While processing the raw from Venezia from Rialto bridge., I found that the darkroom image displayed zoom artefacts that were luckily absent from the identically zoomed exported image.
Does darktable not do full-resolution processing of the visible area, using the selected method to then scale it to the preview area?
I’ll have to experiment my comments were based on what was said beginning here…
I’ll also try to review what settings I had…for this picture. I felt the export using scale factor of 1 lead to a jpg that matched the 1:1 preview …where as the jpg exported reflected the full screen preview more if exported using an export scale factor that matched that zoom…
Sure about the zoom to 100%. I asked for the lossless export for a specific reason. As you noted here you have been using the default scaling algo lanczos3. If you choose to scale the image for export you might observe the same overshoots you sometimes see while in darkroom.
This is even more obvious if your image raw file is either noisy or data have been pushed a lot as in this image. If you see artefacts like this you should concider to switch to cubic/bilinear.
There is one more thing i forgot to mention. If you switch off filmic and decrease exposure until there are no clips the reflecting white spots on the water are extremely interesting to be viewed with different demosicers. In these cases LMMSE is by far the most stable followed by RCD.
Do you think there is a predictable bias for certain types of demosaicing depending on what channels are clipped or maybe if they are all are clipped??? Just curious …I have no clue…
Many demosaicer calculate along the greens so green channel clipped will generally give the worst artefacts. The amount is strongly dependent on the local gradients but not predictable in a useful way. Just be aware of strong noise or clip-noise and use lmmse in such cases I would say.
