[samsam]

What I would like is for Maxwell to output an image channel that represents how 'converged' each pixel has become. Then I could use this to selectively blur/denoise those pixels that are still very noisy while leaving the more converged pixels untouched.

This would work especially well with forground DOF effects which seem to take ages to clean up.

[Kabe]

If this would be so easy to spot convergance then I would be wondering
why Maxwell would not oversample those areas itself. It could introduce
some biasing, but why waste 90% of your render times on pixels where it
doesn't make much of a difference?

Kabe

[samsam]

:Reading recent posts about "What point does a render become a waste of time?" thinking - what about an output delta channel? i.e. a channel that shows the rate of change of each pixel. Perhaps that would indicate which areas could do with more or less noise reduction? This would be relatively easy to record - its just a series of timed 'screen grabs' which would be then be interpreted/collated into their rate of change.

I assume this is impossible to do within the renderer itself as I assume it tries to solve every pixel at the same time so it can't focus on individual pixel areas.

So effectively this would be an independent 'frame buffer' grab process running in parallel with the renderer with some form of integration calculation at the end of the rendering.

[Fernando Tella]

I think it would be nice to have at the beginning a totally biased render and as the render goes on it converges to unbiased, so at the last steps we would have a more usable image. Ok, it is not completely unbiased, but is converging to it. A noisy image is not realistic too and is less usable.

[samsam]

What a great idea! - How about it NL?

[Fernando Tella]

Maybe we could even change interpolation parameters in real time.

[Kabe]

I assume this is impossible to do within the renderer itself as I assume it tries to solve every pixel at the same time so it can't focus on individual pixel areas.

This is most likely exactly the opposite:

The renderer solves only one ray at a time, which is btw the reason why
there's noise anyway. Each pixel needs a couple of rays, but there's no
dependence from other pixels afaik.

Kabe

[samsam]

Kabe - Thanks for the correction. Flunked my rendering exams as you can see.

Here's a cool link - nothing to do with rendering

http://www.spikedhumor.com/articles/177 ... uence.html