I was really talking in general terms. The new filmic system while providing for many options is in my opinion a-real-winner in that it optimizes our process with minimal fuss … the t-e should be, in the same vein, a simple tweaking tool despite having also many options.
Currently, I feel that hunting through the options to find the best fit in the masking does not tie in to the filmic philosophy.
Posting an image does not solve the overall question/problem.
To prevent user from hunting, we need some automated process. To automatize a process, it needs to be fairly basic and perfectly known, hence predictable. Tone EQ is a versatile tool because it is very simple in what it does (apply exposure compensation depending on input luminance), but then it can be used in numerous ways (add or remove contrast, as a 1D tonecurve or as 2D local tonemapping). At this point, I have no way to automatize anything, it depends on too many variables.
The mask contrast depends a lot on the guided filter surface smoothing. Problem is there is no way to predict how much the 1D contrast needs to be enhanced depending on the 2D surface blurring parameters in use, because the guided filter adapts to the image features and that will change from picture to picture.
So, so far, I can only say to users : if blurred mask lacks contrast, add some more ; if blurred mask is too bright, remove exposure. You get an histogram and a box-graph of the histogram center 80% as scopes to help you. There are only 4 settings for the mask post-processing :
- size, to define the scale of the features you want to surface-blur,
- feathering, to define how much you want to tape the mask to features edges,
- exposure, to center the mask average brightness in the settings range,
- contrast, to center the mask brightness span in the settings range.
But I fear that massaging the mask view will only help hiding bad settings.
Filmic is more simple because it only deals with a 1D tone curve, so there is no need to predict interactions between 2D blurring and 1D contrast.
2 Likes
My flow there is usually to go straight to the ‘masking’ tab, click auto exposure compensation and auto contrast enhancement, then adjust them manually (you get a preview of the range covered right under the ‘mask post-processing’ header, with orange-ish warnings if there’s much clipping), then I go to the ‘advanced’ tab to check the histogram and then it’s a back-and-forth game between the masking tab and advanced tab, trying to make sure the mask histogram covers more or less the full range (BTW, @anon41087856, I think it’d be nice if one could simply adjust the contrast enhancement and the exposure fix e.g. with scrolling the mouse with some modifier, or to repeat those sliders under the histogram for direct access; and yes, I also know it’d be nice if the day lasted 50 hours and / or you received more support – I do what I can about the 2nd one, can’t do much about the length of the day).
Also @anon41087856 : it seems that currently (some? all?) modules that come after tone EQ are applied to the mask if you switch to mask display, including any masks (I saw the effect of a drawn mask, but can’t remember if it was with local contrast or with contrast EQ). This includes filmic, which is enabled by default in the scene-referred workflow, so you need to turn them off if you really want to see the mask, you need to go to the ‘active modules’ tab and turn them off, and then back again.
Yeaaaah, more non-standard Gtk widgets in GUI \o/ 
I will see what I can do to make the tone EQ mask use the general dt mask API (which bypasses image operations).
Seems like I have work for at least the next 2 years…
It must be me but I have neither ‘auto’ settings on my system that I updated from git one hour ago.
I meant the picker / eyedropper icons. I have to do more adjustments more or less all the time, but it’s a starting point. Screenshot here:
OK … now understood … thanks
I wonder if this is the issue I sometimes get with the tone equalizer. After setting up what I think is a good mask covering the whole range of my image’s tonal values, I hover over the darkest parts of the image and it shows on the advanced tab as somewhere in the middle of the histogram, and other brighter parts are very close to the darker parts on the histogram. In other words, it sounds like a contrast problem (or lack of it), but adjusting the masking settings doesn’t seem to get me anywhere. Of course, it might be that I still haven’t mastered the art of masking though…
I really like the idea of the TE and have had some good success with it, but I agree with OP that masking could still be made easier. I’m afraid I don’t have any constructive suggestions though. I would be interested in seeing a false colour or posterization implementation…
No, this does not apply to how the mask is applied, only how it is displayed.
Since the filter applies smoothing, it means that a bright spot will alter the local average, brightening darker areas around it (and the dark areas lowering its brightness in return).
You can play around with the edges refinement to make the filter follow edges more exactly. Or, you may be better off with manually adding masks to exposure (that’s kind of like what tone EQ does).
Turn on mask display (but turn off the modules that come after tone EQ in the pipeline) to see your actual mask.
1 Like
That kind of statement is unusable for me as a designer, it’s too broad.
Please, let the solutions (false colors, posterization, whatever to massage the mask) for later. The first stage of design is to accurately frame the problem and split it down into elementary, specific and achievable tasks. For this I need to understand at what stage the problem is: masking internal algo, GUI, wiring between both, or user education/misunderstood expectations.
From specific problems, I can derivate specific solutions. But making things easier – although a valid user expectation – doesn’t bring us anywhere closer to a solution. There is no “make things easier” handbook.
Please provide specific examples : what initial image do you have, what did you do, what did you expect as a result, and what result you got. From there, I can do my job. Otherwise, it’s just empty goodwill.
3 Likes
I understand, I just wanted to add my thoughts to the thread, and I’m aware that it may simply be that I haven’t yet mastered the tools already at my disposal.
But I’ll share a file here that may show some of what I’ve experienced. This photo is not great but has some tonal variety in the vegetation that I wanted to play around with. After leaving the default scene referred options alone (exposure, filmic, etc.), it is already decently exposed, if a little flat. If I then enable the tone equalizer, go straight to masking and click the eye droppers for both exposure comp and contrast comp, I get a good distribution across the histogram. But turning on the mask, you can see that the vegetation is pretty much all masked in the same tonal range. Hovering over a dark hedge puts the line somewhere in the middle of the histogram, and it doesn’t change much when hovering over a lighter piece of foliage. In fact, I even get some inversion where some of the lighter leaves show up as -6 EV and some of the darker ones at -4 EV.
This is obviously a mask contrast problem, but increasing the mask contrast comp quickly results in the mask clipping. I end up juggling all the sliders and getting some improvement, but it’s finicky and I have trouble getting much contrast between foliage tones. I guess this is what I meant by “could still be made easier”, which I realize is not very helpful.
So, is this simply user error and/or my expectations wrong?
IMG_5488.CR2 (25.4 MB)
This is what I get using your process (and disabling filmic, that is applied later in the pipe on top of the mask – so maybe that is the problem):
(For reference, not disabling filmic gives this:)
The mask does mostly what it is asked : splitting features (sky vs. ground). The overblurring of the dark areas is a side-effect of the guided filter, which is not exposure-invariant, and as such will blur more the lowlights than the highlights (that’s a math problem with the internal variance computation, algorithmic fix is on the way with @rawfiner and needs testing/refining).
Meanwhile, if you want less blurring, hence more details, you will have to tape the mask more closely to small features (using higher feathering value) and perhaps brighten it a bit:
Now, the problem is the sharper the mask is, the more local contrast you may remove from the picture, with contrast compression settings like:
Actually, having a blurry mask is the guaranty that local contrast will be preserved, since the exposure blobs will get the same exposure compensation.
For example, this is the result of the default mask feathering = 10 that gives the above blurry mask:
And now, if I push the feathering all the way to 100:
(here is the corresponding mask for reference:)
So, again, I’m not sure if this is understood : a blurry/blotchy mask is actually better preserving the local contrast of the picture than a sharp/detailed one. Having details on the mask is usually bad news, except if you plan on increasing contrast through the tone EQ, in which case it will also increase the local contrast aka acutance aka perceived sharpness – but you might consider ditching the guided filter (details preservation) at all in this case and use the simple 1D tone curve approach.
Then, having balanced the sky exposure vs the ground one, you can use filmic to restore perceptual contrast:
4 Likes
Many thanks for your detailed reply.
I do understand this. I think my expectation is to get a bit more variance in the shadows (blurry, but distinguishing between the lighter and darker shades of foliage). Something a bit like this, which I’ve hacked together using the soften module just to show differences in foliage tone:
Perhaps this will change a little with the refinements you say you are working on with @rawfiner.
Thanks again. It’s a great module. I think I just need more practice with it and to alter my expectations.
Ok, so if it is the result you are expecting, indeed fixing the variance computation to make it exposure-invariant should be the key.
2 Likes
Hi all, @anon41087856,
I’m posting in this thread, because it’s linked to my issues with the ToneEQ & the masking tool … and I wonder if I’m not in the same situation as the OP.
To my great disappointment, I was struggling yet again with the “Tone EQ” module Aurélien indulged us with … and once again foolishly checked (old) the “Zone system” module.
Correct me if I’m wrong, but I believe the “zone system” (not the module, Ansel Adams’s) is the foundation of Aurélien ToneEQ.
And it looks like the “Zone system” module is trying to achieve the same as what Ansel Adams proposes, as the image is split in x (10 by default, an be customized) zones by luminosity/lightness. Is that correct ?
I now know that Aurélien’s ToneEQ is working in a Linear space (in RBG, I think), while the Zone System is in Lab (so in Log, if I get things properly).
The thing is, I find it way (repeat) easier to achieve the desired result with the Zone system module than with the ToneEQ.
And I think it may be due to the “preview” that the Zone system module has
Aurélien,
- have I misunderstood or overlooked anything in my above sentences ?
- is there something inherently different between the 2 modules, i.e. neither from a space (linear, log) nor from an algo one, but from the tool/module intent perspective ?
- Once thing that tickles me is that, in ToneEQ the user sets the exposure correction of the selected (Ansel Adams’) zones. But in the “Zone system module”, it seems the user chooses the zones he want to give more space to (i.e. more EV) or less space to( less EV) … by keeping the same spectrum of lightness from 0 to 1 (or 100%) … I thing I get that the “compression/expansion” of a zone to the benefit of another could be seen as the slope use from one zone to another. I’m not sure this is correct …
- is such a previewing utility something your ToneEQ could (theoretically) be using/proposing ? I’m always struggling with the Masking tool. Always
Thought I think I watch all your videos.
Many thanks … and don’t hesitate to kindly indicate if anything in the above is badly stated or if one of the questions is unclear ?
Guillaume
2 Likes
No. First of all, the zone module of dt works in Lab, where L is the cubic root of the luminance Y, and splits the display-referred range (from 0 to 100%) into as many zones as requested by user. Meaning the width of the zones changes.
Adam’s zone system relies on log2(luminance), which is EV (each zone has 1 EV of width). Tone EQ process pixels in linear but masks the zones in log2. So it’s consistent with Ansel Adam’s and analog darkroom legacy, and the zones have always the same width.
But the main difference is Tone EQ works in scene-referred, between almost 0 and + infinity. That’s why the masking needs to be adjusted to the current dynamic range.
The other difference is the zones module doesn’t blur surfaces and doesn’t preserve local contrast. That makes the (invisible) mask behave in a more intuitive way, but may flatten details quite a lot. However, local contrast preservation is optional in Tone EQ (although enabled by default).
Zone system module doesn’t care about geometric zones, actually. It’s a different way to produce a tone curve. What Tone EQ does is closer to analog dodging and burning : build a geometric mask by splitting the image into luminance intensity, and assign an intensity change to geometric regions selectively.
The mask is the preview, there is nothing else to show. Tone EQ shows the mask as used in the processing. The zones module shows a lie, the nice blurring and segmenting doesn’t match anything done by the algo and is only a GUI thing. If you configure the Tone EQ masking properly, you should get a preview that matches the zones module.
Again, a bit of history will put things in perspective :
Tone EQ is only an exposure module where the mask is built in an automated way. Nothing more.
4 Likes
The problem that I typically face is trying to develop ‘texture’, mainly in the highlights (for instance snow or clouds). I have 'till this time relied on the use of shadows-and-highlights, which does work but can also be fraught with problems (edge artifacts). I find it frustrating that t-e, that is such a fine tool, will not allow me the precision of making this kind of adjustment.
The t-e interface is really superb, is there any way that it can be more precise like a scapel rather than a blunt brick?
I already answered that many times. Either disable the details preservation, in masking tab, or increase the feathering parameter by a lot (which should end up the same if > 100).
1 Like
You are right … using a guided mask and feathering set at 125 I was able to perform substantial cloud shading.
A few things noted: It would be nice if the feathering scale was extended to some reasonable level so that this could be a more ‘normal’ usage. The mask does not reflect this more subtle shading so is less useful. Lastly I tried to restrict the cloud shading with a parametric mask … the mask worked fine but problems were indicated.
I should have said I am running the daily git.