Hi all, There are a lot of Path Tracing filtering algorhytims why Blender does not use them?
There is a pretty nice one, I think Blender can implement it
http://www.karsten-schwenk.de/papers/papers_radfilter.html
There are a few issues which a lot of denoising filters have (and are a major reason why Cycles does not have one).
- It can essentially mean a trade from high frequency noise to low frequency noise
- It flickers a bit during animation
- They scale up poorly with sample count (ie. will not work well with high values)
What you want are perhaps papers published less than a year ago (maybe the brand new one just released by Pixar). The results of the bleeding edge techniques tend to be much better.
Thank You Ace Dragon, but I hope You have heard about Lux Render’s improvement in Bidirectional renderin method in Path Tracing. Now it renders 10 times faster. I think Blender has to implement something more efficient and noise removal. And I think You know about Corona Rendere which uses Path Tracing. Corona is more faster than Blender and less noisier, I hope Blenders developers will do a lot nice things in upcoming versions. 
that radiance filtering looks like something to keep away from. it will certanly produce flickering in animation. I’d rather ask for Vapoursynth algorithms to be implemented as filters/modifiers in the VSE
On a second thought, this or a similar approach could be precious for the preview window: a simple checkbox “Smooth preview” to toggle it on/off.
But for such a questionable usefullness i guess no dev will spend the time needed.
To use what I like to call “ass-numbers” The Blender foundation has to be very pragmatic in what they spend their Money, Time and People on, please note each one of those is its own form of currency and is not necessarily interchangeable with the others. But for the sake of argument we will say that all together those three things are going to have a value of 10k per attempt to solve a problem just for the sake of a thought exercise.
But let’s assume that the blender team has 1000 problems to solve, and each problem costs 10k to solve, but they only have a 100k a year budget (Numbers I made up to illustrate context) Which problems would it be the most pragmatic to spend their attempts on? People like results, but they rage about failures, so most likely they will pick 6 or 7 safe bets at least and maybe one or two more ambitious ones.
So it’s not to say that no dev will spend the time needed on it, but it’s more to say that they will be likely to wait till the technology matures before working with it.
…and that’s why i said that: too questionable the usefulness to spend on it. There are other areas to work on now. I was not complaining at all.