On Mon, Apr 9, 2018 at 7:09 AM, Matthieu Moy <li...@matthieu-moy.fr> wrote:

> ----- Original Message -----
> > From: "Chris" <bydersjo...@gmail.com>
> > Seems as though the pink colours appear in my image when I disable
> 'highlight
> > reconstruction'.
> Yes, this is expected. The module is precisely here to remove this pink
> color.
> > With this disabled the Green & Blue channels have a lot less
> > data in the highlights than the red. Enable highlight recon, and it
> seems to
> > scale all the RGB to a simple range.
> The problem is the following:
> * All channels (R, G, B) are initially coded on a range, say [0, 2^14-1]
> for a 14-bits sensor RAW image. Channels that clip have the value 2^14-1
> (well, I'm slightly oversimplifying because saturation of the DAC comes a
> bit before this value, but ...).
> * The white balance module scales each channels, and usually scales the
> green channel less than the other. So the clipped pixels won't have the
> same values on R, G and B anymore after white balance. So, blown out areas
> are not white/grey anymore, they have color.
> The "highlight reconstruction" module does two things after that:
> * Desaturate the blown-out areas (done for all modes of the module)
> * Optionally, try to recover details. The default mode does not try that,
> it just clips highlights and turn them into plain white (which may become
> grey if you apply a negative exposure compensation). The "Reconstruct in
> LCh" is quite good to recover details when at least one channel is not
> clipped (e.g R and B are clipped, you can't expect much on colors, but the
> G channels can be used to recover an L channel). "Reconstruct color" does
> more or less the same, but works well on different images.
> In any case, setting the "clipping threshold" to 2.0 means the module will
> only work on pixels that have an L value 2 times greater than the clipping
> threshold. Which in practice means that the module will have no effect. The
> manual says "The default is usually satisfactory without any need for
> additional adjustments", which in dt is usually to be read as "There's no
> reason for you to change this value, but it's a parameter that appears in
> the internal algorithms so we've exposed it to the user just in case".

My experience with clipping threshold was different, at least on this dng.
Setting it to 2.0 allowed me to recover the cloud data and come up with a
pretty close match to what Chris achieved using LR.  Result is here

> > Odd, because I do not get this pink appear when I reduce the highlights
> in
> > Lightroom... and I am still using this same .dng file.
> I never used Lightroom, but from what I've read, LR's philosophy is to do
> a bit of processing automatically (or "behind your back", depending on the
> point of view). OTOH, dt makes almost everything explicit, and gives you
> control on everything, including settings that are not really meant to be
> changed. You may or may not like it, but it's unlikely to change ;-).
> The pink was there in the RAW (after white balance), LR just removed it
> for you.
> --
> Matthieu Moy
> https://matthieu-moy.fr/
> ____________________________________________________________
> ________________
> darktable user mailing list
> to unsubscribe send a mail to darktable-user+unsubscribe@
> lists.darktable.org
> Bill

darktable user mailing list
to unsubscribe send a mail to darktable-user+unsubscr...@lists.darktable.org

Reply via email to