[Maxer]

I know this has been asked for before but I was thinking of something a little different this time. Instead of filtering the entire scene for noise it would be better and more efficient if the user could specify a region where Maxwell could concentrate its rendering power to clear it up faster than normal. Or perhaps Maxwell could give priority to objects that were closer to the camera so that they clear up faster than far away objects. Regardless of how it works some kind of noise reduction should be an option.

[sidenimjay]

perhaps a "focus efforts here" map....

could be a zdepth or grayscale as these can easily be generated from the host app or even just created in photoshop from a screen capture.....

at one point i remember a matte image that would block everything else from rendering or showing up in a render....dont recall what it was called tho

would be very handy

[Blitzor]

Or how about determining the value of a null pixel/subdivision prematurely by interpolating adjacent pixels based on geometry? When MR decides to render a particular pixel or subdivision of a pixel, the interpolation becomes more accurate and true. This way we don't get as much noise. An algorythm can be picked, such as gaussian to interpolate.

[Mattia Sullini]

...Instead of filtering the entire scene for noise it would be better and more efficient if the user could specify a region where Maxwell could concentrate its rendering power to clear it up faster than normal. ...

It just makes me remember of something i've seen elsewhere....

[JTB]

...Instead of filtering the entire scene for noise it would be better and more efficient if the user could specify a region where Maxwell could concentrate its rendering power to clear it up faster than normal. ...

It just makes me remember of something i've seen elsewhere....

But reproducing what's nice, is never a simple "copy" procedure... I think that if it's working, we got to have it...
It is not a matter of who did it first... not for customers.

[Maxer]

Yes I did see this feature some where else but it just makes so much since to have it in Maxwell. Let's face it noise is our number one enemy and we need some way to fight it other than bruit force.

[Mattia Sullini]

But reproducing what's nice, is never a simple "copy" procedure... I think that if it's working, we got to have it...
It is not a matter of who did it first... not for customers.

I definetively agree. That's a really useful feature. And if i were NL i wouldn't bother to try not to make it too similar...actually they still have a huge credit to claim for, there!

[ishook]

Fry has this, It's pretty useful. I think though in fry you can't region render (from 3D studio at least), so this is really the only way to render just an certain area. Maxwell CAN region render, so thats awesome.

If MR could find the noisy areas by itself, and focus its efforts on those areas, that would be neat. Wouldn't even need user input.

[Fernando Tella]

If MR could find the noisy areas by itself, and focus its efforts on those areas, that would be neat. Wouldn't even need user input.

That would be much better than Fry's method indeed.

[Tim Ellis]

If MR could find the noisy areas by itself, and focus its efforts on those areas, that would be neat. Wouldn't even need user input.

That would be much better than Fry's method indeed.

Something like a caustic searcher and refiner?

Wouldn't this increase render times though, for the image to be rendered and then refined, almost like multiple AA passes?

Edit - Perhaps an mxi refiner option next to the 'Resume' option in MXCL. So you can load an already rendered mxi and refine it.


Tim.

[Fernando Tella]

I don't know much about how Maxwell core works (let's say I know nothing at all ) but maybe it could works as a counter that works on the fly while rendering (computers are very good at that, right?); it would count how many samples (or dots, or finished rays,...) each part of the image is getting trying to level the whole result by putting more effort on the part that got less samples.

Playing the Devil's advocate against my own thoughts, I'm thinking about those multilight scenes where only physical sky is producing noise and if you turn it off everything gets clean... I guess it should count also the samples from each light source to make a good statistic and see which rays finally get to a useful part of the image and don't waste cpu efforts on not seen calculations.

Well, this may be harder than it seems. Good luck Juan and co.!