nardes wrote in post #18863529
Yes - I used the Topaz DeNoise AI Plug In in PS CC.
Topaz recommend that you use the Plug In on the Raw File before you perform any other processing as the Plug In works best on the Raw sensor data according to Topaz.
I've read people mentioning that, but how is it done? Is it a plug-in that goes into a specific RAW converter, or does it process a RAW file itself in a folder and change it before it goes into any converter? Does it have it's own simple converter?
Noise is certainly visible on the original Raw file when viewed at 100%,
You said "dark museum lighting" and also "ISO 12800", but the dark is redundant with the ISO and we know the amount of light is low; what matters is the quality of light, especially its color. The real reason that many places which require high ISOs are much noisier than studio shots at the same ISO is that one or more color channels are extremely weak, and are virtually under-exposed. Indoor traditional incandescent is extremely lacking in blue light, and outdoor shade on a sunny day is very lacking in red light. I get less visible noise in 100% crops shooting birds on top of a bush in an open area under an overcast sky at ISO 6400 than I do in the shade at ISO 2500. If you were operating with well-diffused magenta light, you would get the least noise (especially chromatic) at any given high ISO with Bayer CFA cameras..
but Topaz does an excellent job of mitigating it without destroying the genuine fine detail, such as seen on the scales of the Green Tree Snake, which was located in a glass cabinet smeared with dirty finger prints and illuminated by horrible mixed lighting.
It is probably doing a lot more AI than most basic noise reductions. What I see with a lot of noise reductions is that they still assume that the areas of detail required are the finest ones possible at 100% pixel view, but what they really need to do is assume that as the capture gets very weak, the target detail should be coarser, and extremely coarse when the image looks like mostly noise. Most NR assumes that the noise is signal, apparently, with very weak captures. From your NASA examples, it looks like Topaz rejects fine dots of detail, but sharpens edges, which is how I think a good AI noise reduction should work. Keeping edges sharp does not require keeping anywhere near as much noise as keeping dots sharp.