New Sigmoid Scene to Display mapping

ariznaf · January 20, 2022, 1:35pm

Not me who was asking for dramatic sky (depends i what you called dramatic) just not so dull.

You don’t allways have a tripod or the time to do that.
Most of my photos are when travelling or going for a walk, not a prepared photo.

That would be an appropiate answer if other tools were not able to get better results.

What was discussed here was whether filmic and /or scene refered workflow was in the origin of making it more difficult to expand those lights or not.

The color balance trick provided by @s7habo seems to work quite well.

And highlight recovery in color preservation is problematci in DT as it creates artifacts.

Just addressing people to not clip their photos would not solve the flaws.

I think we are going a bit out of topic here, and @jandren won’t be happy.
The topic was about comparing sigmoid and filmic and whether it would be good to have it in mainstream.

Most of us are having a look to sigmoid due to difficulties in expanding highlights (in skies for example) and seeking for alternate paths.
May be related but not the initial intend of the OP.

rvietor · January 20, 2022, 1:42pm

But when an important zone in the image has clipped pixels, you have irretrievably lost information. Any program that claims it can recover that information is lying, plain and simple. The best you can hope for is some kind of approximation. And getting a good colour approximation is even harder than getting an approximation of luminosity, even when only one channel is clipped (with three channels clipped, I wouldn’t even talk about “approximation”, it’s more like “guess”).

And I don’t see where switching between sigmoid and filmic has any relation to highlight recovery…

ariznaf · January 20, 2022, 1:54pm

Of course you have lost information, but some tools do a better job recovering or giving the photo a more natural aspect than others.

Most times the clipped areas are more or less uniform in color, and the luminance can be retrieved from the remaining channels.

The color preservation in highlight recovery does a decent job… if you overlook the artifacts it generates many times.

You are right, there are too separate things: highlight recovery (nothing to do with sigmoid) and the difficulty in expanding highlights to the midtones when using filmic and the linear scene referred workflow compared to the traditional workflow (where sigmoid or other tone mapping tool could help, or at least some would expect so).

The debate about recovery emerged when it came to give examples about skies.

nosle · January 20, 2022, 2:56pm

This is only true to some extent. Scene referred doesn’t improve your camera sensor. Under exposing means noise. Filmic can handle raising shadows quite gracefully but it introduces just as much noise as other software because it’s in the data.

Often the sky or other blown areas are of tertiary importance. There’s no real info there for your scene but you just don’t want it looking odd and attracting undue attention. Graceful handling of highlights mean you can trade in noise for recovery in areas of quite low and unimportant detail. Most software makes this a possible trade off.

Edit: Just to add that I write this as a recovering clipophobic. I have thousands and thousands of underexposed images.

paperdigits · January 20, 2022, 4:03pm

You can’t gracefully handle what isn’t there. That is my point.

s7habo · January 20, 2022, 5:04pm

Just consider the recommendation @anon41087856 gave regarding the difference between global and local tone mapping.

With the global tone mapper (filmic) you treat the whole dynamic range and with local (tone equalizer) you treat certain brightness ranges. This allows you to treat brightness ranges separately.

Example:

This photo is underexposed because the highlights on the walls of the houses should not be overexposed due to the very large dynamic range:

If I increase the exposure to see the market place better, the highlights are overexposed:

With filmic I have compressed the dynamic range and accordingly the highlights are there again, but the contrast there has been lost through the compression:

This is also logical because almost 4 EV in highlights have been “squeezed” from the dynamic range of the camera into 1 EV of the dynamic range of the screen:

To get contrast back there, I now have to treat this area locally. So I have to,

move it into the middle grey (darken it)
“spread it out” from there to the highlights.

So, after the white balance, I can first darken the sky with the help of the second instance of the colour calibration module and thereby increase the contrast between the sky and the clouds:

I can also use colour balance to increase the contrast even more:

And now I can use Tone Equalizer to treat only the highlights (walls of the houses) separately to increase the contrast there. First darken them (move them into middle grey) with first instance:

And then, with the help of the second instance, without perservation of the details (which means, to use it like a normal tone curve), I can expand this area back into the highlights:

And last, I can use diffuse and sharpen module to increase local contrast a little bit:

nosle · January 20, 2022, 5:04pm

Well most raw software can so it’s clearly and empirically possible.

paperdigits · January 20, 2022, 5:49pm

Nope. When channels are clipped, you loose that data.

Sure you can guess what should be there and put data in there, and darktable is weak here (but new goodies coming), but its still a guess and the application is inventing data that isn’t in the raw file.

anon41087856 · January 20, 2022, 6:00pm

Local tone mapping, and their followers : the halos.

nosle · January 20, 2022, 6:11pm

I think this discussion has slipped into highlight recovery which is part of the same problem of managing highlights but is still a distinct issue that I know you are working on.

Filmic is getting increasingly better but it’s still difficult to control the highlights of the global tonemapping. I agree that you have limited “space” to do so but It’s usually enough and workable with a more flexible curve. I understand that filmics answer to the issue is local tonemapping with the tone equaliser?

paperdigits · January 20, 2022, 6:16pm

Yes, or masked exposure module.

anon41087856 · January 20, 2022, 6:16pm

OGM NOOOOOO. I have explained that, and Boris has shown you the graph.

Stop with wishful thinking, any curve will give you the same result because it’s constrained in 3 points and display peak white is 100% while diffuse reflective white is 92%. 8% is 0.12 EV for specular highlights and emissive stuff.

No bloody global mapping, no curve, no one-size-fits-all method will give you that.

USE LOCAL TONE MAPPING.

Underexposed · January 20, 2022, 6:39pm

Second color calibration?

s7habo · January 20, 2022, 6:43pm

Yes. Second instance. First is for white balance, since I use gradient mask in second one for darkening the sky.

Underexposed · January 20, 2022, 6:44pm

Yes, but you called it color balance. Don’t you mean color calibration?

s7habo · January 20, 2022, 6:45pm

Yes. I corrected it. Sorry.

ariznaf · January 20, 2022, 6:57pm

Yes, tone ecualizer, but with tone eq I have a problem many times:

when you select any patch in the sky all of them are in the last step of the curve.

So you cannot expand the highlights using a toot that has no control points to separate that tones.

May be it is due to the linear scence processing, but then in tone eq nodes should be not equally spaced or be shown in a logarithmic scale with morre controls in the highlights.

Your image is underexposed in general terms to not burn highlights.
And you use exposure to compensate it and the tone eq to revert it and expand lights with good result.
But that image is not backiluminated and has not so much DR than a sunset or when the subject is in the shadows.

When you have the sky overexposed (let us keep the clipping effects apart and thinkg of an image with no clipped channels).

May it be that the opposite path works?
I mean lowering exposure a bit, and using tone eq to adjust?

The use of color balance to darken the sky is providing a good improvment for the photos I have tested, thank you.

I would not like to keep talking about highlight recovery here, as it really does not too much relation with tone mapping or sigmoid vs filmic comparation.

But it @anon41087856 local contrast technics in gamma compensated work flow can produce halos, but usually in high contrast border thansitions.
Usually are selections and strong tonal changes what produces your halos.

In a sky that LR or other software recovers from one or two channels being blown you have not that problems (except if you are really mad at dramatization).
In clouds or sunsets there are no hard borders or abrupt transitions, so recovering color from neighbourhood and aplying info about luminosity from the remaining channel usually does not work bad.

In any case it works much better than having a grey patch or a magenta color where it should be a not so saturated yellow.

s7habo · January 20, 2022, 7:44pm

If that is so like here…

…then you go to the masking tab and adjust the mask…

…and now you have it:

The photo is a little blurry but, before:

After:

Also done as I have described above.

s7habo · January 20, 2022, 7:50pm

Yes.

ariznaf · January 20, 2022, 7:58pm

Great, thank you.
Have to study it and do tries to see if I can master it.

I will put this thread in my bookmarks bag.

In my image, all samples from the sky are over the last node to the right.

But I had not used the mask postprocessing options.
One more thing to study and master, thank you.