Evaluate the new sigmoid tone mapper just merged into master ...

[jorismak]

I’m not saying that filmic is awesome, that filmic rules all…

Not even saying sigmoid is good or bad or anything (really, I do see the appeal, and if more people are happy about it, then it’s a good thing, right?).

I’m just very very VERY curious to get a raw file where someone had the issue with filmic completely changing things like skin tone. And even more so if sigmoid doesn’t do the same.

I’m just curious, specially because I have never seen it. And because the differences between sigmoid and filmic are minimal from my testing, and the differences are there in how highlights are handled. So colour differences in different parts of the image, I’m very curious about!

[jorismak]

This is a good example where sigmoid does a certain thing that is sometimes very nice, sometimes not. Depending on your shot. For most sunshine / sun disc shots, it does something nice :).

The image with highlights properly reconstructed, and then just exposure, so low that the sun isn’t clipping. So no filmic, nosigmoid.

Just, to take a look at the ‘data’ of your shot:

You see how the sun ball suddenly becomes very bright? This is no tone mapping at all, this is how bright the scene was. (well, because the sun was clipped it’s a guess, I think this shot would be much better if it was taken without clipping anything).

Now, filmic with just a click on the ‘auto white’ picker and then changing to luminanceY mode:

And here sigmoid in rgb ratio mode with the skew a bit down:

Now, there are two ways at looking at the results. “Filmic preserves more of the data” and “sigmoid smooths more”. If you like on over the other, well, that depends on your expected outcome .

Filmic still shows the bright sun disc and the surrounds a bit. If you compare it to the un-tone-mapped data, I think filmic keeps more of it.
This is the ‘against sigmoid’ argument, that it smooths the highlights, rolls them off too smoothly or something, causing to lose highlight details.

But, in this case (and other shots like this), you could very well argue that the sigmoid result is more pleasing! And I kinda agree .

So, in cases where you want the roll off in the highlights to still try to retain sudden exposure changes in the scene, filmic seems to do a better job. But in many (very) bright shots, that’s not what you want to happen, you want to lose a bit of the highlight details to make them more smooth.

Now, messing with local contrast in the highlights can get some details back:

But then I’m fighting results from sigmoid, so although I find this very much a good idea, it’s not ‘the most easy’.

Now, filmic can also be ‘fixed’ easily if you know how, by leaving it at maxrgb but applying a touch of filmic-reconstruct, with the slider more to ‘bloom’. This smooths things over:

So, in my mind (and maybe oversimplifying it a bit): Sigmoid is good for smooth highlight transitions, filmic is good for preserving more of the ‘recorded’ gradation and colour in there.
But the smoothed ook if sigmoid can be undone by using local contrast, and filmic can be made to smooth over sudden changes in brightness by using its own reconstruct parameters.

What you want depends on the shot I guess, but it very much tells you if you get quicker ‘pleasing’ results with one vs the other.

Jandren was joking when he said “maybe I need to rename it to sunshine tonemapper” or something across those lines… but it is a very good module for shots like that .

[kofa (István Kovács)]

I think the comparison with filmic is valid, since until sigmoid arrived, it was the only unbounded → bounded tone mapper available in darktable. It has its strengths, as well as weaknesses, so comparing sigmoid to the old (and current) default is a very relevant topic. I agree with @flannelhead, and will refrain from mentioning filmic, unless in comparisons with sigmoid.

I’m testing it on more images, and find that I use tone equalizer more often to compress the tonal range than I usually do with filmic, but I begin to like the results.

[qmpel (blends colors 🎨)]

Thank you @jorismak for posting your findings. For me this is helpful and very interesting

@jorismak

I’m just very very VERY curious to get a raw file where someone had the issue with filmic completely changing things like skin tone. And even more so if sigmoid doesn’t do the same.

If you have proper lighting, I’m all with you - don’t think that filmic will mess up with the skin tones. But I often have shots, where some faces have bright sunlight. Then it gets a bit messy.

Here is a (snap-)shot that shows the effect:

_MG_0969.CR2 (17.7 MB)

For my taste, all the standard filmic options are a tad off - so I need to start fiddling. Sigmoid on the other hand gives a believable result.

license CC-BY-SA

[PhotoPhysicsGuy (Bob)]

@jorismak

And because the differences between sigmoid and filmic are minimal from my testing, and the differences are there in how highlights are handled.

Thank you for posting this! It’s interesting, I come to a different conclusion from looking at those examples?!

You show filmic:

@jorismak

Now, filmic with just a click on the ‘auto white’ picker and then changing to luminanceY mode:

and the according sigmoid:

@jorismak

And here sigmoid in rgb ratio mode with the skew a bit down:

where you sum up the differences as:

@jorismak

Now, there are two ways at looking at the results. “Filmic preserves more of the data” and “sigmoid smooths more”.

which to me is not the problem in the comparison.

For me it’s the gradient from sky to the superbright sun-disc that has a noticeable twist in color and below the sun towards the horizon that hint of salmon. It’s not smooth, not because of detail, but because of a non-smooth gradient. Even far away form the admittedly extremely bright sun disc. Sigmoid not only is lacking the hue-twists (and with that improves smoothness of brightness) sigmoid also lacks that salmon flavor (I haven’t color picked the region but I would not be surprised if that is actually the same color, but the surrounding hue-twist in filmic pushes the salmon-color perception).

Therefore, I think your conclusion:

@jorismak

So, in cases where you want the roll off in the highlights to still try to retain sudden exposure changes in the scene, filmic seems to do a better job.

Is not what I would arrive at. The definition you see in filmic may be part of some weird interaction of color-brightness perception combined with the several possibilities of how filmic is trying to retain RGB-ratios (which might be the wrong thing to do if you want a high-visual quality dechroma for tonemapped highlights).

I don’t want to drag the topic too much by talking about filmic here, but the conclusions drawn from example pictures comparing filmic to sigmoid are sometimes surprising for me.

(All of the above surprisingly holds true for my uncalibrated and calibrated monitors which is not what I would have expected)

EDIT:

@qmpel

Here is a (snap-)shot that shows the effect:

that comparison speaks volumes and imho emphasizes with skintones what I saw in the sunset picture.

[jorismak]

@PhotoPhysicsGuy

or me it’s the gradient from sky to the superbright sun-disc that has a noticeable twist in color and below the sun towards the horizon that hint of salmon. It’s not smooth, not because of detail, but because of a non-smooth gradient.

1. We all judge on different merits, so this is fine!

2. I’m seeing the same colours when judging the image. Sigmoid in ‘per channel’ mode does something different, but I don’t like it at all. And messing with the ‘preserve hue’ slider introduces a colour which shouldn’t be there, in my opinion.

Toggling between the filmic one and the sigmoid one, the colour (hue) is not different (well, maybe if you start measuring). But just looking at it, specially from a meter extra away, I don’t see a problem between those two.

Now, yes, the wholep point I was saying, is that filmic handles the extreme highlights different. You see here in filmic there is a sort of ‘cut off’ point where it goes to white. Before that point, it tries to hold on to the colour that’s there for dear life. This causes the jump, and the fact it’s not judged ‘smooth’ by most (and I agree!).

But sigmoid makes it smooth by not being as saturated. Sigmoid just starts going to white smoother and earlier. That colour filmic is trying to hold on to, is there in the un-tone-mapped version. Filmic just messes up the transition.

Sigmoid wins the transition, but it does so by being less colourful / saturated. It loses colour-detail.

In cases where you can’t properly judge the ‘true’ colour anyway (like bright sun, or skin tone directly lit by bright sun), this is probably the more preferable method. (Also the reason why the ‘colorbalance rgb - natural skin’ preset desaturates the highlights quite a lot).

But imagine that you know which colour it was, like a bright blue sky. Then what sigmoid is doing is a shame, because it looses the colour.

If the amount of colour in the extreme highlights is of no concern, then filmic luminanceY (or ‘no’) but now even better sigmoid is probably the choice.

I’m looking at some snaps from my own, and yes, when directly hit with sunlight things can get messy. But to be honest, I don’t like sigmoid then at all. I just think skin with bright sun on it is always a mess. I try to use the tone-equalizer to, well, equalize the lighting of it. If that doesn’t work, I just often don’t use the shot. Maybe that’s why I never thought about differences in that.

So, I’m learning use cases here . Can you guys have a crack at this file? I’ll start a play-raw thread for it to not pollute or hijack this thread for a single photo. But I am interested in what filmic / sigmoid uses are here.

Play raw here: Handling shade + direct sunlight with skin-tones, filmic / sigmoid preference?

[priort (Todd Prior)]

To further your comments do you recall part of the original thread…I will have to go back where @jandren suggests to use contrast in rgb contrast in conjunction with sigmoid?? I will go back and edit this comment if I find the reference…but then might impact what modules are used to tweak it and what the highlights and shadows response is when combined with some contrast added from that angle…

[priort (Todd Prior)]

@flannelhead

As far as I know, chromaticity preservation was an explicit design goal for filmic - adjustments such as skewing some tones to yellow etc. were intended to done in other tools / modules. This allows for a kind of surgically precise workflow. Just look at the amount of control exposed to the user also in the filmic tone curve. Other modules have other design goals that lead to different workflows.

I think we are hashing out and discovering what was commented by @flannelhead. And when you want or need that filmic will be the way to go and maybe sigmoid will be a quick path or require less tools to provide manipulation of the “yellows” and other aspects of the tone mapping that will become clear from a bit more use and testing

[priort (Todd Prior)]

Thanks for sharing this one …when I have time I want to play with this a little… If the metric is what do they look like at this point, ie no fiddling … but see merits in maxrgb and no in filmic… if you were to explore adding back color to no it might be a nice base and maxrgb gives pleasing tones on the post and evens out the skin nicely so just needs some color work…

[jorismak]

Well, I went at it.

And I get a filmic version and a sigmoid version where I prefer the filmic version. And also, when flipping back to back, I hope to illustrate what I mean with ‘the differences are not that big’.

Hmm… but I’m kinda doubting myself here. Did I end up at this sigmoid version, because I did the filmic one first? Am I trying to make sigmoid to the filmic thing, instead of it’s own thing? If I did sigmoid first, would I end up with a different filmic version?

The ‘cheating thing’ here is that I use the natural-skin preset of colorbalance-rgb. Mainly because I always use one of the presets as a starting point. But also because I think it needed it.

Now, I did try sigmoid without colorbalance-rgb. And when comparing without colorbalance-rgb, sigmoid did an interesting thing. But I wouldn’t call any of the two a ‘winner’.

Once I introduced colorbalance-rgb, I liked sigmoid better with it in the pipe, but then I prefer the filmic version (although it’s VERY nitpicky).

Also look at the sky in the back, and the wood of the ceiling. Filmic seems to do a bit ‘more’ there.

Now, I didn’t do ‘much’ with filmic. I set the exposure by using the exposure-picker at 60% on the shaded side of the head. Enable filmic, hit ‘auto white’, and maybe reduce it a tiny bit more. I experiment with the ‘highlights contrast’ which defaults to ‘hard’, sometimes I like ‘safe’ more. Depends on the image.

Now, I copied the image, disabled filmic, made sure I haven’t nothing else after exposure (so no colorbalance-rgb or local-contrast, diffuse, etc… at this point yet). I enable sigmoid.
I notice I almost always like the image with the contrast reduced. I almost never like the ‘per channel’ option because it looks too pale to me, and when reducing the ‘preserve hue’ slider, the skin tone shifts to a yellow cast I don’t like at all. So, ‘rgb ratio’ it is.

Again, this is not to say I don’t like it. I’m just trying to learn how other people use it and see things!

(animated png, seems to be working for me)

source frames:

_MG_0969.CR2.xmp (9.2 KB)
_MG_0969_01.CR2.xmp (9.5 KB)

[jorismak]

(And to add more confusion, just a Filmulator 2 sec attempt. As a sort of ‘compare’ to what other tools can do with the colors).

[PhotoPhysicsGuy (Bob)]

@jorismak

Sigmoid just starts going to white smoother and earlier. That colour filmic is trying to hold on to, is there in the un-tone-mapped version. Filmic just messes up the transition.

I can 100% agree on this characterization.

@jorismak

But imagine that you know which colour it was, like a bright blue sky. Then what sigmoid is doing is a shame, because it looses the colour.

But it has to, at somepoint anyway. I think we are discussing where the tradeoffs should be and how they look like?! And this is where I see a huge improvement/addition in what sigmoid offers.

[priort (Todd Prior)]

What if you tone eq the cheek on sigmoid…for me the tonality of the filmic one on the face is too equalized but the sigmoid one is a tad over contrasted on the cheek… I did some matching of patches in another image and it seems like filmic needs a bit more exposure if you try to match areas in an image between the two modes… this might not be the way you would use it or edit but I found when the goal was to get a similar edit filmic needed a bit more ev ie you couldn’t just enable one and disable the other one… but as with lots of this it can be highly bias by the observer in this case me…

PS that observation came from messing around with the pants in the image @Kofa provided of the fish float in the parade of the striped shirt gentlemen…

[afre (Alan)]

@jorismak

(animated png, seems to be working for me)

It works when I click on it to activate the lightbox preview. I have an idea: insert grey blanks in between so our eyes get a chance to reset.

[afre (Alan)]

One thing I keep on reiterating about the filmic module is that it is doing a lot of things at once. Even on default settings. I know AP and others would loathe this comparison but it is kind of like PS editing, albeit more state-of-the-art, where it is challenging to view the effect of each parameter in an objective manner. In comparison, sigmoid (and filmulator) is more straightforward. So, it is truly difficult and perhaps a fool’s errand to compare the two beyond usability.

What I can tell from the _MG_0969 samples @jorismak provided us just now is that filmic appears to emphasize local contrast/detail better than the other two methods. However, it has a least one bug (see the concrete end of the drainage pipe).

[Y69 (Y)]

Sigmoid seems to instantly heal my issues with branches/vegetation in backgrounds - hard to adjust hardness of the transition, too strong coloration of the branches, fiddling with preserve chrominance & reconstruct.

Left: filmic rgb with tuned levels, reconstruction, and preserve chrominance to RGB power norm
Right: sigmoid with a bit lowered contrast and bumped skew.

Filmic auto levels:

Filmic tuned:

Sigmoid default:

Sigmoid tuned:

CC BY-SA 3.0

DSC06179.ARW (23.9 MB)

[priort (Todd Prior)]

The source images are nice then you can load and use snapshot…

[afre (Alan)]

My comment was mainly to point out things for casual readers. I typically open in tabs, arrange and view them the way I like.

[priort (Todd Prior)]

Off topic but for those branches you can go into the lens correction module and override TCA and then tweak a little the blue…you can make nice improvements… Rawfiner shows this as a suggested first step for tough CA esp on branches… also the initial sky treatment you see with filmic likely comes from using clip highlights…I don’t think you will see that with the new code… Nice image to experiment with …thanks for sharing it…

I spoke too soon…maybe consider sharing the raw its a nice image for testing out a few areas

[Y69 (Y)]

Added the RAW source