same kind, that is n_i==n for all i , simply reshape the 1D array to get something interpretable:

cells = polydata.GetPolys()
nCells = cells.GetNumberOfCells()
array = cells.GetData()
# This holds true if all polys are of the same kind, e.g. triangles.
assert(array.GetNumberOfValues()%nCells==0)
nCols = array.GetNumberOfValues()//nCells
numpy_cells = vtk_to_numpy(array)
numpy_cells = numpy_cells.reshape((-1,nCols))

The first column of numpy_cells can be dropped, because it contains just the number of points per cell. But the remaining columns contain the information you were looking for.

To be sure about the result, compare the output with the "traditional" way to collect the point ids:

def getCellIds(polydata):
    cells = polydata.GetPolys()
    ids = []
    idList = vtk.vtkIdList()
    cells.InitTraversal()
    while cells.GetNextCell(idList):
        for i in range(0, idList.GetNumberOfIds()):
            pId = idList.GetId(i)
            ids.append(pId)
    ids = np.array(ids)
    return ids

numpy_cells2 = getCellIds(polydata).reshape((-1,3))

print(numpy_cells[:10,1:])
print(numpy_cells2[:10])
assert(np.array_equal(numpy_cells[:,1:], numpy_cells2))