Colorspace clarification....

(Breaking away from the topic quoted above…following a new thread that was emerging)

To understand this could you provide a link or some background …superior is a subjective non targeted assertion…I am sure there are criteria for the assertion I would just love to know them… I was under the impression that JCh was the “modern” LAB in simplistic terms…and was being used in blending and many of the updated modules??

4 Likes

OMG, are we still on that ?

CIE Lab 1976, do I need to add more ?

Color space aimed at remapping cone response in a way that makes the delta E uniform over the visible gamut ? Aka not designed for pixel pushing ???

Notoriously known for its blue skew ???

Unfit to HDR ??? Not even properly hue-linear perceptually-wise ?

And on the topic of delta E, did you ever look at the delta 2000 corrected formula to account for the failure of Lab to predict delta E (wich should be a simple euclidean norm if the space did what it was designed for) ???

It’s really hard to respect tech bros still swearing by Lab in 2022 after so many evidences, theoritical and practical, of its failure. Not sure if they got the news that the Vietnam war is over either.

7 Likes

Even CIELUV (1976) is better than lab, its blue isn’t so badly stretched out and skewed (but it certainly isn’t hue linear).

No one use Lab. It’s awful.

1 Like

In image processing no one use Lab for gamut compression because blue hue becomes purple, but lms based color space becomes pure black generally due to the negative luminance values.

It’s still the more efficient choice speed/quality wise for of sharpen, local contrast and denoise and it could be used before filmic because it doesn’t breaks anything.

Have you compared darktable’s rgb tone mapping with Lab tonemapping?
With RGB tonemapping :
Red is shifted to pink, orange is shifted to pink too, yellow-green becomes radioactive green and blue is slightly shifted to purple.

Actually the linear to pq conversion is broken and i have pointed this in the 2019.

So all modules that use that color space are broken.

Edit: I really hope that moving the old modules in the xyy color space is a joke :rofl:

But how is Lab is immune to this? Only through some undefined behaviour? LMS colour spaces are a step up over Lab.

Yeah orange does become ugly pink. But blue is even more purple skewed in Lab (and don’t try hue shifts in the blue region in lab)

I can’t see any any hue shift with sharpen, local contrast enhancements, denoise and tone mapping.

For gamut mapping probably the better choice is to use the hsi color model because is easy to detect out of gamut colors( values above 1 in the saturation channel) and it’s robust enough to not have negative “luminance”.
The hue could be restored with some lms color space.

However take this example

Darktable with new modern workflow


P1200121.RW2.xmp (5.8 KB)

Photoflow with hsi gamut compression and cie lab sharpen and tonemapping
P1200121.pfi (49.6 KB)

However as a side note i prefer the simple rgb tone mapping most of the time.

I really don’t care fighting with someone that lacks of any practical sense, just don’t say wrong things to darktable’s users.

This thread could be closed

Please could you clarify what these examples are meant to demonstrate? Are we comparing your ability to edit between the two applications or is there some objective truth that they illustrate?

We could observe in the photoflow version that:
1- unsharp mask in Lab color space before the tone mapping function doesn’t brake anything
2-with Lab tonemapping we have way less artifacts compared to the RGB ratio filmic in darktable.

I suspect if Lab and USM always produced obviously bad results nobody would ever have used them. That they sometimes (even often?) produce good results doesn’t disprove the assertion that they can fail or are less robust than the alternatives.

I have no skin in the game here – I don’t know the answer one way or the other but as a physics major I would assume that manipulations that respect physical reality (linear scene-referred) should be easier to construct and more “correct”.

I would like to see some objective evidence/discussions to convince me though. Perhaps even links to research/papers that make comparisons.

1 Like

For tone manipulation the best color spaces are:

1)RGB per-channel and perceptual color space with the perceptual color space slightly behind
2) cie Lab
3) RGB ratio

For sharpen (not deblur) we need to work on the luminance channel and in a log/gamma color space.
Linear rgb in this case is just an artifacts factory

For deblur i can’t say, but using only deblur as the main raw sharpener is wrong.

How have you arrived at these conclusions?

Are you serious? Show me a commercial raw editor that use rgb-ratio as the main method for tone manipulation, filmic it’s pretty much what Reinhard described in his paper in 1997 and only darktable use it.

All the proprietary lut (sony, Canon, arri…) for the log/hdr to sdr conversion are rgb per channel curves (plus gamut compression).

There aren’t papers to show how much linear rgb is bad for sharpen an image because is so obvious but you could find a lot of discussion on internet and everyone in the end comes with the same conclusion.

Yes i’m serious. You have asserted that @anon41087856 is wrong in stating that Lab is a bad colour space to use (and by implication that the entire direction that darktable is taking is wrong). When one makes assertions about objective reality (especially one with such a big implication to the future of darktable) it is normal to prove your case with evidence.

I (and @priort) have asked for objective evidence and instead you have just provided additional assertions without further justification. Now we’re on the “but everyone says so” argument and “it’s so obvious nobody needs to prove it”.

Anyway pixls is telling me off for too many replies so I’ll just sit here quietly and await some argument to convince me one way or the other.

5 Likes

There are a lot of discussion in this forum too, find it yourself

Try It yourself

You could import in gimp an unsharpened image and try the linear or gamma unsharp mask, the linear usm will have a lot more artifacts

1 Like

I am grateful for the exchange and the opportunity to learn something and to hear other opinions. It would be great if we don’t degrade into a Republican vs Democrat mentality…we can see how well that is working for the US…

I would agree with Chris that making any statement without a reference or clear cut example that can be used for discussion point might lead to a hollow debate rather than a fruitful discussion…

One example for me was this one…

When I look at the images I am not immediately aware how they demonstrate the points we are talking about ( I acknowledge part of this could be me being thick). The photoflow one looks like it has a lot more contrast…beyond that I was not sure how it was comparable to DT to support a theory or assertion. (Again likely I am slow on the uptake)

In the original thread for this photo there was a nice DT edit …that subjectively I thought was on par or actually nicer to these in this case and IMO which means really nothing…but the point is one edit in DT and one edit in photoflow don’t necessarily provide a clear demonstration of what is the point just like my assertion that the mentioned edit is “nicer”

EDIT: @age thx for your time and I see you have some suggestions in your post that came as I was writing mine…

Sadly yes, I like the denoise and local contrast module in darktable, but in my opinion it has become a slow, overcomplicated and the worse quality wise raw editor.
The recommended scene referred workflow it’s good only for very innatural jpeg export.