[Artistus]

Hi!
I have some ideas how to drastically improve Maxwell Render speed/quality experience.
I'd like to pass them to Maxwell Render developers.

Firstly regarding Multi-Level Extra Sampling.

I noticed that it would be very helpful in some cases to optimize rendering times and workflow using multiple Extra Sampling levels, if they were available in settings. For example, if the noise gets away quickly on some simple material surfaces or not needed background, there is basic SL applied for it, then for more complex situations (masked in custom alpha channels) - Extra Sampling 1 level, for even more noisy materials or objects - Extra Sampling 2 level and so on. It would be perfect to have about 5-7 Extra Sampling levels and to be able to add those levels by some "+" button in Settings within given range. I think having up to 5 possible Extra Sampling levels will cover needs of most users (if not all) of Maxwell Render.


Secondly regarding Adaptive Sampling and Area Denoising.

We all can see that Denoiser sometimes significantly reduces details in some cases even on highest possible settings. Especially there is an issue with transparent surfaces with refractions and even more if they are textured, have some subtle roughness maps or bump maps. I think, that it maybe could be very helpful to have Adaptive Sampling for cases where Denoiser or Extra Sampling is not worth effort. What if some algorithms from Denoiser are used to calculate noisiness of image areas gradually and create consistent virtual map of the noisiness, which was then used as a mask for Extra Sampling (or/and Denoiser)? Theoretically this could optimize the machine power to get images clear even with complex objects with caustics (and/or SSS) when other zones of the image lost the noise much earlier.
As well it would be helpful in some cases to add Deep Channel to be available to choose for Extra Sampling (level 1).

As for Area Denoising I mean using custom alphas for Denoiser to be done exclusively there. For example Denoiser could be applied on background and simple surfaces (through selected custom alpha), but on transparent objects with caustics, there are more samples calculated.


Thirdly about Combined Custom Alphas.

Here the name stands for it's meaning. You could combine few custom alphas in Combined Custom Alpha image: one calculated during rendering and a bitmap from pre-made image file, or Deep Channel (and noisiness map would be also cool if was available for combine).


Please, tell what do you think about these ideas?
And if there are some alternatives existing already in Maxwell Render, please let me know. I think it will be interesting also for others.

[luis.hijarrubia]

Actually we have a porpouse internally to do something similar. It was an interface to do a map similar to custom alpha, but done after rendering, selecting on object id or material id channels some objets / materials. And then selecting from a slider a percentaje of use of denoiser. As we find a just denoiser yes/no should have very strong edges between zones. This kind of after render maps could be used to change the extra sampling map and do a multi-level multisampling as you suggest.

[Artistus]

It's great to hear that you are already working on something what can somehow cover ideas of multi-level and adaptive sampling. Crossing my fingers for these features to come.

[bobross]

It would be nice if you could "paint" with the mouse over areas in the render view and "concentrate only more samples in these area" type system.

[sinatropus]

It would be nice if you could "paint" with the mouse over areas in the render view and "concentrate only more samples in these area" type system.

great idea. or that you could actually click on material or object on the render view to pick them for extra sampling or denoising. everything after or during the render is very welcomed.