Planar Tracker Stabilize --> Match Move Workflow?

[Crossposting from WeSuckLess]

So, I'm trying to do some pretty basic shot cleanup on a product spot (removing unwanted water droplets from a bottle).



The shot has a moving camera (simple roll rotation from above looking down on the product), so my normal workflow in Nuke would be to stabilize the bottle with a planar track, do the cleanup on the stabilized plate, then match-move it back to the original camera motion -- pretty simple setup with two transform nodes linked to a roto planar tracker doing all the work needed.

In Fusion I can do this with a normal point tracker by creating two Xf nodes linked to the steady & unsteady outputs of the Tracker and it seems to work fine, but generally for a shot like this I'd prefer to use a Planar Tracker to reduce jitter.

When using the Planar Tracker node though, the workflow appears to produce some artifacts around the edges of the image that I imagine are caused by a tiny difference between the outputs of the stabilizer and the match-move operations (which should in theory be perfect inverses of one another).

Here is my setup -- you can see the little black dots around the edges of the image that dance from frame to frame:



In HQ rendering you can see it's actually a barely missing fraction of a pixel where you can see the misalignment of the stabilize / match-move:



Am I missing something here, or are the stabilize / match-move outputs of the Planar Tracker node simply not pixel accurate?

System Information:
+ 2013 Mac Pro
+ OSX 10.11.6
+ Fu9.0.1
+ OpenCL [Off]

It is probably introduced due to antialiased edge that is produced in first stabilization.

That said, it is never a good idea to transform the whole plate. If you need to stabilize for paint work, only transform the patches and merge with original footage. Or better yet, do live cloning without stabilizing, so you minimize the blurring from transforms. In Nuke I would link tracker data to a layer, put all clone strokes/patches to that layer and use only full pixel shifts for clone source positioning. This will minimize blurring. How to approach this in Fusion... sorry, can't help here.

Thanks for the reply Hendrik --

Yes, I completely agree about not transforming the whole plate, no reason to incur a filter kick on the entire image. For these kinds of patches I only composite the cloned strokes themselves back onto the original plate, the initial stabilization is just to make working easier.

That being said, there are times when I need to paint right at the edge of the frame and I would like to find a workflow that doesn't create these artifacts (I think you are right it is probably antialiasing). In Nuke I have never noticed this issue, even when working with cubic filtration on transforms, etc.

Also, these artifacts are not created when working with the standard Tracker + linked Transform node workflow, only the Planar Tracker (which cannot be linked to Transform nodes the way the standard Tracker can be, as far as I know) -- possibly because of the nature of transform concatenation, but I'm not sure.

To overcome the edge issue you can extend the edge outward, one pixel should already do the trick.