问题
I am trying to write a vtk filter in python ProjectDepthImage Doing the projection isn't the problem. It's controlling the execution of the vtk pipeline.
Basically, I have a callback on 'UserEvent' (which is triggered when a user hits the key 'u' while the render window is active.) This calls the function: render_point_cloud() I have found that I need to call both dif.Modified() and pdi.Update() in this callback or I will get unexpected results. However, by doing this this RequestData() gets executed twice, as indicated by the print statement.
I really want to use the VTK pipeline correctly, and I want RequestData() to be executed only once when the user hits 'u'
I have to call Modified() on dif because of the way vtkWindowToImageFilter works: http://www.vtk.org/doc/nightly/html/classvtkWindowToImageFilter.html#details
Best post on vtkPythonAlgorthim that I have found: https://blog.kitware.com/vtkpythonalgorithm-is-great/
import vtk
from vtk.util import numpy_support
from vtk.numpy_interface import dataset_adapter as dsa
from vtk.numpy_interface import algorithms as alg
import numpy as np
import matplotlib.pyplot as plt
from timeit import default_timer as timer
import time
# import vtk
from vtk.util.vtkAlgorithm import VTKPythonAlgorithmBase
class ProjectDepthImage(VTKPythonAlgorithmBase):
def __init__(self):
VTKPythonAlgorithmBase.__init__(self,
nInputPorts=1, inputType='vtkImageData',
nOutputPorts=1, outputType='vtkPolyData')
self.__ren = []
self.__renWin = []
self.__display_pts = []
self.__viewport_pts = []
self.__world_pts = []
self.__sizex = []
self.__sizey = []
def SetRenderer(self, renderer):
if renderer != self.__ren:
self.__ren = renderer
self.Modified()
def GetRenderer(self):
return self.__ren
def SetRenderWindow(self, render_window):
if render_window != self.__renWin:
self.__renWin = render_window
self.Modified()
def GetRenderWindow(self):
return self.__renWin
def RequestData(self, request, inInfo, outInfo):
print 'Executing'
# all the depth values
inp = dsa.WrapDataObject(vtk.vtkImageData.GetData(inInfo[0]))
depth = numpy_support.vtk_to_numpy(inp.PointData['ImageScalars'])
# update the viewport points, check to see if render window size has changed
self.__update_viewport_points(depth)
# transformation matrix, viewport coordinates -> world coordinates
tmat = ren.GetActiveCamera().GetCompositeProjectionTransformMatrix(
ren.GetTiledAspectRatio(),
0.0, 1.0)
tmat.Invert()
tmat = self.__vtkmatrix_to_numpy(tmat)
# project to world coordinates
self.__world_pts = np.dot(tmat, self.__viewport_pts)
self.__world_pts = self.__world_pts / self.__world_pts[3]
# for storing the point cloud
points = vtk.vtkPoints()
vertices = vtk.vtkCellArray()
polydata = vtk.vtkPolyData()
out = vtk.vtkPolyData.GetData(outInfo)
for i in np.arange(self.__world_pts.shape[1]):
pt_id = points.InsertNextPoint(self.__world_pts[0:3, i])
vertices.InsertNextCell(1)
vertices.InsertCellPoint(pt_id)
polydata.SetPoints(points)
polydata.SetVerts(vertices)
out.ShallowCopy(polydata)
return 1
def __update_viewport_points(self, depth):
# if the render windows size has not changed, just return
if (self.__sizex, self.__sizey) == self.__renWin.GetSize():
self.__viewport_pts[2, :] = depth
return
# save the new size
(self.__sizex, self.__sizey) = self.__renWin.GetSize()
# new display points
self.__display_pts = np.ones((2, self.__sizex*self.__sizey))
count = 0
for i in np.arange(self.__sizey):
for j in np.arange(self.__sizex):
self.__display_pts[0, count] = j
self.__display_pts[1, count] = i
count += 1
# new viewport points
viewport = self.__ren.GetViewport()
self.__viewport_pts = np.ones((4, self.__display_pts.shape[1]))
self.__viewport_pts[0,:] = 2.0 * (self.__display_pts[0,:] - self.__sizex*viewport[0]) / (self.__sizex*(viewport[2]-viewport[0])) - 1.0
self.__viewport_pts[1,:] = 2.0 * (self.__display_pts[1,:] - self.__sizey*viewport[1]) / (self.__sizey*(viewport[3]-viewport[1])) - 1.0
self.__viewport_pts[2,:] = depth
# new world points (of the right size)
self.__world_pts = np.ones(self.__viewport_pts.shape)
def __vtkmatrix_to_numpy(self, matrix):
"""
Copies the elements of a vtkMatrix4x4 into a numpy array.
:type matrix: vtk.vtkMatrix4x4
:param matrix: The matrix to be copied into an array.
:rtype: numpy.ndarray
"""
m = np.ones((4, 4))
for i in range(4):
for j in range(4):
m[i, j] = matrix.GetElement(i, j)
return m
def render_point_cloud(obj, env):
start = timer()
dif.Modified()
pdi.Update()
end = timer()
print(end-start)
# ren, renWin, iren - vtk rendering objects
ren = vtk.vtkRenderer()
renWin = vtk.vtkRenderWindow()
renWin.AddRenderer(ren)
iren = vtk.vtkRenderWindowInteractor()
iren.SetRenderWindow(renWin)
# create cube
cube = vtk.vtkCubeSource()
cube.SetCenter(0, 0, 0)
cubeMapper = vtk.vtkPolyDataMapper()
cubeMapper.SetInputConnection(cube.GetOutputPort())
cubeActor = vtk.vtkActor()
cubeActor.SetMapper(cubeMapper)
ren.AddActor(cubeActor)
# set camera intrinsic params to mimic kinect
renWin.SetSize(640, 480)
ren.GetActiveCamera().SetViewAngle(60.0)
ren.GetActiveCamera().SetClippingRange(0.1, 10.0)
iren.GetInteractorStyle().SetAutoAdjustCameraClippingRange(0)
ren.GetActiveCamera().SetPosition(0.0, 0.0, 2.0)
# has to be initialized before filter is update
# not sure why
iren.Initialize()
# dif (depth image filter)
# Filter that grabs the vtkRenderWindow and returns
# the depth image (in this case)
dif = vtk.vtkWindowToImageFilter()
dif.SetInputBufferTypeToZBuffer()
dif.SetInput(ren.GetVTKWindow())
dif.Update()
# pdi (project depth image)
# Gets output of dif and projects the coordinates into world coordinate system
pdi = ProjectDepthImage()
pdi.SetRenderer(ren)
pdi.SetRenderWindow(renWin)
pdi.SetInputConnection(dif.GetOutputPort())
pdi.Update()
# for displaying the point cloud
mapper = vtk.vtkPolyDataMapper()
mapper.SetInputConnection(pdi.GetOutputPort())
actor = vtk.vtkActor()
actor.SetMapper(mapper)
actor.GetProperty().SetPointSize(2)
rgb = [0.0, 0.0, 0.0]
colors = vtk.vtkNamedColors()
colors.GetColorRGB("red", rgb)
actor.GetProperty().SetColor(rgb)
ren.AddActor(actor)
# needed for first image in loop to be current
ren.Render()
dif.Modified()
pdi.Update()
iren.AddObserver('UserEvent', render_point_cloud)
iren.Start()
来源：https://stackoverflow.com/questions/35440195/vtkpythonalgorithm-controlling-pipeline-execution