Irradiance Particles a bit slow?

[ThE_JacO]

Irradiance Particles and the whole lot of the revamped stuff I wouldn't call production proven to be honest.
There's been some debate afaik whether to include them or not, in the end they probably ended in because people would otherwise complain, but on MI's side there's still a fair bit to iron out in there.

It's normal that it seems slow, many (clued) people agree and haven't managed to get them to behave any better.

If anybody can prove otherwise I'm all ears, because right now it simply looks like something I'm not going to be in a hurry to use.

[debris]

I got the same impressions last night when I started testing XSI7.
The "IP_Tutorial" coming in the /doc-folder lists the rendertimes for the different settings and I can't seem to reproduce them at all.
I noticed that it helped extremely to use only mia_architectural shaders for the materials when it comes to IPs. Even the simplest scenes render ridiculously slow when you use phong or lambert shaders (which you shouldn't do anyway when it comes to secondary illumination stuff).

[liltbrockie]

hmmm how very strange! :-D .... no wonder they didnt even mention it in the training videos eh lol

[ThE_JacO]

If Soft doesn't include a feature because it's not up to their standards they get bitched at because other softwares dump it in regardless of its state. If they include it people bitch that it doesn't work (not meaning you :) this thread is fine )...
There's no winning for developers when 3rd parties are involved :D

It's also possible that it's actually valid and useful but nobody has figured out exactly how or when.
I can't honestly say I devolved much time to it.

[liltbrockie]

Yeah i wasnt having a go... just an observation... i think its great they introduce experimental features for us to try!!!

[DougNicola]

...I noticed that it helped extremely to use only mia_architectural shaders for the materials when it comes to IPs...
Thanks for the tip! :)

[Jacobo]

Downloaded your scene and yes, veeerryy veeeryy sloooooww... I agree with Jac-O... Features like these are never to be trusted and used until they are developed to a point where they are more streamlined and optimized (so we're looking at MR version 4 or maybe even 5 until it becomes more robust) And production testing is also very important...

[JamesCG]

Hmmm I downloaded your scene and learned this.

After turning on Interpolation the IP Optimize phase was fine but the IP Final Pass took FOREVER. However, after it was done the render was blazing fast. I moved the camera about to see if that would cause a new IP Final Pass to take place and was pleased when it didn't (of course I turned off rebuilt map after everything was done and before I moved the camera). I could then change the Interpolation point setting to smooth out the look of the scene without any noticeable slowdown.

Im not sure if the doc tutorial if RRT (Real Render Time?) takes into account the IP Final Pass or not.

So in conclusion with Interpolation on:

Emitting Importons: Decent
IP Optimization: Ok
IP Final Pass: DEAR GODS WHY
Render: Very quick, easy to change the Interpolation Point number to smooth out things, no recalculating anything when I move the camera.

Hope this helps, I'll be playing around with this feature this weekend.

~ JamesCG

[liltbrockie]

i'm surprised to hear that moving the camera does not need a rebuild as i thought the whole idea of importons was that the rays were fired from the camera!

[Lone Deranger]

I guess it depends on the severity of the camera move and the type of geometry.
If it's a small move you might get away with it.
Doing a 360 on complex geometry probably wouldn't get you great results with interpolating the IP cache from just one frame alone.

[greyface]

Actually I'm finding them much faster than FG/GI..

[liltbrockie]

Actually I'm finding them much faster than FG/GI..

care to post an example scene to back up your claim?

[greyface]

Rebuilding on every frame is not necessary in most cases. If you set the importon diffuse high enough, the initial map should cover most angles, as long as the camera is placed in a way that it 'sees' the whole scene.

I'll be posting tests soon.

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.