For [Fuji X-Trans] raf files in particularly the enhance details occasionally, as in less then 50%, will increase the apparent resolution but on some of those it also comes with increased saturation and I wonder if the better separation between different colors (sec to increased saturation) accounts for some of that perceived sharpness. I didn't run any objective tests for sharpness nor do I intent too.
As I understand it, one of the more counterintuitive features of machine-learning is that the authors of the software aren't always able to determine what attributes in the source image the neural network has isolated and used to produce the result in the target image.
We do know that Adobe used a special training set for X-Trans files:
Enhance Details uses an extensively trained convolutional neural net (CNN) to optimize for maximum image quality. We trained a neural network to demosaic raw images using problematic examples, then leveraged a new machine learning frameworks built into the latest Mac OS and Win10 operating systems to run this network. The deep neural network for Enhance Details was trained with over a billion examples.
Each of these billion examples contained one or more of the major issues listed above that give standard demosaicing methods serious trouble. We trained two models: one for the Bayer sensors, and another for the Fujifilm X-Trans sensors.*
So presumably the enhance details feature is especially sensitive to the objectionable artifacts that are sometimes produced when Lightroom's generic sharpening algorithms operate on X-Trans files.
I haven't performed any rigorous tests, either, but my plan is to keep Iridient X-Transformer handy and try it as well as enhance details on Fuji images that don't respond well to standard Lightroom sharpening, then select whichever looks best. Somewhat cumbersome, but without knowing exactly what is happening under the hood in Lightroom and X-Transformer, I think it's going to be impossible to predict in advance which of the products will produce the better result.
———
*
Adobe Blog, Enhance Details, by Sharad Mangalick, Feb. 12, 2019
