On jeudi 14 février 2019 12:23:03 CET Dusenberg wrote:
> I have just been puzzling over the 'pink highlight' problem on several raws
> I recently shot (not HDR), all of which have blown highlights caused by a 
> camera operator silly error :)  I recently created a custom style for my
> camera and I thought that was something to do with the problem, so I'm glad
> it's a known characteristic for which a solution is available - don't
> over-expose!
> 
> I also noticed that Highlight Reconstruction was switched on on these raws
> and I hadn't done it as I like to start from a blank canvas, so switched it
> off. Reading the trail of replies however, it seems the only way to
> fix/reduce the pink problem is to use Highlight Reconstruction - so I'm now
> wondering if DT automatically switches Highlight Reconstruction on if it
> detects blown highlights; does anyone know?

As far as I know, "highlight reconstruction" is _always_ automatically 
switched on for raw files (and never for jpeg or png files). It just has no 
visible effect if there are no over-exposed areas. The default setting is 
"clip to white", I prefer using "reconstruct in LCh", which usually gives a 
bit more detail in slightly over-exposed areas (clouds are a typical example).

Keep in mind though, that you can get blown areas for two reasons in 
Darktable. First is of course over-exposure, in which case the raw data is 
clipped. 

But the basecurve that is (automatically) applied to raw files can also cause 
blown regions. Darktable picks a curve based on the detected camera (for the 
default case), and e.g. the Sony basecurve clips rather soon (at input values 
of about 90 on a scale 0..100). 

For that reason, I tend to switch to another curve (Canon or Leica, neither of 
which clips the highlights), or use the "filmic" module (which has a learning 
curve). That very often means also a correction in the "exposure" module!

Remco


____________________________________________________________________________
darktable user mailing list
to unsubscribe send a mail to [email protected]

Reply via email to