In the attached image you can see the a 200% zoom of the typical test image from DPReview, taken with a X-T3 and exported as 16bit tiff.
On the left, the DxO one and on the right the Darktable.
Both the pictures have just the basic edit applied, as CA, lens correction, a bit of sharpness highlights and shadows correction.
As you can see, in the DxO version, you can clearly read the little text but in the DT one you don’t.
I’ve played with demosaic a bit and other settings without getting it better.
After playing around more with DxO photolab, I noticed that INSIDE the editor, playing with the raw file, the preview I get is IDENTICAL to the one from Darktable, zoomed at 200%.
The differences emerge once the raw is exported to TIFF!!
The darktable tiff is identical to the preview I get during editing, the DxO one isn’t!
The DxO tiff is clearly a lot better in fine details like these. What could it do under the hood?
I know that is a “pixel peeper” thing but I’m intrigued by these kind of things and curious about it.
I suspect that DxO is using machine learning. I don’t think it is a coincidence that DxO is supporting x-trans after they developed Deep Prime.
Deep prime, AFAIK, uses machine learning for combined demosaicing and denoising. The demosaicing results for bayer sensors is better with deep prime on.
Ok, I didn’t know that. Thanks!
I’ve tried and it’s the deep prime denoise that produces those better details.
The difference is massive.
Would be able to see something like that within darktable in the future?
Is it something doable? Commercial software are moving that way and it could be a game changer in many aspects.
