[Solved] Why do CAT and Filmic render clipped areas so weirdly

[Leniwiec (Michał)]

Hello

There is one thing with new workflows, which remains mysterious to me, and pulling me back from ditching “the old ways”, i.e. White Balance and Base curve.

As much as I like the whole concept and overall colour rendering of CAT in Colour calibration module, I find it hard to use when dealing with not “safe” images, which contains overexposed areas.

Perhaps my comments below each picture will be more clear:

Modern combination of Filmic RGB, WB set to “Camera reference D65” and CAT16 “As shot in the camera”. Of course, not the best photo in the world, but clearly shows what happens to the bright lights, both natural at the very top and artificial on the bricks.

Here the only thing changed is going from CAT to the old White balance “As shot”.
Surely blown highlights are still blown, but I’d say in a much nicer way - more smooth and not suddenly cyan

Combining Base curve with CAT gives quite acceptable (to me) transitions between non- and over- exposured, but somehow brightens the image, taking away all the details from the top-center pillars.

Finally, my safest and most robust combination since darktable 2.4 - White balance and Base curve.
Sadly, colours tend to be yellowish compared to nice CAT rendering, however, that setup is more gentle in terms of smoothing transitions from clipped and non clipped areas, not introducing cyan in highlights and showing more details in those extreme areas.

So those are my edits in JPGs and below is the RAW published as Creative Commons, By-Attribution, Share-Alike as well as recipe for the “worst-case scenario”.

_MG_5096.CR2 (31.8 MB)
_MG_5096.CR2.xmp (15.5 KB)

Bonus

After talking with my friend about DXO Pure RAW, in the evening I remembered about my old free license for DXO Optics Pro and its ability to apply PRIME denoising and export as linear DNG.

So if anybody likes to experiment - enjoy

_MG_5096_DxO.dng (83.4 MB)
Creative Commons, By-Attribution, Share-Alike

[eyedear]

this is the best I can wring out of filmic with highlight reconstruction and some tone equaliser

_MG_5096.CR2.xmp (6.5 KB)

[spidermonkey]

Thanks this is interesting. Slightly off topic but I couldn’t help noticing the ‘early music’ plimsoll shoes.

[charlyw64]

IMHO that is just the consequence of a too late transistion into a bound color space.

Your camera captured out of gamut colors (that the human eye can‘t perceive), any changes made on the basis of that linear scene data pushes even previously well defined areas over the edge and into that territory of unperceivable colors - which then get mangled by the conversion filters and since some of the colors aren‘t even in the realm of the camera captured colorspace the results are undefined.

Unfortunately IMHO the whole concept of scene referenced editing is a complete mess and should be abandoned ASAP as there are no edits except for direct pixel copies that are safe to perform on that data before it is transformed into a well defined, bound colorspace.

[Leniwiec (Michał)]

Oh!

Now it looks like there was quite a lot of data hidden in the upper part of the image! And you recovered them

Also, I see I’m not the only one willing to switch off the chroma preservation in Filmic

However, I’d be really glad to see someone find a way to restore the contrast or so called “punch” of the overburned Base curve version without losing the highlights.

[Jade_NL (Jacques)]

_MG_5096.CR2.xmp (9.8 KB) darktable 3.7

This is a quick edit and definitely could be better when spending the time on it, but it does illustrate that just the basic modules and playing with the sliders can get you a rather decent result when dealing with an over-exposed/blown/problematic image like this.

Best way to solve this in the future is not to overexpose the image while shooting it.

[charlyw64]

@Jade_NL

Best way to solve this in the future is not to overexpose the image while shooting it.

Sorry that is the worst adice I have read for a long time. The subject of the shot clearly should be the musicians and you would not do yourself a favour by undrexposing them to preserve the inevitable blown hightlights of the bright windows. You could bracket if you want a HDR (the only way to deal with the dynamic range of that scene).

[Leniwiec (Michał)]

Hmm…

Well, surely underexposing is quite logical technical solution, but also what Charly says is - I think - the point. Windows and lamps would have gone blown out anyway, but the decision was up to me How would I want to expose. There is always a trade off.

At that time I was a big fan of even overexposing to get the most signal to noise ratio in the shadows or main subject.

This topic helps me to, I’d say, understand the nature of each slider in Filmic or CAT. I see we all have the same set of tools and sliders, but everyone can use them differently and get desired results.

My mind is quite reluctant to understand why e.g. Filmic’s white point slider creates cyan artifacts if moved to the right - I think to myself: “Well… moving it to the right means giving more space for the highlights, yeah? Sooo… they should be blending more and more smoothly, right? No…”

I know I’m a strange man
I have set the robust and proven workflow and I should be happy and silently move on from one photo shoot to another, but somehow I can’t rest and constantly force myself and the software to achieve new sucesses in post production.

[Jade_NL (Jacques)]

@charlyw64

The subject of the shot clearly should be the musicians and you would not do yourself a favour by undrexposing them […]

I fully agree, but there is a lot of room in this shot to save a lot of the blown parts (not saying you can save it all). Have a look at the RAW and you might see and hopefully understand, why I gave this advise pointer.

This was shot with a Canon EOS 6D, a rather good camera so lowering the exposure wouldn’t hurt you all that much, even if shooting at 12800 ISO. Much could be brought back with the added bonus that reconstruction of the now much less blown parts would be much easier.

[Aliks]

Darktable can cope with 2 light sources, so I put in a second instance of Color Calibration, masked to the four musicians, and also used the same mask to bring out a bit more detail on them using exposure, and local contrast.

Looks OK to me, and does illustrate the power of scene referred IMHO.

_MG_5096.CR2.xmp (16.3 KB)

[eyedear]

Yeah. I think I pull down the highlights too much. The middle top left where it is overexpose I think there is no more information there. I would rather have that at almost pure white if I spend more time on it.

@Leniwiec

Also, I see I’m not the only one willing to switch off the chroma preservation in Filmic

Yep I have tried my best to use the other color preservation on almost 500,000 images in my library it works in about 1% of my images. I find the contrast to be weird. It does preserve the chroma when you are above 4. But so far I find the no preservation works the best for me. If I do use the power norm color preservation then I would have to add in local contrast and also add basic color fullness in color balance rgb to get back the contrast and color which means more time tweaking the photo again.

I think my settings mimic the base curves while preserving the editing in linear space.

I find that a value of 3-3.5 ev is the ideal setting white point slider to get a natural contrast and color. And then use the tone equalizer to pull down the highlight or push up the shadow. Much easier to control. I think this is the wrong method but so far it’s works for me. Hahahahaha

[s7habo (Boris Hajdukovic)]

@charlyw64

Sorry that is the worst adice I have read for a long time. The subject of the shot clearly should be the musicians and you would not do yourself a favour by undrexposing them to preserve the inevitable blown hightlights of the bright windows. You could bracket if you want a HDR (the only way to deal with the dynamic range of that scene).

My photo. Only the highest highlights in the windows above are very little overexposed. Photo taken with Nikon D7200 from hand. ISO 800.
The main subject is the altar retable.

Before:

After:

One more. Nikon D850. Exposure for the highlights. Handheld.

Before:

After:

EDIT: I forgot to mention that I used darktable for this.

[eyedear]

@Leniwiec

However, I’d be really glad to see someone find a way to restore the contrast or so called “punch” of the overburned Base curve version without losing the highlights.

this is the best I can do without using filmic

_MG_5096_02.CR2.xmp (6.6 KB)

hope it helps

[nosle]

Raw software simply have to be able to cope with blown highlights. (I think dt can) Arguing for perfect capture of the dynamic range at all times is crazy talk completely divorced from reality and the history of photography .

Now personally I’m a chronic underexposer. I do recognize however that much better photographers than I am constantly blow their highlights with great looking photos as the result.

Managing highlights and dynamic range was one of the purported benefits of filmic. Going cyan in a way that isnt automatically or very easily tamed isnt really following through.

Cameras loose colour fidelity in under exposed areas. Its not a very good idea to place your main subject several ev’s submerged.

[MStraeten (MartinSt)]

@nosle

Raw software simply have to be able to cope with blown highlights

They can mitigate the effect of clipped highlights, but they can’t restore lost information. So if you want to avoid losing information thats up to you when you take the photo.
Cyan highlights are an indicator that channels didn’t clip equally. It’s up to you to decide if this is intended or not.

[nosle]

Sure* but getting back detail is different from having blown highlights turn cyan. I understand the thinking that “its just the colour science at work” thing is that the purpose of raw software is to manipulate the data into images. The colour science is only of interest in so far as it enables successful manipulation of images.

Again im sure there are tools to deal with it in dt Its the arguments not the software I’m arguing against.

* actually no because almost all other software had good inpainting implementations than successfully mask loss of detail.

[charlyw64]

@MStraeten

Cyan highlights are an indicator that channels didn’t clip equally. It’s up to you to decide if this is intended or not.

Nope, they are a sign that the software does switch to a bound, well defined color space too late in the pipeline (no contrast, sharpening, color manipulation is safe to be performed before the image is converted to a proper bound color workspace - which doesn’t need to correspond to the display colorspace). Remember that a scene referenced workflow as used in darktable would have deal with colors that the camera captures but will not be perceivable by the human eye. Manipulate the data at that level and you risk creating even more absurd colors even outside the gamut of the cameras sensor - which later needs to be translated to something that a display medium can show and the human can see. Darktable has gained preservation of those undiscernible colors at the cost of losing any way of correctly editing or displaying the image as anything other than an arbitrarily tone mapped HDR…

Unfortunately the means to convert early enough in darktable have been either deprecated or have suffered teminal changes to make them work before that transistion has been made.

As an example take sharpening. When performed on (as an example) a very red colored flower that is barely within the human range of vision will create higher contrast along detected edges that will contain areas that the human eye can‘t distinguish from the unsharpened surrounding - but the process of cramming that into the output of the filmic module can either drop the overall contrast even more to „preserve the false colors created by the sharpening“ or clip them (which loses the effect of the sharpening).

All other programs do work on large color spaces which have well defined translations into smaller output color spaces.

[paulmiller]

The coloured highlights in image #1 are the result of an interaction between the Highlight Reconstruction module and Color Calibration / Filmic. Try turning Highlight Reconstruction off…

[MarcoNex]

Here is my attempt, DT 3.6

_MG_5096.CR2.xmp (10.0 KB)

As highlights are blown out, I tried to bloom them a bit to try to create an atmosphere.
Without filmic color preservation, the blue of the violinist’s dress seems unrealistic to me

[Leniwiec (Michał)]

Impressive! That’s scene-referred, I presume?