I think one is the working site maybe?? @elstoc can likely confirm the difference if any in the twoâŚ
When 4.0 is release, weâll release a new URL for the docs. Both of what youâve linked are the same thing at different URLs: the master branch of the docs repo.
Going to have a play with this later, cause Iâm curious to get to grips with this.
I noticed that latest commits to filmic (or something else in the last few weeks ) seems to suddenly removed the need to use highlight reconstruction for me.
In areas which are clearly indicated as sensor clipped (and thus are magenta with no modules enabled and lowered exposure ), i needed to enable highlight reconstruction to get rid of the magenta a few weeks back . Now i donât often anymore , as if filmic sees that itâs extreme luminance and doesnât preserve the magenta .
Like flannelhead said , only in some cases do i still get a bit of magenta and start toying with highlight reconstruction.
I donât think so.
Certain norms (e.g. luminance Y) help, as does preserve chrominance = no.
I just had a quick try, and what - I think - already has been said in the topic:
Set preserve chrominance to ânoâ and things appear right almost directly.
Whichever preserve-mode I pick, filmic v5 doesnât do better (except ânoâ).
Luminance-Y comes close, but the trousers of the sailors get a bit blue, while they stay white / warmer with just set to ânoâ.
In the top-right between the leaves is also a bit of overexposed sky. When set to v6-ânoâ I donât have to do anything to it, not even filmic-reconstruction. In other modes itâs magenta and I have to grab the highlight-reconstruct module (didnât try filmicâs to be honest).
preserve-mode set to ânoâ does make the highlights really âpop outâ. The people in the middle-right with white shirts for example, they look really bright (blown out?). The captain with the full beard right of the tail of the fish, his beard also seems rather blown out. But changing preserve-modes doesnât really bring it back. Toying around with the tone equalizer, you can bring the highlights down and the shadows up to lower the contrast of the scene before hitting filmic, but that might not be the look youâre after.
(although dragging tone-equalizer before exposure (so under it) and picking one of the âsoftâ compression presets is not that bad actually).
âmax rgbâ really tries very hard to get any kind of color out of it. And if your highlights are really white (like the trousers here) it just turns them into some color thatâs not really correct.
Itâs not an easy one, I give you that. Makes me want to go back to some pictures and try some other preserve-modes :).
I posted a filmic v5 version (the 1st image, in fact), and I think it does a fine job out of the box. Try opening these in tabs, and switch between them:
ah, my dropdown boxes bug out with recent gtk versions, so once I click something it doesnât always âstickââŚ
using the scroll wheel to go through modes is more reliable⌠filmic v5 with preserve-mode set to ânoâ and filmic v6 (with no) are very very close, and both work well.
@flannelhead can you explain - to someone not knee-deep into color maths - why âpreserve chrominanceâ set to no is still hue preserving?
Am I correct in explaining it as 'it does not try to keep saturation, or âhow much colorâ, but it does keep âwhich colorâ, by not causing color shifts?
So the trousers have some blue in them somewhere, but with preserve-chrominance to ânoâ they are just so high in luminance that they appear white?
Chrominance is the colour correlate of the norm. If the colour of the source is incorrect to begin with or one modifies one without the other, then one would end up with funky colours.
Sorry, English is not my mother tongue.
Websterâs defines correlate (noun) as
1: either of two things so related that one directly implies or is complementary to the other (such as husband and wife)
//brain size as a correlate of intelligence
2: a phenomenon that accompanies another phenomenon, is usually parallel to it, and is related in some way to it
//⌠precise electrical correlates of conscious thinking in the human brain âŚ
darktableâs docs only mention chrominance to note that itâs not the same as chroma:
chroma is not short for chrominance, which is the color part of a video signal (the Cb and Cr channels in YCbCr, for example).
Those two donât help me understand what you mean.
I should have just stuck with component. In short, chrominance as it relates to the norm is image divided by norm. I am a very tired: someone please correct me gently if I am misspeaking on the matter. Essentially, chrominance is the direction component of the vector and the norm is the magnitude. If you know what I am talking about, it is easy to understand why a change in direction or magnitude would change everything, especially something as sensitive as colour.
Yes, there is that. The CIE definition is a value that can be calculated and used as an intermediary for other calculations. @jdc is the gentleman to ask. Just donât do it in this thread.
There is a nice video there taken from the perspective of painter in the digital realmâŚI think he defined it as chroma / max chroma possible for a given hue⌠outside the scope of my knowledge really but makes sense I think
Glad you asked. The ânoâ mode first applies the filmic curve to each channel individually. This causes a desaturation in the highlights and a hue shift. However v6 then restores the original hue (âwhich colorâ) but keeps the new chroma (i.e. âhow much colorâ). This is followed by gamut mapping to decrease the chroma in order to fit within the output color space. So I think your explanation is pretty accurate.
If I remember correctly, thereâs also an effect on the shadows, where the individually applied curves would cause a hue shift + oversaturation; and filmic v6 counters both (in contrast with highlights, where the (lowering) effect on saturation is retained).
Actually the shadows donât get any special treatment there, so filmic v6 with the no chromaticity preservation mode gets that resaturation of shadows. However, thatâs followed by gamut mapping so there should be no hue shifts in the shadows.
Thank you! And for comparison, how does v5 behave?
Has the behaviour described here changed?
- for the non-chrominance preserving method (independent RGB channels), we still enforce hue to be passed along from input to output. We still keep the desaturing behaviour in highlights but we set the original chroma in shadows to avoid resaturating.
Actually yes, you may be right I forgot thereâs also a function that changes the chroma a bit before final gamut mapping. Too tired to read the code now though, getting back to this tomorrow