When I use my iPhone I shoot in raw. Often I use the Adobe DNG Convertor to import the images into darktable. Except from some issues with exposure (the raw images are very dark) and lens correction, these DNG images work just fine.
Today however, I tried to use the ‘neural restore’, but I got an error ‘raw denoise failed for image’, is this a known limitation, I am doing something wrong or this a bug?
Are you really using a raw image? Most cellphone images are demosiac images.
It might be the case that Adobe DNG convertor demosiacs the images…
not likely.
Is that Apple Proraw by any chance? I think those are not considered raw files. If your DNG files are in effect already demosaiced, you should see that in darktable (the demosaic module won’t be active for them)
That is indeed the case.
The ‘denoise’ tab worked just fine on that image. I was actually very pleasant surprised, because darktable produced a much better image than topaz!
@anry, If I recall correctly, you are the much of the brains behind this module. Could the module be ‘image aware’ such that the ‘raw denoise’ tab is not shown if the image is not a raw image?
Apple ProRaw is actually a linear DNG, it does not have CFA data. It should have been processed by linear model in raw denoise, and it requires demosaicing done by darktable. However, I see that datktable does not support Apple RAW DNGs, at least I was not able to open RAWs from iPhone 12 Pro and iPhone 17 Pro in darktable 5.6.0 on my laptop.
Does it mean you are not importing original Apple RAWs, but converted instead?
Modules and controls in modules are shown unconditionally. We have information about unsupported images, only after we attempt the preview or processing. Starting such action without user explicit request is unnecessary costly.
That is correct, I am not able to open them directly.
Yes indeed. I export them as ‘original’ from Mac Photos (synced through iCloud). Then use the Adobe DNG converter, and the import the converted DNGs instead. This work reasonable well.
Okay, this makes sense. Although is it really that costly? I would think that it just a simply check if the data is mosaic or not? And that info is already available in darktable itself? Or is this too simplistic?
It is not about checking if data is mosaic or not. Linear DNGs should be processed by linear model. The current situation is a real bug.
But I need to warn that there will be not much help from raw denoise on computational RAWs like Apple ProRaw.
I see; the other denoise tab did a great job, actually better than topaz 
Thanks for the explanation!