We have recently introduced a new denosier in Lighroom that is based on machine learning that was pioneered by Michael Gharbi before he left Adobe to found a startup. Since then, Zhihao Xia and I have been the research leads behind this technology, and we have recently extended the denoiser to support more input types.

While extending the support to more input types sounds straightforward at first, it is rather tricky to pull of. After all, domain gaps are a common nuisance in machine learning and it is easy to come up with a denoiser that works well on the training data but fails to do a good job on real-world imagery. And there is a high variance in how noise looks in the real world, it can even look different in one part of an image than in another due to the sensor having different temperatures in those regions.