Sorry I’ve been a little too fast to comment on that, I’ve not compares since inpaint opposed have been introduced but now that I’ve said that I’m testing it right away !
Having checked re opening old raws where I remembered struggling with highlight details I compared with RT playing with highlight compression and exposure on both softwares and I can say :
The inpaint method compares to old edits effectively brought more details out of the highlights
I could not find significant differences in the amount of details I could recover from the highlights
I totally retract that part of my previous post, I’m sorry as it was rather inconsiderate from me of the work being done and results achieved.
With sigmoid and inpaint opposed (as mentioned by @hannoschwalm) most of the “deficiencies” of dt are gone imho. It’s now mostly workflow and ui differences.
When sigmoid came along I was impressed enough to consider switching back from RT to dt. For most photos handling dynamic range becomes very easy and fast with sigmoid. Filmic should be avoided imho unless you really know your stuff, judging from play raws and showcases a lot of people still fail to take care of the pink issue.
RT still has better sharpening, pixelshift and the highlight recovery still works better. I find that I often have to switch between inpaint opposed (which RT has gotten from dt I believe) and colour propagation to find what works for a certain photo.
Personally I try to use DT periodically because there’s a lot to like in it on paper but when I compare the final result with RT edits I often find the RT ones better (although for some pictures I prefer DT), not sure exactly why. I think part of it is that the colours are better in RT our of the gate for my camera, but even for things where DT should have an edge like use of masks I find that RT’s local adjustment hold their own.
“Personally I try to use DT periodically because there’s a lot to like in it on paper but when I compare the final result with RT edits I often find the RT ones better (although for some pictures I prefer DT), not sure exactly why. I think part of it is that the colours are better in RT our of the gate for my camera, but even for things where DT should have an edge like use of masks I find that RT’s local adjustment hold their own.”
I read in a thread in the forum on dpreview that such problems are due to the scene-referred workflow, which would only have disadvantages apart from HDR on screens. I don’t have the knowledge to judge this statement, but it makes me think.
If that statement comes from who I think it did, I would suggest disregarding it. He clearly has some sort of irrational hatred of DT, that seemingly means he can’t say anything nice about it. And several of his claims are provably wrong.
Without searching, let me guess: charlyw64?
He is the guy who, whenever a new darktable version is up in the DPReview news, says it’s unusable and best avoided – and still, he comes back every time and spews hatred and lies. @stevie : yes, ignore him, really.
But this topic is mainly about ART, so:
Surprise, surprise, ART also has a scene-referred workflow. Maybe charlyw64 is wrong?
(quoting @lphilpot) Yes, darktable has a linear / scene-referred workflow.
So has ART, fwiw. If you didn’t notice, it’s because of the “what” Vs “how” I mentioned above. (ART vs Rawtherapee - #17 by agriggio)
CLF LUTs are […] well suited to be applied in a scene-referred workflow, e.g. to implement “look transforms” or also “output transforms” (i.e. tone mapping operations from scene to display/output) (ART and OpenColorIO - #14 by agriggio)
a clear separation between tools operating in “scene referred” space and tools operating in “display referred” space, similarly to what is happening e.g. with darktable (ART (the software) news - #235 by agriggio)
And, if I understand correctly, so has RT, or at least the features are rather similar (@jdc , please correct me if I misunderstood your references):
innovations in DT (scene refere, Sigmoid, new color science,…) are hotly debated on the forums and that similar processes on RT (Ciecam, Sigmoid, Munsell correction, Auto WB…) are currently in the shadows(Devs - How Hard Would It Be To............? - #29 by jdc)
I also revised the GUI to make it more consistent, notably by including Sigmoid and Log encoding in Cam16 image adjustments. […] Note that the notions of “scene conditions (reference)” and “viewing conditions (display)” have been present in RT for more than 12 years. Local adjustments - Cam16 and Log encoding improvments)
That sounds very interesting and maybe worth it to help those with potato computers? I mean, only if you wanted to do it because it sounds like you were doing the work on that. If you don’t feel it was worth it, no problem! But the Diffuse or Sharpen module can’t be used by all Darktable users…
So, I know this is controversial for some, but if your not interested in pixel peeping, just ignore it. I’m mainly writing this, because I noticed interested into the topic in other threads.
I did print the Lago di Garda image in larger size (60x40cm) on a nice paper, and I’m pretty happy with the result. The details a really shining.
Here is my final call regarding the diffuse or sharpen settings. I did watch APs 2h video again and did a revised preset for very fine details (see below). You can compare the results in these 100% crops. I show the ART version, the original version with the standard sharpen demosaic no AA filter preset, the version of @123sg and my refined preset (which I’m happy about, because IMO its at least as good as the ART version). Compare for yourself, but be sure to watch in full size.
For those interested in the DoS settings, here is a screenshot:
You probably can’t apply these settings blindly. Must tune them per picture, mostly with edge sensistivity and edge threshold, I think. If to strong, reduce iterations. Also combines very well with the detail threshold in the parametric mask settings.
Did you apply your preset to the same base as Steven’s? Somehow, I think I see not only sharpness but tonal differences, which then also influence perceived sharpness.
I’m actually quite liking DoS and even just using the 4th order speed to sharpen works quite good. I did notice if I use Contrast Equalizer and in Luma bump up the center section a bit on long telephoto shots of say a city and buildings it really messes up the sharpness of the buildings.
Interesting. I’ve pasted the “noAA” image over Steven’s and yours, and set them to difference mode. Yours shows more differences (differences emphasised using the Levels tool):
Yet, I perceive Steven’s as sharper, more detailed. Or was your aim to replicate ART’s Capture Sharpening as exactly as possible? In that case, you did a very good job! Differences emphasised:
This is interesting. I actually perceive@qmpel’s new version as more detailed, although I don’t think there’s much difference in actual resolving power, so as to speak. Possibly finer detail would be more descriptive.It also looks cleaner, less ‘artifact-ish’.
This is viewing at 100%… that makes a difference to, obviously.
I was curious so I tried that preset proposed by @qmpel on a recent photo shared of the bee.on the flower… earlier in the thread and it wasn’t the best out of the gate… I was in a rush this morning and just took a quick look so I should check again… But I have found just recently that I use an approach not like I have been before where I basically accepted the preset and then tweaked the iterations for the most part… I now feel that you can essentially apply a strong effect… say alternating -100 -50 -100 -50 settings or something like this ie quite strong to sharpen and then its very easy to dial it down and tweak it to a very nice result by using the two edge settings and opacity… I have also started to put an instance of astrodenoise in front of D&S so that I can tweak a bit of the grain/fine noise before D&S see’s the data…You can also try the other way for sharpening and just use a small effect on the sliders but slowly drop the threshold slider to a negative value… This has a strong impact on edges and might be more bang for the buck than bumping iterations… Finally the details slider can be used if you enable the parametric mask for even more fine tuning… so there are for sure a lot of small things that one can try to get the look that suits the individual…
Yes - it’s not the effect that’s generally accepted as “sharpening”. I would call it a deblurring of the finest details (won’t get much smaller than a radius of one, I’d think).
So, I can understand @kofa , if he perceives the other version as “sharper”. I wonder how it looks like on a 4K Monitor, myself viewing on 2560x1440 with relatively large pixels. Will probably have to upgrade sooner or later And yes, @kofac I did try to mimic the ART way closely, as for my taste it provides a very fine level of detail…
I think that’s a pretty solid way to go. Just leave the grain level out of the sharpening process. It’s more or less what ART does, i think (the contrast threshold slider for the capture sharpening).
And yes, 
)