MR - quick test

Discussions about rendering in SOFTIMAGE©
User avatar
Maximus
Posts: 1121
Joined: 09 Jun 2009, 13:45

MR - quick test

Post by Maximus » 10 Jan 2012, 01:22

Hello guys,

was just bored today and tried some things with mental ray, especially with Global illumination. You know me, i always complain about it! :)
Anyway i've setup a fairly simple scene, focusing on to achieving an translucency/subsurface scattering effect, simulating a lamp.

Everything is in scale and i actually couldnt get a clean render using FG or GI or both. That was pretty frustrating.
So i decided to switch to ctrl_ghost, and used IBL mode, irradiance particles + importons, Final Gather, unified sampling all together.
I've first used IBL + irradiance particles + importons to generate a IP map, which i then used with final gathering.

Well the result was kinda unexpected on how good it came out, so i decided to post it (the render itself is crappy but its just a test!)
Consider that with FG or GI i couldnt get a clean render like this, there was always splotches, artifacts, and crap all around, even with insane parameters.

Just to give you an overview of whats in it:

Ambient occlusion set to Color Bleed
Subsurface scattering + Translucency + checkerboard map plugged into transparency slot of architectural
1 portal light, 1 photometric light
hdri for illumination, and mia photographic exposure.
Unified sampling set at min1 max 50 quality 5

Image

this image took // INFO : RC 0.4 info : wallclock 0:05:04.28 for rendering

So for once i'm satisfied :D
This bring me to why are we still forced and struggling to use Final gathering :(
I really wish we get an overall update on the framebuffer refresh for mental ray, to use those things more flawlessly and without havin to hit refresh twice for a render, that many times cause a freeze of the software when you are using heavy scenes..

I found the "new" mr things awesome, but sadly they are poorly implemented and exposed, i really didnt expect to clean this render this quickly and at this quality. I've left the shading basic because with this "white" look you can just see artifacts from poor GI sampling, if i started with textured you wouldnt see problems if they pop up.

Anyway, enough boring you!

cheers, Max

Kzin
Posts: 432
Joined: 09 Jun 2009, 09:36

Re: MR - quick test

Post by Kzin » 10 Jan 2012, 08:35

you can get easier and faster setup if you only use ibl and ip's. its a one click solution because you dont "waste" time to setup and tune your gi, it will work out of the box (it is not as fast as arnold setup wise but way faster then vray for example).
and i agree, you can get fast results nowdays in mr, but most people dont try newer features.

why stuck with fg?
ip's only computes the diffuse gi contribution and no specular. so they dont go thru transparent surfaces or cant be reflect from mirror surfaces.
so its best to have a direct light in your scene, like portal or ibl. and its not clear what will happen to ip's in the future, if they developed further or kicked, nobody knows (there is one statement that is not really clear on this topic).

integration is a problem. for example the rendering with ip's and fg together is not integrated in the right way like it is in standalone, so what you get is not correct or lets say you should have more control of intensitys for both. we had a discussion on this topic in the mr forum. i had the problem that i could not replicate a rendering because of this, so i throw away the idea of using ip's and fg together for the moment.

juergen
Posts: 4
Joined: 09 Jun 2009, 07:56

Re: MR - quick test

Post by juergen » 10 Jan 2012, 08:37

Hi Max,
first thanks for sharing your test
maybe its just the image compression, but the floor looks a little bit noisy, I got the same result because of
unified sampling, I tested the whole weekend with it (plus ctrl ghost), first impression... unified is incredible slow compared to adaptive

and if I may ask , how you tell the ctrl ghostshader to calculate only IP and Importons?...okay, there is rebuild on off, but no rendertype IP map file only checkbox

thanks ciao j

Kzin
Posts: 432
Joined: 09 Jun 2009, 09:36

Re: MR - quick test

Post by Kzin » 10 Jan 2012, 08:50

juergen wrote: unified sampling, I tested the whole weekend with it (plus ctrl ghost), first impression... unified is incredible slow compared to adaptive
thats a typical reaction from alot of poeple, including me. but its not true at all and it comes from misunderstanding what and how unified does.
if you have alot of sampling going on in your scene, unified can easily 30 or 40 percent faster then adaptive with way better detail rendering.
you have to think different in terms of sampling compared to adaptive.but i recommend to read these articles, this will bring in some light on this topic. ;)

http://elementalray.wordpress.com/2011/11/23/unified-rendering-toyota-built/
http://elementalray.wordpress.com/2011/11/24/the-unified-rasterizer/
http://elementalray.wordpress.com/2011/11/30/unified-sampling-visually-for-the-artist/

juergen
Posts: 4
Joined: 09 Jun 2009, 07:56

Re: MR - quick test

Post by juergen » 10 Jan 2012, 09:18

Kzin, thanks for the elemental ray links, but I already bookmarked the side....
okay, have to admit I didn´ t any motiuon blur scenes..maybe here its faster with better quality,
but the overall workflow is not that complicated if I understood it correctly, use miamats, set your glossiness, AO, arealight samples to 1
and play with the unified settings...
and for me it´s really hard to get a good balance between exploding rendertimes and a nice quality...but have to play more with it :)

Kzin
Posts: 432
Joined: 09 Jun 2009, 09:36

Re: MR - quick test

Post by Kzin » 10 Jan 2012, 09:48

juergen wrote:Kzin, thanks for the elemental ray links, but I already bookmarked the side....
okay, have to admit I didn´ t any motiuon blur scenes..maybe here its faster with better quality,
but the overall workflow is not that complicated if I understood it correctly, use miamats, set your glossiness, AO, arealight samples to 1
and play with the unified settings...
and for me it´s really hard to get a good balance between exploding rendertimes and a nice quality...but have to play more with it :)
its not only motion blur, also glossis or dof or ibl. if you have a scene without alot of sampling, so simple shadows, no glossis, or no small details or fine lines, then adaptive will be a better choice. unified is not better for all shots in general.

User avatar
Maximus
Posts: 1121
Joined: 09 Jun 2009, 13:45

Re: MR - quick test

Post by Maximus » 10 Jan 2012, 16:16

juergen wrote:Hi Max,
first thanks for sharing your test
maybe its just the image compression, but the floor looks a little bit noisy, I got the same result because of
unified sampling, I tested the whole weekend with it (plus ctrl ghost), first impression... unified is incredible slow compared to adaptive

and if I may ask , how you tell the ctrl ghostshader to calculate only IP and Importons?...okay, there is rebuild on off, but no rendertype IP map file only checkbox

thanks ciao j
You need to type manually the path in there.Turn off final gathering from render manager, open ctrlghost, Activate IP activate irradiance particles, click on "rebuild" and type a path like "c:\Desktop\myIPfile" it'll save ur ip map there.
After u saved it, switch to Rebuild off, and activate final gathering in the normal rendering panel, and use "irradiance from map" this way it reads from that IP file.

Regarding the floor its not noisy to me, it might be the compression, the image is totally clean.
And as Kizin said, you need to use Unified when you have a lot of things going on, i use it with IBL system because it produces grain which normal sampling cant get rid of, also shadows/ambient occlusions and every heavy sampling, i think its better. The problem with MR to me is not the speed, is the quality.
I gladly wait 15 hours for a render but i dont want a single artifact or problem. Thats me :)

juergen
Posts: 4
Joined: 09 Jun 2009, 07:56

Re: MR - quick test

Post by juergen » 10 Jan 2012, 18:53

ok works, thanks Max