Oldschooljedi said:
g-force,

in your current script you write

"Function Contra(clip denoised, clip orig)
{# re-write of Didée's contra sharpening routine from TemporalDegrain.avs"

So do I need the hqdn3d.dll-plugin as required filter for this as described here?

No, no need for that plugin.

I have a new theory as to what is happening. I think you may have a newer version of MVTools than I have. And that isn't necesarily a bad thing. BUT, here's the problem:

I encoded my newest version of the code and everything looks awesome EXCEPT when Luke lights up his lightsaber in Obi's hut. There is an annoying blur around the lightsaber. So I did some experimenting, and found that if I did a more accurate (but slower) motion estimation, sure enough, that blurring was much less. But now, with that better motion estimation, the motion vectors are locking on to the DVNR smearing that is present. The most noticable problem area I have seen so far is the Greedo scene. On the original, one can faintly make out a ghost of Han when the scene switchs to Greedo. So now when I make the motion estimation better, the algorithm locks on to that ghost and re-inforces it, making it even worse. And this looks very similar to the screen shots of the scene changes that OSJ has posted.

Bottom line is, there may not be a way to have both good denoising and little artifacting due to the DVNR smearing. I mean, I'll keep working on it, but things are looking not so good at the moment.

This GOUT is getting worse all the time.

-G