(The following information comes courtesy of one of the research engineers at the Foundry).

The Card node implements the PtGUI lens model.

The formula used by the Card node is:

r_src = a*r^4 + b*r^3 + c*r^2 + (1-a-b-c)*r
* r_src is the radius between the source pixel and the center of the image
* and r is the output radius from the center of the image
* where the radius is scaled so that width/2 is 1
(unless you toggle "lock to vertical" where height/2 is 1)

So the parameters should map *unless* PTStitcher is calculating an optical axis that is not at the centre of the image.

You may need to use "lock to vertical" if the smaller side of the image is vertical. You may also need to check that Nuke has the correct pixel aspect ratio for the format if it is not square.

The only issue is the assumption that d = 1-a-b-c. The effect of this is that the mid point of one side will be locked in place. Connect a checkerboard to a Card node and adjust the b parameter - you will see the mid-point on left and right do not move.

The work-around for this is to use the scale parameter on the Card to effectively set d. I believe it works out as scale = (s, s, 1) where s = 1 / (a + b + c + d). If you look at the NukeX LensDistortion node, the translation to "Card parameters" includes this scale to correctly reproduce the lens distortion.

If you can get PTStitcher (or whatever software you use that exports a .pts file such as AutoPanoGiga) to apply the lens distortion it has calculated, you could apply that to a checkerboard image then put that into the NukeX LensDistortion node. This will calculate the lens distortion again using our model and translate it to the "Card parameters" - that way you have a sanity check to see what the values should be for the Card node.