I have recently started going the filmic / tone equilizer route. Results are amazing. Therefore I scanned my library to find pictures which might of interest for this dev process.
I found one but encountered a strange behaviour in DT. I have done nothing to the picture - just loaded the raw withouth enabling any modules manually. Overexposed parts of the image are pink instead of white. Loading the same raw in RT shows “clean” white.
I imported your image to my installation of dt 3.0.2 and it looks the same.
Turning off OpenCL for a test and also did an export of the untreated image, it looks the same in the exported .jpeg
Thx for the quick replies. Setting the white point to something lower in the raw blackpoint/whitepoint module did the trick as well as decreasing the clipping threshold in highlight reconstruction.
So I assume that this behaviour is intended - since the affected regions are not within the color profile - right?
Wikipedia provides a good description of gamut. If you have pixels out of gamut these pixels cannot be displayed correctly. At least one channel will be clipped, resulting in a more or less ugly color cast.
Usually it doesnt make sense to decrease the clipping in highlight reconstruction because that also affects existing information in further channels - you can conserve their information by selecting ‚reconstruct in lch‘.
In your case its a wrong setting of the raw white point for your camera. This setting can vary over the iso range of your cam, so you can define a presets to be applied as default for all images taken with this cam at specific iso values
Those pink areas aren’t a gamut problem, but a clipping problem:
the green channel of the sensor is overloaded, so it is clipped at the maximum value a pixel can have.
Ideally, all three channels should clip together when you get an over-exposed area.
Unfortunately, the white balance correction amplifies the red and blue channels relative to the green channel, so a clipped white area will show up as magenta.
This is corrected by the highligh reconstruction, which uses the raw white point to recognise clipped areas. So if the raw white point is set too high, highlight correction will not see any blown areas, and happily leave them with the (wrong) magenta color.
@rvietor’s is the right answer. When you expose an image with sensor-saturated highlights, the data on all three channels piles up in about the same place on a histogram. When white balance is applied, those channel pile-ups separate, and cause a color cast on the saturated pixels.
Highlight reconstruction is about pulling that data to joy, or one can just set the white point to the lowest pile-up and push all of the saturated areas to white. The former is definitely preferable, but the latter can suffice if the saturated areas are small or about light sources in the scene. IMHO.
I agree, the raw over exposed indicator clearly shows the clipped regions. Activating the gamut check the magenta areas are also shown as out of gamut (independent of the profile choosed). But you are right, the primary problem is the overloaded green channel of the sensor. The “out of gamut” is only a consequential problem.
I have been struggling with the same issue a while ago. The nice people here helped me out by pointing to the Highlight reconstruction module and switch that to 'Reconstruct in Lch" and start lowering the clipping threshold.
