Here’s the link to the latest weekly snapshot of darktable 4.1, https://drive.google.com/file/d/1YEOIksXECOX3ngK0V6OFZko04JStXhaX/view?usp=sharing. The list of latest changes is at Pulse · darktable-org/darktable · GitHub
3 Likes
@wpferguson I noticed in this build that the highlights reconstruct color option has gone. Can you explain or add information as to why? I am not sure if this is a negative factor, but I just noticed it was missing. Sometimes it has been the best option for processing in the past.
BTW, I really appreciate your weekly builds. What would make them a little better for me would be a heads up what is new and improved. I do look at the github link you post but it is not very helpful for me usually. Not a complaint, just a hopefully constructive comment. Again thanks and have a good weekend.
See the discussion here : Further work on highlights reconstruction by jenshannoschwalm · Pull Request #12623 · darktable-org/darktable · GitHub
Thanks for this information. I have just read the post. I am not sure if removing reconstruct in color is good or bad since I don’t have the option to test it anymore. I probably would have preferred the option to stay unless there was a maintenance or coding problem. But I am not gong to whinge to the developers over this issue.
Not quite true, if you have used it in the past: you should still be able to create a style or preset using that algorithm from an image where you used it. But have you test it in the past?
I’ve never been able to get a correct result with ‘reconstruct color’, it gave artifacts every time I used it, and with whatever demosaicing I tried. So I’d say the only reason to keep it (*) would be: it is possible to get correct results with it, and the new methods are all significantly slower.
(*) That is, keep it as an available choice for new edits. Like the deprecated modules, it should stay to maintain compatibility with old edits, unless/until there’s a way to convert to a newer method without changing the result.
I already noted on this in the macOS build topic: the Pulse link is useless as it depends on the instant in time when you click it, not when @wpferguson posted it…
@wpferguson @MStraeten May I suggest you use the “compare” link instead, which is fixed? You just need to update the (short) commit hashes between your last and new build every time. For example: Comparing c640f33...6e5bd4d · darktable-org/darktable · GitHub
1 Like
I don’t think I’ve ever got good results from Reconstruct in Color. Its been spoken about as broken for some time. I support hiding it, fwiw.
1 Like
I find reconstruct in Lch has been a good option for me, but I have on odd occasions had good results with reconstruct in color. An example would be where you have an undesirable clipped highlight on a colored object such as a bronze statute and filling it with the surrounding colour is a quick and easy fix.
I will live with the developers opinions on what is best. I was just surprised it had disappeared.
I’ve actually taken tonusong color balance rgb mad just tinting it whatever color is good, then painting in a mask.
1 Like
reconstructing highlights is one of my hate jobs in editing because you are always behind the eight ball. However, clicking through all the options and seeing which works best is one option I use. Lch is one of the most reliable to recover the texture if all the channels are not clipped and reconstruct in color can be useful when just inpainting the surrounding color helps. I am currently working through a video by AP were he demonstrates his guide laplacians approach which has never worked in my hands. Sometimes I play with highlight reconstruction in Filmic as well.
I am just unsure why they felt the need to remove an existing option, even if they personally did not like it. I just have to trust they had good reason.
Not remove, just hidden in the GUI. If you find value init, more power to you, make a preset.
If you catch AP in this video …he actually states that with the current pipeline that those methods are not compatible…At the time he made that video all there was was the LPHR and this as he shows when you watch is in his words meant for specular highlights or to restore a gradient to be further corrected with filmic HLR. Now we have inpaint and segmental and they seem to do a nice job… If you use bloom in filmic and set the mask right it will end up resembling reconstruct in color if the areas are truly blown…at least that has been my experience… I think v6 gamut checking is making it further cute to try and correct these. I think I am migrating back towards v5. I seem to get it to handle better … of course I will end up on an image that convinces me otherwise I am sure… 
But unless I find an image I have previously used it on I am unsure how I can find it to make a preset. Any suggestion? Thanks.
Its still in the current stable release. You can apply it using that version then load the xmp, and make your preset.
Have you try the new inpaint algo?
Before responding to your question I found an image of a person with clipped highlights on the forehead. The inpaint option did an excellent job on that picture. 
1 Like
Now find a picture that you think the reconstruct in color algo works with and try to use the inpaint.
Well let me comment on that because i was the one suggesting to remove it. The three of us working on dt HL reconstruction (@Iain @garagecoder and me) did a huge amount of tests and validation of the new algos, we didn’t find a single image where “reconstruct in color” (RIC) was performing at least as good as “inpaint opposed”. The second argument to remove RIC, in the vast majority of cases it produces lots of artefacts, these artefacts are a) not predictable and b) depend on scale / zoom in darkroom so you never know what you get while exporting. So a situation we decided to mark the algo as “deprecated”. There also have been github issues about this, it’s just impossible to fix.
“Deprecated” means - yes you must be able to use old edits, the algo will be there but not for fresh edits.
4 Likes
Thanks for taking the time to respond to my comment. If I every find a previous image I have used RIC with I will make a preset to retain the option for future images, but if we can never be sure what we will get on export then I see why you chose to remove it. I want to directly compare one day the inpaint and RIC. Are they fundamentally aiming to achieve the same sort of fix? I believe I liked the RIC effect on some sunsets and reflections on bronze statures and jewellery. I have been most impressed with Lch for clouds in the sky and unavoidable clipping in high dynamic range lighting.
Thanks for all your development efforts with DT. I really appreciate what you all do for us.
You can still get that with the actual module…ie color reconstruction and its a bit more configurable so really you have not lost much and as other have said, RIC as you call it looks different at different zoom levels and often has weird banding and streaks in the final result so its not good …and finally as AP mentions in his video the older methods are based on WB ie the notion of HL going to white where as the modern wb uses a CAT to produce or correct for an illuminant and that is not white so these method are not really compatible… at least badly explained by me this is how AP explained it in his video…