Reconstructing some highlights in an alley

Antonio.guerreiro · February 17, 2021, 1:15pm

Here is an image for you to play with, and I would also like some advice about reconstructing highlights.

There are some very small details on the right wall that are blown out. The best option in the “highlight reconstruction” module seems to be “reconstruct color”, but I would like to smooth the effect a bit. I tried to use the “reconstruct” tab on filmic, but I got some artefacts and could not get rid of them. I would like to inpaint some colour on those highlights, but with less saturation than the results produced by the “reconstruct highlights” module.
Any ideas?

Here are some details:
Using just “reconstruct color” on “reconstruct highlights” module:

Using “reconstruct” tab on filmic:

Captura de Tela 2021-02-17 às 09.59.03

And here is my edit (darktable 3.4):

DSC04027.ARW (47.4 MB)
DSC04027_03.ARW.xmp (31.4 KB)

This file is licensed Creative Commons, By-Attribution, Share-Alike.

Tim · February 17, 2021, 2:27pm

I did not do an all out edit, only a few steps to work on the blown highlights. Color calibration, filmic RGB, and local contrast. dt 3.4.1

DSC04027.ARW.xmp (5.8 KB)

Jade_NL · February 17, 2021, 4:14pm

This one isn’t all that simple to “fix”. This is what I came up with (200% magnification):

This is a combination of filmic (settings can be seen in the shot + 5 itterations), highlight reconstruction->reconstruct colour (0.970), lens correction (distortion+vignetting) and chromatic aberrations turned off.

I think that the lens/aberration settings also play a roll in this one.

BTW: This is RawTherapee at 300%:

Off topic: @heckflosse: You might want to have a look at how your ‘poor man’s haze’ addition handles this on. I’ve been looking at the results by using what is shown in the terminal. Just about every time it is good (enough), but this one misses the mark. I know that this image might not be all that suitable, but just wanted to let you know.

gadolf · February 17, 2021, 5:44pm

Nice image, thanks for sharing.
Harsh light, harsh contrast.
My take.
Basic scene-referred/modern workflow, but I confess I didn’t care too much about those highlight details, and focused more on the overall mood.

DSC04027.ARW.xmp (14.8 KB)

heckflosse · February 17, 2021, 6:22pm

If poor_man_dehaze prints negative values in terminal, then they should not be used. It works only (“correct”) for low ISO images.

Thomas_Do · February 17, 2021, 6:26pm

Used filmic highlight reconstruction.

DSC04027.ARW.xmp (16.7 KB)

Suki2019 · February 17, 2021, 6:27pm

hello

I find that the RGB curve does a very good job with such things.

…quick editing on the laptop

Reconstructing some highlights in an alley_DSC04027_01.ARW.xmp (12.9 KB)

crop

paolod · February 17, 2021, 6:46pm

Try turning off the “preserve chrominance” setting in filmic - it can badly exacerbate those artifacts at the boundary between bright and dark pixels.

Also the interpolation setting in Darktable preferences might have some effect on the artifacts - see discussion at Darkroom: jagged, coarse lines, sharp transitions in zoomed-out preview (and also rotation, lens correction?) · Issue #8016 · darktable-org/darktable · GitHub.

Jade_NL · February 17, 2021, 7:17pm

Had a bit of time to spare, so I also did a normal, darktable play raw edit:

highlight.alley.arw.xmp (12.2 KB) darktable 3.5.0

Antonio.guerreiro · February 18, 2021, 12:02am

That’s interesting, your image shows no artefacts, but it does on my laptop when I load your sidecar file:

Didn’t try to export the image, though, I’ll do that and see if the artefacts are still there.

afre · February 18, 2021, 12:03am

Version disparity? He is probably using dev.

Antonio.guerreiro · February 18, 2021, 12:25am

Using the rgb curves does seem a nice trick to deal with that. I tried it myself, but I also put an instance of color balance after it to desaturate the highlights a bit. The results look very decent, those little highlights seem more “tamed”:

ggbutcher · February 18, 2021, 1:37am

Sooo… I downloaded the image to see what the librtprocess highlight recovery would do. It just de-magenta’d them, still sharp and white.

I just decided to go abstract, grayscaling the image then applying a red-channel curve that lifted red to start, but went negative at a convenient place to turn the highlights cyan. Now, what highlights?

Suki2019 · February 18, 2021, 5:08pm

RGB Curve in darktable is a great tool