Chaos Corona Forum

Official Stuff => News => Topic started by: Ondra on 2014-05-29, 16:14:43

Title: Blog: progressive importance sampling
Post by: Ondra on 2014-05-29, 16:14:43
https://corona-renderer.com/blog/research-corner-progressive-importance-sampling/
Title: Re: Blog: progressive importance sampling
Post by: jjaz82 on 2014-05-29, 16:52:58
wow .. congratulations for the excellent work..

Research is the vital point of these software

keep going :)
Title: Re: Blog: progressive importance sampling
Post by: blackice on 2014-05-29, 17:45:43
well done!
>B.
Title: Re: Blog: progressive importance sampling
Post by: maru on 2014-05-29, 17:51:36
Looking at things like this makes me think my school was such a huge waste of time and money. :(
Title: Re: Blog: progressive importance sampling
Post by: Avan on 2014-05-30, 00:13:26
Simple for understand, but it looks amazing ! Can we see the feature in bulds?
Title: Re: Blog: progressive importance sampling
Post by: crazyman on 2014-05-30, 00:44:46
Keymaster:

- Can you set custom amount pings or pongs will algorithm perform, before it start path tracing? Or is it predefined values defined by algorithm being used?
- How this affect animations? Does importons/photons calculated once for all frames while doing static animations?
- What if objects changing it's position/material properties while animated? Does algorithm need to calculate all pings-pongs for every frame?
- If so does it produce unwanted "pixel shimmering" ?
Title: Re: Blog: progressive importance sampling
Post by: Ondra on 2014-05-30, 00:59:17
the number of iterations is configurable. We ended with something like 30 batches of 1 million photons+importons each.

Animations, shimmering, etc. works exactly the same as other unbiased algorithms. We did not do cache sharing between frames, although it would be possible.
Title: Re: Blog: progressive importance sampling
Post by: juang3d on 2014-06-03, 10:17:10
So if I understood correctly, this won't flicker with animation, right? because you are not using photons to light, but to drive the paths, is this correct?

It seems to be a pretty good system.

Cheers.
Title: Re: Blog: progressive importance sampling
Post by: Ondra on 2014-06-03, 10:36:58
It would not flicker, but it would shimmer
Title: Re: Blog: progressive importance sampling
Post by: crazyman on 2014-06-03, 12:06:47
I think you can avoid shimmering in maxwell, if all tied to same seed number. Am I thinking right direction?
Title: Re: Blog: progressive importance sampling
Post by: rampally on 2014-06-03, 14:05:21
It would not flicker, but it would shimmer
can  you avoid that shimmering too!!!
Title: Re: Blog: progressive importance sampling
Post by: crazyman on 2014-06-03, 14:08:16
It would not flicker, but it would shimmer
can  you avoid that shimmering too!!!

wut?
Title: Re: Blog: progressive importance sampling
Post by: juang3d on 2014-06-04, 13:06:51
It will shimmer?

What do you mean? it will be brighter over time or something like that?

Cheers.
Title: Re: Blog: progressive importance sampling
Post by: Ondra on 2014-06-04, 13:14:37
No, because it is unbiased. The pixel values will oscillate between the true values. It will demonstrate as a light uniform high-frequency noise
Title: Re: Blog: progressive importance sampling
Post by: juang3d on 2014-06-04, 18:13:48
So with enough time, it will be clean, right?

Do you think that there could be a solution to avoid that noise? (In the future)

Cheers.
Title: Re: Blog: progressive importance sampling
Post by: iliyang on 2014-06-04, 21:35:01
The method is indeed unbiased, however calling it "progressive importance sampling" is slightly misleading. The term would be appropriate if the distribution caching and training was actually done during the whole rendering process. Currently, it's run as a pre-pass.
Title: Re: Blog: progressive importance sampling
Post by: Ondra on 2014-06-05, 00:10:31
yep, I had to take quite a few liberties to change the name into something an artist would actually remember ;)
Title: Re: Blog: progressive importance sampling
Post by: Javadevil on 2014-07-12, 04:00:30


Awesome, really interesting combining all these different methods into one to get  faster and accurate results.

Nice work