Anybody has good materials/knowledge or experience in this field. Please share! ?

I would also like material with a good introduction. Like smooth boundaries would just lead to reflection and refraction, which raytracing is great for, and do not lead to noise. Pure diffuse surfaces - oh - need path tracer for this (or KD-tree?). What is the base for this denoise filter? I start with Metropolis-Algorithm and for me it seems ideal in finding the important samples. But somehow this algorithm does not know anything about the error margin, or does it? Is there some rigorous approach towards an error margin which also uses internal state from the tracer?

With error I can just low pass filter in a post-process until the error is below 1% for all pixels. Some parts of the scene will lose detail.

I mean, if metropolis detects that it needs more samples at some point, it transforms the path into a nearby path to get them. So noise should be homogen across the picture.