Resample to Dataset terminology

Am I the only person who can never remember the difference between Input and Source when using this filter? I always have to look it up. Could we possibly relabel these as “Input Variables” and “Source Geometry” or would that make it too easy?

Irritatingly yours,

Dennis :wink:

You are not the first to bring this up (, but thanks for bringing it up to help raise the priority.

Changed target to this spring release, marked as required. Oh, Dennis, you are sooo very correct.

1 Like

In Kitware’s defense, I have just discovered the tips if you hover over the fields which clue us in!

How about a legitimate question about Resample to Dataset.

I am trying to take a cylindrical membrane surface and Resample it’s variables onto a copy of it that has been rotated.
So, Input is the original cylinder, which is discretized every 2 degrees, and Source is a copy of the original cylinder that has been rotated by 0.5 degrees.

The "Compute Tolerance has to be increased to 0.02 in order to get a reasonable distribution on the rotated geometry. What is this tolerance, is it a physical distance, and how do I determine a reasonable value a priori? What bad things would happen if I used a non-optimized value (i.e., too large of a tolerance)?

The Computed Tolerance must be very small because I get a very bad distribution on the rotated geometry (almost single value)

Thanks for any help.

1 Like

Trying to sample values from a surface is always problematic as an x, y, z coordinate will seldom be exactly on a surface. In fact, in your case, none of your coordinates will be exactly on the surface. When you rotate your cylinder 0.5 degrees, the resulting points will deviate off the chords joining the 2 degree samples in the original mesh.

Choosing a tolerance is always problematic. Too small of a tolerance (particularly on surfaces) and points won’t be identified in the cells you want them to be. But too large of a tolerance and you might be sampling values in a mesh where you shouldn’t be. I wasn’t privy to the decisions in deciding what the computed tolerance should be, but I believe it picks small numbers to error towards only sampling points truly in cells.

You will have to play with the tolerance until it gives you the results you like. I also found that changing the Cell Locator to Tree Cell Locator seemed to improved performance.


In my case, where I have a 1-to-1 correspondence of nodes/elements with slight displacement between the two structures, it doesn’t seem to change anything if I change Tolerance from 0.02 to 1000.0. I assume this is because it uses the closest point no matter what.

Since I am trying to write a general purpose routine which will be used on meshes of undetermined size and refinement, I think my only possible problem is having Tolerance too small, so by using 1000.0 I avoid this problem. I am always remapping onto an identical but slightly perturbed mesh, so that is my plan unless you can see a potential pitfall.

The name change has been completed in

Input is now Source Data Arrays and Source has been changed to Destination Mesh.