/Off topic
Ooh, Are you willing share any of your LUTs? I am always looking for some new good ones.
Here’s the aforementioned training set: 400 images (4.43 GB). Please let me know when you have downloaded the images, so I can free up the storage. It’s not a very large server.
got it, thanks
Thanks for all the images, especially for the huge set from @bastibe.
All tests confirmed the current code so there is a pr in github
Can you please share some before/after samples?
Don’t understand this. Before / after what?
The pr has just been updated with some fixes. A am not aware of remaining issues, might also have fixed windows build issues. My tests show equal quality compared to Bayer.
Fuji friends, could you spare some of your precious time and do some hard testing?
Of special interest are
- Transitions from dark to clipped, do pixel peep here!
- Smooth transitions like partly blown out sky.
Before and after highlight reconstruction of some of the images. I am a RawTherapee user and I don’t have any experience with DarkTable. I was just curious to see the results.
@chaimav
Note: I have not yet fetched Jens-Hanno’s latest wizardry,
so here is the status from a few days back:
Addendum: image (d), below, is made using the latest Jens-Hanno-code.
a) highlight reconstruction module = off:
b) highlight reconstruction method = default (i.e. clip highlights):
c) highlight reconstruction method = inpaint opposed:
d) highlight reconstruction method = segment based:
Nice, eh?
Have fun!
Claes in Lund, Sweden
This is magic
Does anyone know if this code is easily portable to RawTherapee?
I will let Ingo @heckflosse know about it. We worked together on other stuff and I bet he can tweak an extra 10% of performance.
The code is in dt, a port should not be a problem.
Addendum: Jens-Hanno’s latest wizardry (segment-based)
is now added to the image suite above.
For that specific image, segment-based highlight recovery
seems a bit better than the inpaint opposite method.
(See, for instance, the darkblue contours around the white clouds
in the inpaint opposite method — but segment based is a bit fluffier.)
/Claes
I’ve run some tests and it seems to do pretty well, didn’t notice any flaws yet, but have only tried a few images.
There seem to be some minor bugs remaining so keep tuned in the pr