It is an interesting problem. I tried a variety of programs with both versions, the gradient only and the gradient with noise. Their sample is better than any. I cannot duplicate (after a couple hrs) their results yet, which shows my lack of understanding of their parameters.
Their Deconvolution sample is from a new, yet to be released
, version of TGV Restoration. I have not been alble to test it either, but it's supposed to be released soon (the latest PI release had specific updates to facilitate that addition). What is not clear to me yet is how well small detail survives. Edges are perhaps simpler than small detail and they did mention slight issues in the sharp corners of the target, although they probably did test how well small point sources of light (stars) and planets survived. They're mainly doing final stability testing now, as far as I've read.
IP also has an annoyance that it will not open tifs that have a profile embedded. I have to re-save as no compression, PC encoding, no ICCs.
That's strange, because I can open TIFFs with compression (LZW or ZIP) or uncompressed, with an embedded profile. Maybe it has to do with layers? Otherwise I'm sure the PIteam would want to have a look at such a problem file.
IP's sharpening does shift luminance AND increase color saturation. I either have to strip the luminance, then re-combine, or send it back to RT as a tif to check colors again. There is a good chance I will just go RT to PI then output.
That doesn't sound like how it should handle sharpening, I assume you are referring to Deconvolution. Maybe it's related to the above TIFF issue? I've had such a color issue with an earlier release when using the CIE Y deconvolution (not in RGB/K components), but after I reported it they quickly fixed that bug that had crept in when they recoded some of the routines.
I agree that RawTherapee is also a very good converter, but having good alternatives is only positive. We get to pick and choose what suits our workflow best. And who knows, perhaps some of the advancements in PI can be replicated in RT ...
P.S. I have also detected a deconvolution issue, but only when Luminance CIE Y was the selected target. I've just (a few hours before the weekend) reported it as a bug
on the PI forum, hopefully it will be resolved soon. The same bug didn't happen on all files, so maybe it has something to do with a specific combination of file parameters coming out of Photoshop.
P.P.S. There may have been a change in the default color management settings, perhaps during an intermediate update. I'll have to check further to see what exactly changed the previous behavior.