Hi,
I have data files in a 3D box having density generated by my hydrodynamic simulations. I wish to make a 2D projection plot on a plane through the box instead of a regular slice plot. Is there a filter or any script on Paraview with which I can make such a visualization?
Regards,
Alankar
Please precise. Ideally include a visual.
Hi Mat,
Sorry for the late reply.
I was trying to have an utility or a script in Paraview that would integrate the requested 3D scalar field f(\vec{x}) along a line of sight \hat{n} , given by the axis parameter (e.g. \hat{i},\hat{j} or \hat{k}) and a weight factor scalar field w(\vec{x}).
g(\vec{X})=\frac{\int f(\vec{x}) w(\vec{x}) \hat{n} \cdot d \vec{x}}{\int w(\vec{x}) \hat{n} \cdot d \vec{x}}
Thus create a 2D projection from 3D scalar field dataset.
These examples are taken from the yt python module which creates such projections
Projection along z-axis:
Looks like a job for a Programmable Filter.
Hi Mat,
Can you please refer to some article or share any example script that may be somewhat similar to this? Then I can work on that and modify the script to get my job done.
Hi Mat,
From the link you shared, I think the example code under Changing Data Type section is most relevant to what I wish to make. So I tried to first get that working. Here is the state for that.
filter-test.pvsm (370.6 KB)
After this, I tried modifying that code to get a simple yz plane projection (x-normal). I succeeded partially as you will be able to see if you load the following along with the data I’m sharing. Can you please give me any suggestion on this to get it working?
proj-filter.pvsm (491.0 KB)
Data:
OR
from my google drive:
Hi Mat,
I think I got this working on doing a very simple projection on yz plane (x-normal). I plan on to generalize to projections along the other orthogonal planes (can be done easily I guess). But to have it on other arbitrary planes require finding intersection of line of sight from each pixel in projection to data which right now I have no idea how to implement on Paraview Programmable Filter.
I have also got a query on will this method (you can have a look from the state I’m sharing) will be efficient on large datasets run on parallel client-server interface?
proj-filter-working.pvsm (490.9 KB)
I’m afraid that is not possible in a programmable filter, you only have access to the data.
Hi Mat,
What is your opinion about this as far as the Programmable filter I currently coded is concerned?
I have also got a query on will this method (you can have a look from the state I’m sharing) will be efficient on large datasets run on parallel client-server interface?
proj-filter-working.pvsm (490.9 KB)
Please share also the code of the filter directly.
Hi Mat,
Here is the code. I tried to make a vtkImageData of the filter. For comparing, I was also creating a matplotlib figure on the array which I’m creating for the vtk data. I get the correct expected image in matplotlib figure but not for the vtkImageData on Paraview (very strange behavior at the boundary). Can you please have a look at this and also comment on the efficiency on large datasets run on parallel client-server interface?
I’m new to vtk data structures. Any suggestion will be really helpful.
from paraview.vtk.numpy_interface import dataset_adapter as dsa
from paraview.vtk.numpy_interface import algorithms as algs
import vtk
import vtk.util.numpy_support as ns
import numpy as np
pdi = inputs[0]
verbose = True
normal = 'x'
if verbose:
print(pdi.VTKObject)
print('Data fields in the file: ',pdi.PointData.keys())
#get a hold of the input
numPts = pdi.GetNumberOfPoints()
dimensions = pdi.GetDimensions()
extent = pdi.GetExtent()
spacing = pdi.GetSpacing()
xmin, xmax, ymin, ymax, zmin, zmax = pdi.GetBounds()
#create the output dataset
ido = self.GetImageDataOutput()
resolution = None
if (normal=='x'):
resolution = (extent[3]-extent[2],extent[5]-extent[4]) #pixel for the projection plot
resolution = (1,*resolution)
ido.SetDimensions(1,resolution[1],resolution[2])
ido.SetExtent(0,0,0,resolution[1],0,resolution[2])
ido.SetSpacing(0,spacing[1],spacing[2])
elif (normal=='y'):
resolution = (extent[1]-extent[0],extent[5]-extent[4]) #pixel for the projection plot
resolution = (resolution[0],1,resolution[1])
ido.SetDimensions(resolution[0],1,resolution[2])
ido.SetExtent(0,resolution[0],0,0,0,resolution[2])
ido.SetSpacing(spacing[0],0,spacing[2])
elif (normal=='z'):
resolution = (extent[1]-extent[0],extent[3]-extent[2]) #pixel for the projection plot
resolution = (*resolution,1)
ido.SetDimensions(resolution[0],resolution[1],1)
ido.SetExtent(0,resolution[0],0,resolution[1],0,0)
ido.SetSpacing(spacing[0],spacing[1],0)
ido.SetOrigin(*pdi.GetPoint(0))
ido.AllocateScalars(vtk.VTK_DOUBLE,1)
origin = pdi.GetPoint(0)
#choose an input point data array the projection of which is to be made
ivals = pdi.PointData['density']
ivals = np.reshape(ivals,(dimensions[0],dimensions[1],dimensions[2]),order='F')
weight = pdi.PointData['total_energy'] #np.ones(numPts , dtype=np.float64)
weight = np.reshape(weight,(dimensions[0],dimensions[1],dimensions[2]),order='F')
#All cells are assumed to be of uniform size in data
if (normal=='x'):
proj_img = np.zeros((resolution[1],resolution[2]),dtype=np.float64)
weight_img = np.zeros((resolution[1],resolution[2]),dtype=np.float64)
for i in range(resolution[1]):
for j in range(resolution[2]):
proj_img[i,j] = np.sum(ivals[:,i,j]*weight[:,i,j])
weight_img[i,j] = np.sum(weight[:,i,j])
proj_img[i,j] = proj_img[i,j]/weight_img[i,j]
ido.SetScalarComponentFromDouble(0,i,j, 0, proj_img[i,j])
if (normal=='y'):
proj_img = np.zeros((resolution[0],resolution[2]),dtype=np.float64)
weight_img = np.zeros((resolution[0],resolution[2]),dtype=np.float64)
for i in range(resolution[0]):
for j in range(resolution[2]):
proj_img[i,j] = np.sum(ivals[i,:,j]*weight[i,:,j])
weight_img[i,j] = np.sum(weight[i,:,j])
proj_img[i,j] = proj_img[i,j]/weight_img[i,j]
ido.SetScalarComponentFromDouble(i,0,j, 0, proj_img[i,j])
if (normal=='z'):
proj_img = np.zeros((resolution[0],resolution[1]),dtype=np.float64)
weight_img = np.zeros((resolution[0],resolution[1]),dtype=np.float64)
for i in range(resolution[0]):
for j in range(resolution[1]):
proj_img[i,j] = np.sum(ivals[i,j,:]*weight[i,j,:])
weight_img[i,j] = np.sum(weight[i,j,:])
proj_img[i,j] = proj_img[i,j]/weight_img[i,j]
ido.SetScalarComponentFromDouble(i,j,0, 0, proj_img[i,j])
np.save('D:\\proj_img.npy',proj_img)
import matplotlib.pyplot as plt
if(normal=='x' or normal=='y'): plt.figure(figsize=(16,32))
else: plt.figure(figsize=(16,16))
plt.pcolor(proj_img.T)
plt.savefig('D:\\test-img-proj.png')
#
# This part of the code is to be placed in the "RequestInformation Script" entry:
#
from paraview import util
pdi = inputs[0]
pdi = self.GetInput()
ido = self.GetOutput()
extent = pdi.GetExtent()
dimensions = pdi.GetDimensions()
normal = 'x'
if (normal=='x'):
resolution = (extent[3]-extent[2],extent[5]-extent[4]) #pixel for the projection plot
resolution = (1,*resolution)
util.SetOutputWholeExtent(self,[0,0,0,resolution[1],0,resolution[2]])
elif (normal=='y'):
resolution = (extent[1]-extent[0],extent[5]-extent[4]) #pixel for the projection plot
resolution = (resolution[0],1,resolution[1])
util.SetOutputWholeExtent(self,[0,resolution[0],0,0,0,resolution[2]])
elif (normal=='z'):
resolution = (extent[1]-extent[0],extent[3]-extent[2]) #pixel for the projection plot
resolution = (*resolution,1)
util.SetOutputWholeExtent(self,[0,resolution[0],0,resolution[1],0,0])
Also on changinng normals variable and hitting Apply makes the output on Paraview produce plots with garbage values. Then changing OutputDataType from vtkImageData to Same as input sometimes I’m getting the correct result (although with the boundary issues) but sometimes that isn’t working as well. This is seems wierd and pretty unexpected.
Here’s the state: proj-filter-testing.pvsm (608.2 KB)
Here as I was saying, you can see some nan-values (“red” in both plots) produced at the edges creating some strange artifacts not present in the numpy array.
One really hacky unelegant way of doing this is applying ExtractSubset to negelect the edge. But doesn’t solve this problem. As is done in the following figure where you can see that the edges are clipped off that gave problemtic Nan values:
Hi Mat,
Can you please comment on the boundary Nan values I’m encountering?
Also on the efficiency of this filter on large datasets (around 50 GB)
On parallel, your script will be executed multiple time on part of the data.
However, if you have efficiency in mind, you may want to explore the possibility of implementing a C++ plugin instead of using a programmable filter.
No idea for the NaN values, that would require investigation.
