View Full Version : Need help optimising rendertime with interpolated radiosity!

03-18-2006, 09:26 AM
Hello everybody!

I'm doing a visualisation on a book shop, and would like to know if I need to increase/decrease the MES to get rid of the splotches, or do I play with the tolerance?

Rendertime with antialiasing and monte carlo took about 3 hours. I need to do a flythrough of the model, I can't wait 3 months for a 30 second video.

Thanks for all your help guys! (In advance...)

PS: Last image used MES: 1m with 0.3 Tol. It appears the bigger the MES gets, the better the image looks. It doesn't make sense to me, a little light on the subject or even a interpol radiosity tutorial will do fine!

03-18-2006, 09:36 AM
I would suggest you look into a spinning light alternative.
Radiosity won't get much faster than that with LW's renderer.

If you MUST have radiosity, fprime is your only real choice within that two month time frame.

03-18-2006, 09:51 AM
The scene has a lot (more than 50) downlight spots which have been fixed in position according to the architects plan. I've disabled some of the lights, for example every second light in a row or so.

I just need to understand how tolerance and mimimum evaluation spacing effects renders. I understand the basic theory (I think), but I want to know the concise steps to take to for ex. eliminate blotches or so.

I've tried FPrime, but it takes too long to clear up the noise with all the lights. I want to use interpol radiosity as I personally feel I can benefit from the radiosity caching!

Thanks for the reply!

03-18-2006, 10:11 AM
I've managed to stumble accross a setting that appears to be working for my scene:

Rays per evaluation: 6x18
Tolerance: 0.3
Minimum evaluation spacing: 1m (although it seems like a big amount to me)

Render time with Classic enhanced, Low is a wopping: 4min 4sec!!!

I'll be retrying a render with FPrime running it for the same time!

Holding thumbs!

03-18-2006, 02:13 PM
Was just wondering if anyone out there would be as kind as to give me a few pointers on the interpolated rendering. You know, what to and what not's.

I just feel the above image I posted was pure luck, and it didn't come out the way it did because I know my way around IR.

So help pleeeeeze!

03-18-2006, 02:36 PM
Is Microwave free?

03-18-2006, 02:37 PM
Sorry, just checked, it's not. Silly question!

03-18-2006, 02:51 PM
Hey bryphi77, I only notice now you're from Philladelphia!!! Cool, my best friend immigrated there with his wife about a year ago!

Anyways, baking sounds a cool idea, but I'm not sure what is gonna happen when LW 9 is available. I'll stay cool for a while and keep on tuning interpolated rendering which is my main concern now!

But thanks for the cool idea! Checking Evasion's site now!

May I rephrase my main question in this thread to try and make myself clearer:

Let's say I'm rendering a room that is 10mx10m across, how would I go about setting the Minimum Evaluation Spacing value for IR.

Is it normally a small value, say a tenth of 10m, i.e. 1m, or even smaller, 100th of 10m giving me 10cm, which will give me theoretically 100 sampling points accros the room in 10m should I be having the camera head on.

Do you catch my drift? I think I'm talking in circles confusing myself! Hee hee! Have mercy on me!

I need to know, is it better to set the tolerance first and then tweak the MES, or vice versa!

If you came this far reading what I wrote...thanks for your patience!

03-18-2006, 02:54 PM
Just looking at a couple of threads from Otacon! Wooow! That guy is the master of interpol rendering!

Otacon, if you're out there!!! Please help!

03-18-2006, 04:43 PM
I can help you with render settings, but I can also tell you that Interpolated probably won't be faster than Monte Carlo in an animation, unless you put up with some flickering. Interpolated is faster because it's less accurate, and those inaccuracies (lack of shadow in small crevices for example) will become quite obvious during an animation because it will be different from frame to frame.

The way to get smoothness with speed from Interpolated is to turn on motion blur ( any value above 0 ), set the the radiosity to low settings and the AA passes to high or extreme. This creates lots of fast radiosity calculations that are averaged together when the image is done.

This works great for stills where it's very hard to tell where it's inaccurate, but I've tried it with animations on extreme AA, and when I set the radiosity high enough to avoid flickering, it ended up taking longer than monte carlo ( given the fewer passes it needs to look smooth ).

I generally start interpolated settings like this:
*Ray/Evalutaion: high, like 8x24. Doesn't add much time and helps.
*Tolerance: 0.3 When set anywhere between 0 and 0.2, it almost alway takes longer than MC, and looks worse.
*Min Evaluation spacing : 1 meter
*Anti-aliasing: enhanced medium

This gives you fast test renders to get what you need for your scene - it's always different. I will usually end up reducing the Evaluation spacing to as little as 10mm, the Tolerance down to .2 and the AA up to enhanced extreme, but they all add time to the render.

I recommend doing interpolated radiosity tests with a camera in a plain room/box with an open top, white background and no lights, so you've only got 5 polygons to render and you can see any blotches. And when you've got good results, compare it to a monte carlo render.

03-19-2006, 01:16 AM
Thanks Toby!

I've left my PC running tonight, working on a 360 rotation for the shop model, and I must say, the flickering is not too bad, but in some spots its just rediculous.

I think I might try baking the illumination into the walls of the room, then I'll add gMil to the furniture and see what that does.

I've got a couple of ideas I want to try today such as reducing the lights, getting rid of the high luminous polygons (that represent lights in the roof) and replace them with proper visible lights that I can tweak.

How would you go about setting up the large 4 tube flourescent light boxes in the roof. I was thinking of using an area light with a cool bluish tint. I just need to make it visible.

Thanks for the reply. Updates will follow!!!

03-19-2006, 02:51 AM
I think baking is a good idea. With LW surface baker it would be a lot of work, but I've heard a lot of good things about Microwave, if you're doing this for a living I'm sure it would pay for itself.

Luminous polygons don't really work for lighting unless your radiosity settings are very high, so area lights are a good idea. Keep the quality setting as low as you can, if you have enough anti-aliasing passes then a low setting will be smooth. A setting of 2 is sometimes good enough, and renders MUCH faster than 4.

This sounds like a lot of hard work, let us know how it turns out!

03-19-2006, 03:12 PM
If I may ask, why is microwave better than Lightwave's baker? It appears to give you more options but what about ease of use, speed and does it look better?

03-19-2006, 08:24 PM
Thanks...I checked their website. Version 1.0 from April 03,2003 is the lastest version.. Are they planning any updates that you know of? Or is the software pretty firm and not in need of any bug fixes?

03-20-2006, 11:28 AM
I did baking today with a test room of 10mx10mx3m. Rendered Monte Carlo at over just about 40mins. After baking, turned off the lights, upped the luminosity to 100% on the walls, and rendering one frame in average 3.5seconds!

Thanks for the wonderful advice, I love baking!!!!

03-20-2006, 11:39 AM
Bryan, does Microwave bakes on Lower mesh *automatically* ?
In other words, does it creates alone a lower poly mesh, or do you have to make reduction manually and then bake on it somehow ?