Tweet
Hi !
i d'ont know where to post this thread, i hope i'm in the good place.
anyway,
After seeing (and using) the aeronoise tool in Fryrender engine, something is running in my head…
As it bump and make noise in my head and make me insomniac, I need to share for any advice… and perhaps it’ll have a little interest…
Well, let’s go !
The aeronoise tells me it is possible to focus the render on a part of the image. Seeing that, I believe the same thing is possible in PPT.
But !! Telling manually where to process is not really optimal imo
As Vlado says somewhere in this forum, if a render engine knows by advance where are all the difficult zones in the rendering, there is no rendering at all, image is already made… That is why aeronoise is a manual tool.
But how to say to the PPT where are the tricky zones to help him to concentrate cpu efforts where they have to be made? why not using existing techniques in vray to help it ?
Start postulate :
- Irr map makes a 3D pattern of variable sized samples, distributed depending on geometry but also light distribution and material complexity (displace, frosty, etc..)
- The Irrmap result in 2D is a pattern of samples, with different sizes, ideally placed to define the GI in the image.
- The irrmap is fast to evaluate because of its undersampling capabilities.
The global idea :
Normally, When PPT runs, it fills all the image with fired rays, increasing progressively the global sampling identically (more or less, depends of options) in the whole image. After a certain time, some areas are visually good, and some other needs more sampling.
Here is the trick :
Imagine a 2D map generated by IrrMap where, for example, each sample in the image is identified by a specific color.
When the PPT is started, it takes account of how many rays fired falls in each sample, and, let’s says for the example, when a sample have received 1000 rays, then it is marked as “finished” and PPT avoid it for the rest of the pass, focusing on samples which have not yet received 1000 rays, and so on until the last sample have received 1000 rays.
With this technique, little samples wich receive same amount of rays than bigger ones.
When all samples in the image have their 1000 rays, there are all marked “todo” and PPT starts a new sampling level.
With this technique, we are very near of an aeronoise tool, but not manually governed. This is an irrmap prepass converted in 2D map wich describes where to focus to the PPT. I think this is really more optimized in this way.
Of course it is just an idea, perhaps it is better to group different samples size in a b&w map with, let’s say 256 groups of samples, ordered by sample size, but the global idea is to give an optimization map to the PPT wich leads in faster renderings.
Does this idea make sense ??
i d'ont know where to post this thread, i hope i'm in the good place.
anyway,
After seeing (and using) the aeronoise tool in Fryrender engine, something is running in my head…
As it bump and make noise in my head and make me insomniac, I need to share for any advice… and perhaps it’ll have a little interest…
Well, let’s go !
The aeronoise tells me it is possible to focus the render on a part of the image. Seeing that, I believe the same thing is possible in PPT.
But !! Telling manually where to process is not really optimal imo
As Vlado says somewhere in this forum, if a render engine knows by advance where are all the difficult zones in the rendering, there is no rendering at all, image is already made… That is why aeronoise is a manual tool.
But how to say to the PPT where are the tricky zones to help him to concentrate cpu efforts where they have to be made? why not using existing techniques in vray to help it ?
Start postulate :
- Irr map makes a 3D pattern of variable sized samples, distributed depending on geometry but also light distribution and material complexity (displace, frosty, etc..)
- The Irrmap result in 2D is a pattern of samples, with different sizes, ideally placed to define the GI in the image.
- The irrmap is fast to evaluate because of its undersampling capabilities.
The global idea :
Normally, When PPT runs, it fills all the image with fired rays, increasing progressively the global sampling identically (more or less, depends of options) in the whole image. After a certain time, some areas are visually good, and some other needs more sampling.
Here is the trick :
Imagine a 2D map generated by IrrMap where, for example, each sample in the image is identified by a specific color.
When the PPT is started, it takes account of how many rays fired falls in each sample, and, let’s says for the example, when a sample have received 1000 rays, then it is marked as “finished” and PPT avoid it for the rest of the pass, focusing on samples which have not yet received 1000 rays, and so on until the last sample have received 1000 rays.
With this technique, little samples wich receive same amount of rays than bigger ones.
When all samples in the image have their 1000 rays, there are all marked “todo” and PPT starts a new sampling level.
With this technique, we are very near of an aeronoise tool, but not manually governed. This is an irrmap prepass converted in 2D map wich describes where to focus to the PPT. I think this is really more optimized in this way.
Of course it is just an idea, perhaps it is better to group different samples size in a b&w map with, let’s say 256 groups of samples, ordered by sample size, but the global idea is to give an optimization map to the PPT wich leads in faster renderings.
Does this idea make sense ??
