Thanks, I’ll take a look.
Just a note: the approach I was using 30 years ago had no computer science aspect, and obviously no neural network involved. It was “simply” a questioning of the expert (recognized as such by his peers) to discover (and help him discover) his intellectual approach.
It is an approach essentially based on cognitive sciences (ergonomics, psychology, sociology, etc.). At the time, in 1993, when I presented the results with the work of the “expert”, which was summarized in 2 pages, I was taken for a madman… and yet.
Thank you
Jacques
And yeah, the “process during output” design, personally doesn’t have much use to me, other than curiousity. personally I think the most useful implementation of tools like denoise (really all the AI tools, but i’ve noticed artifacts, unnecessary sharpening, and what I guess I’d call HDR-style adjustments in a lot of AI tools lately) would apply the adjustments to the tools the user has activated in an app like Rawtherapee, and define the solution in the image profile/.icc file. Basically if you want it to adjust color to its best ability, but you don’t want it to adjust white balance, or i dunno, to ALSO use the shadow/mid/highlight:RGB adjustments solely, you disable hue/color adjustments you want ignored, hit the “i’m lazy, do it for me” button, and any adjustments are applied to the .icc (or whatever image profile format), instead of just being applied during the .jpg or .png rendering. So the image can be easily fine-tuned post-AI adjustment, or not really fixed because the best tools available weren’t utilized. Heck. there’s nothing wrong with an “i’m too lazy to choose specific tools, use any you want” button as well haha. Also it’s by far the most powerful, fastest way to create a data-set for the specific user. It ends up with its best guess, and also the user’s discreet corrections to the original output, ALSO also, it may allow for image data being saved/integrated into the model to be reduced, it’s at minimum, 3 “image records” per .raw file, as .icc files are miniscule compared to actual image files. Not that you’re worried about this stuff at the moment i’m sure. but i worked in software for a long time, and some of our best solutions, came from idle chatter about what we wished we could do haha. this is mostly just the stuff I usually bring up when discussing processing software… If I had to distill it down to one sentence, it’s that from the software UI/AI control, I think the fewer pre-sets, or post-output adjustments a tool offers, the less useful, powerful, or creatively honest, a tool is.