ART goes towards 1.0

Final quality. Ordering is important, as @anon41087856 has waxed poetic; indeed, my first “exposure” to darktable’s documentation included some statement about how the ordering was pre-deterimined, fixed, and not changeable to save us from ourselves. LR doesn’t even tell you that; they just do it and you accept it.

Now with 3.0, you can change the order of operations, see what happens. I applaud the change; with my hack software rawproc, I’ve had hours of fun stitching together tools in orders Mother Nature never intended, doing egregious damage to images, and learning in the process. Now, you can too in dt.

Yes, the math can be daunting, but there’s some worth understanding if you want control over your images. The whole tone curve thing is one, IMHO.

Hi,

I understand, and I’m trying to keep it as simple as possible. In this case, I couldn’t come up with an analogy that was convincing and not misleading… However, perhaps this can help:

Just play with the sliders and you can see what happens in terms of a (hopefully more familiar) tone curve…

More replies later during the day

3 Likes

A quick comment on this. Ordering is important in the implementation, but from the user’s point of view (and speaking about ART, which is the topic of this thread… :slight_smile: ) the order in which you perform the operations doesn’t matter. ART (like RT) always applies the operations in a fixed order, that is not controllable by the user. So, I’d say just tweak the tools in the order you prefer, the result will always be the same.

Clearly, doing similar things by applying different tools (e.g. my example above of using CDL instead of tone curve) does have an impact on the output, because then the pipeline will be different. But that’s a different thing…

Similarly to the above, note that there’s no (direct) relationship among the placement of a tool in the GUI, the chronological order in which you apply the operations, and the actual order of the pipeline. The first is dictated essentially by my taste, the second by your preference, and the third by some considerations about math, color spaces, performance, ease of implementation, and other things you might not want to care about…

(As a final note: I am planning to add a table to the ART wiki with the order of the tools in the pipeline and other implementation details, for the interested people – but I don’t know when it will be ready)

2 Likes

Yes,that is why in the french translation I added between parentheses the term lightness and contrast after slope and power respectively in the UI of Color manager. With the help of @sguyader
we retain this “solution” .

Me, I’ve learned a lot about how things work by experimenting with order of operations. Indeed, I was able to explore the whole scene-referred thing back when we had those challenging discussions here by building toolchains in rawproc that I couldn’t do in the other softwares. But, I eventually came to recognize the invariance, especially in the early part of the toolchain, so I made a group tool to accommodate.

I am glad you’re working on an order presentation in ART; while most won’t want to change it, I think knowing it is important to dissecting things like, “Why doesn’t my image look good like it did in LR?”… :smile:

Thanks Albert, really appreciate your explanation and for bringing it back on topic. You’ve confirmed what I was beginning to doubt - that the order in which I choose to make changes does not affect the underlying order of processing in the pipeline. Although I didn’t quite follow what you meant about using CDL instead of the tone curve having an impact on the output, is this something I should know? In my earlier example of stretching the histogram, does it really matter for the final image quality whether I use Tone Curve, Log Encoding or Slope/Power in Colour Correction early in my workflow?

Why I use ART as my main tool is that I feel it has the power and complexity of RT and darktable, but it is more streamlined and is not too overwhelming. So I would say you have definitely succeeded in your mission. Thanks again for your work and your explanations on this forum.

Two more items on my wishlist for your consideration:

  • The option to include subfolders in the file browser, i.e. recursive search
  • A single reset button for each tool, i.e. to reset all parameters in the tool without doing each slider one by one. It could be next to the power button, or somewhere near.

And by the way, that geometry link you posted was great for visualizing what slope and power do. Is it accurate to think of it like a tone curve, or is it only a weak analogy? I’m wondering because if we think of it like a regular tone curve, it would seem that Offset can be used to push the blacks left and right, and Power gives a fair amount of control over shadows, but not highlights. Is that an accurate interpretation?

if that’s the only thing you do, then no, it doesn’t really matter.

yes, the graph shows exactly what happens…
the general wisdom is: slope → highlights, offset-> shadows and power → midtones. while there’s a reason for that, I’m not sure find it fully convincing. but if it’s helpful I could change the labels of the sliders to something like “slope (highlights)”…

Well I would then go on to do further edits if that’s what you mean, including color correction, sharpening, noise reduction, etc.

As for changing labels, if you think the text in brackets is accurate enough, then by all means. But if you think it’s not really accurate or will make the GUI cluttered, I understand. It’s your baby and I don’t want you just to customize it just for me. From that geometry graph, it seems it’s not as simple as just thinking in terms of highlights, midtones and shadows. Each of the sliders affects the whole tone curve, but in different ways and to different extents. Would you say the interpretation by @srgmro is better (slope=lightness, power=contrast)?

well, in the video world you would call them lift (offset), gamma (power) and gain (slope), though aces people might not be happy about that (because of the display vs scene referred editing stuff… plenty of threads about it if you are curious – but let’s not bring that debate here)

2 Likes

I would say that this is not an interpretation, but rather a way to memorize the main effect of the slider.

The does’nt seem equivalent at least from video SW company: An In-Depth Look at ASC-CDL Based Color Controls - Pomfort
With those interferring sliders, I wonder how video technicians transform the intents of artistic director into SOP parameters :thinking:

After such a treatment, the data are no longer “linear”, so I suppose it is rather located at the end of pipeline in the non linear part.
Should not be useful to be able to stretch the histogram near the beginning of pipeline by increasing black point and exposure? or compressing the histogram?

Until now it was simple, but @agriggio brought to us ASC-CDL in its full glory. To contemplate it, try mode: separate RGB channels instead of standard.
@agriggio in that mode, the global saturation slider seems missing. But you can defer that until we get our PhD in color grading.

Indeed, they might not be equivalent. That’s why they have different names, to avoid confusion. But their intent corresponds to lift, gamma, gain. See e.g.
https://theasc.com/ac_magazine/September2009/CASPart2/page4.html

In fact, it’s somehow in the middle (see below)

See here: agriggio / ART / wiki / Pipeline — Bitbucket

That’s by design. You can still manipulate saturation by manipulating the individual channels.

Thank you @agriggio for the pipeline description and all the efforts you do to develop ART and answer my uninformed questions.

I don’t think it is at all imprtant for us. That was just to warn you that it is present in the ASC CDL spec 1.2.

Thanks for posting the pipeline info.
Let me know if I should start another thread for this, but I’m still wondering if there is an optimal workflow to follow in ART? I realize that the order in which I do things doesn’t affect the underlying pipeline, but I’m still getting the impression that the final output quality can be affected by workflow.
Are there some general guidelines or should I not worry about it?

1 Like

if you have some trouble getting the results you want, feel free to ask questions and I’ll give you my opinion. other than that, I don’t feel qualified enough to provide general advice, sorry…

I’m not sure it’s the quality per se that is affected by the tools you use, in general. It’s mostly a matter of taste and artistic decision. Whether you choose to render an image flatter, contrastier, or with shadows crushed or lifted, etc… In the end it’s your choice. But it’s just my opinion.

Sorry if I wasn’t clear. I’m not looking for general advice on digital processing. I’ve been doing it for many years and I’m happy with the results I’m getting from ART. I was just wondering if there were some general workflow guidelines for ART specifically to ensure optimal quality (from a technical standpoint, not artistic).
I wondered if I was doing anything wrong by just following by own preferences with tool order. For example, if I chose to use Log Encoding before Colour Correction, would that be an issue.

To be honest, I think it’s all the darktable noise that’s confused me! Not sure if you’re aware but quite a few users are confused now with darktable 3.0 and module order.
Not to worry, thanks!

Thanks, and yes I agree. It was just the technical side of things I wasn’t sure about. There is lots of talk about scene-referred, display-referred, LAB, RGB, etc. all over the forum and some seem to have strong opinions about going from one to the other in the wrong order. Maybe it’s not an issue with RT/ART…

I think the only thing that is effected by the order you use the tools is how you see the effect of the next tool you use, but the order you apply the tools in doesn’t effect the quality of the output since the pipeline is still fixed.