Help with highlight reconstruction of clouds

I am struggling with a number of photos like this one to properly use either the highlight reconstruction module or filmic’s highlight reconstruction to handle the small clipped parts of these clouds; it seems no matter what I do in filmic, I end up with this blue-ish tint:

clouds_filmic_blue_tint

Can someone show me how to more skillfully use highlight reconstruction for this type of situation and make the recovery white? I’d appreciate any other suggestions on this photo too. Thank you!

Here’s the RAW:

_MG_0814.cr2 (21.4 MB)

This file is licensed CC BY-NC-SA 4.0.

Hi @garibaldi,
the new highlight reconstruction option “inpaint opposed” works perfectly on this image.
It is already available in the master build. I would think it will be available in the next release? (just me thinking I have no inside how the releases work :sweat_smile: )

What a lovely view! The clouds add so much freshness to the shot.

Here’s my rendition with a bit of highlight reconstruction and AgX

Here are two versions, one with the new HLR and Sigmoid, and one more suitable for darktable 4.0.1 with FilmicRGB and the old HLR.

4.1.0, Sigmoid:


_MG_0814 Sigmoid.cr2.xmp (13.8 KB)

and 4.0.1, FilmicRGB

_MG_0814_FilmicRGB.cr2.xmp (16.1 KB)

Notice that the xmp with the Sigmoid version will probably not open in darktable 4.0.1 or older.
Other than tone mapping and HLR, they are more or less the same.

1 Like

Since it is on master, it will be part of 4.2. The only scenario for it not to make it would be q revert because someone finds a major issue. That’s extremely unlikely at this point since we have been using it for a few months now. Because how much better it is over the current default, it will be the default in 4.2.

1 Like

The while level for your camera seems to be wrong. Turn off highlight reconstruction, turn on raw clipping detection, and lower the white point until the magenta is marked as clipped.

I found that 12800 is a good value.

Having the wrong raw clipping value impacts all highlight recovery methods.

If you are on darktable 4.0 or older, try highlight reconstruction with recover in LCh (but disable color calibration and set white balance to as shot):

Or, if you are on at least 4.0, you can try keeping white balance as camera reference and keep color calibration enabled, and use guided laplacians as your highlight reconstruction method (but this time, it did not work for me).

If you are on the development branch, use inpaint opposed as the recovery method (it is the new default):

10 Likes

Good catch on the white balance. I think it is the first time I see a wrong white point on a cr2 file.

Exif values for this file are not that low but whatever works…

image

What software … art has AGX luts but nothing in DT or RT right… Blender??? Other??

1 Like

Great answer…even with the default white level …things are fine as long as you are using legacy wb and not clip highlights with the modern wb ie color calibration… Going to legacy wb and a touch of tone eq totally restores that small segment and no blue/cyan stuff…

The work done with the new HLR to make it compatible and mostly free of this sort of things is nice…

Any OCIOv2 software (you can tell I’m a fan). Blender works. I use Olive! (Open source NLE video editor)

In Darktable you can use a LUT, but it’s a LUT… results may vary.

(Looking at ART right now, thanks for the heads-up!)

My version… ( filmique rvb ).

_MG_0814.cr2.xmp (21.1 KB)

4 Likes

I was exploring this here…

FYI

1 Like

Hi @garibaldi,

here is my edit :slightly_smiling_face:


_MG_0814_02.cr2.xmp (19.6 KB)

1 Like

Rawspeed has a single value of 13653 for the Canon 50D
RawTherapee has different values base on the ISO. For the 320 ISO, it uses 12700.
All 3 are different (ExifTool, Rawspeed and RawTherapee).

Combined with the test Kofa did, I would use 12700 on this image.

OP, for other ISOs, this is what RawTherapee has. You can create Auto Presets in dt to adjust the white point based on these.

“white”: [
{ “iso”: [ 100, 125 ], “levels”: 13300 }, // typical 13432
{ “iso”: [ 160, 320, 640, 1250 ], “levels”: 12700 }, // typical 12790-12810
{ “iso”: [ 200, 250, 400, 500, 800, 1000, 1600, 2000, 2500, 3200 ], “levels”: 15630 }, // typical 15763-15733
{ “iso”: [ 6400, 12800 ], “levels”: 16200 } // typical 16383

1 Like

Tried to create contrast by colors - maybe the result is a bit too colorful, but I like it.


_MG_0814.cr2.xmp (46.4 KB)

Thanks for sharing this nice image!

1 Like

Fujifilm? Wasn’t this taken with a Canon 50D?

I was trying to reference the commit I used.

I did an edit to avoid confusion. The values are for the 50D camera.

1 Like

Today I learnt something new about DT. I opened this image in DT and had the infamous magenta highlights in the clouds, I turned on the raw clipping detection option, adjusted the white point value until the magenta areas were revealed as clipped. I turned off the raw clipping detection and the image looked great. What a simple fix.

I then tried the method on another picture with very bad magenta cloud issue. Again the method worked by just lowering the white point until the magenta clouds disappeared. However, this new image needed further highlight reconstruction to restore the blue sky and cloud detail. I experimented with all the options available in the current Windows weekly build. Only the inpaint opposed and the segmentation based method worked well.

Thanks for the hint about changing the white point. It is easy to adjust and needs to become part of my go to fix if I see magenta highlights.

1 Like

I think you can do this but really I don’t think it’s a routine editing tactic. If the value is correct as expected by the camera thing are generally fine. In DT these things have recently been looked at and addressed as the move to scene referred and modern wb didn’t really play nice with in particular the default clip highlight mode. In many cases I find looking at all these play raw files in older versions that using the correct white point and legacy wb was enough to make thongs behave. Now with inpaint and segmentation it’s even better