PDA

View Full Version : Modo's Render Engine?


anobrin
02-05-2009, 10:54 AM
Hi
What Kind of render engine does MODO use??
(Bias, etc)
from what I can see it looks just as good & Fast as V-RAY

Grolicus
02-05-2009, 05:21 PM
I'm no expert whatsoever, but I think it's biased, multithreaded, does bucket rendering, and uses irradiance caching for radiosity calculations.
Render quality could stem amongst other things from the fact that a lot of physically correctness is applied, for example with lights.
It also has a fast built-in multithreaded render previewer that uses another technology (progressively refines).

Check out some videos under the Rendering tab on Luxology.TV .

anobrin
02-05-2009, 05:36 PM
Thanks I'll check them out.

Novakog
02-18-2009, 06:40 AM
That's a pretty accurate description - it's a biased raytracer, although I think if you use the correct settings it could be considered capable of unbiasedness (it can do Monte Carlo global illumination instead of IC, energy conservation on materials, etc.)?

If you want examples of renders, the Lux gallery (available here (http://www.luxology.com/gallery/image/)) also has I think around 2000 images.

CaptainObvious
03-01-2009, 08:11 PM
The only features I can think of in modo that's biased would be the irradiance caching and the refinement shading. Beyond that, it's largely unbiased.

I would like to point out that "unbiased" does not mean what most people seem to think it means. It has nothing to do with rendering at all, least of all physically accurate rendering. It just means that the rendering converges to statistically accurate results. Irradiance caching can be physically accurate, but it is always biased. Per-pixel brute force GI can be physically inaccurate, but it is still unbiased.

modo's renderer is basically a ray tracer. It supports scanline rendering as well, but there are a lot of features that are unsupported by the scanline rendering (such as instancing). It defaults to an "automatic" mode that picks scanline rendering as long as possible, but switches to ray tracing when you're using features that require it.

It uses bucket rendering, and has a buckets-to-disk feature that lets you render images without keeping the final thing in memory (which basically means you can render images of any size; the maximum I've done was 32k by 32k on a machine with 1.5 GB of RAM).

Novakog
03-02-2009, 12:38 AM
The only features I can think of in modo that's biased would be the irradiance caching and the refinement shading. Beyond that, it's largely unbiased.

I would like to point out that "unbiased" does not mean what most people seem to think it means. It has nothing to do with rendering at all, least of all physically accurate rendering. It just means that the rendering converges to statistically accurate results. Irradiance caching can be physically accurate, but it is always biased. Per-pixel brute force GI can be physically inaccurate, but it is still unbiased.



Yeah you're right - I'm embarrassed that I only realized that bias was referring to it in the statistical sense about a week ago. Specifically, "statistically accurate results" mean the results you would get with infinite samples at every pixel on the screen (so Monte Carlo sampling, by definition, converges on this).

Mike RB
03-19-2009, 11:09 PM
Unbiased is often misused as meaning progressive rendering with tons of noisy samples. One has nothing to do with the other.

ngrava
03-28-2009, 08:25 PM
I'm also embarresed to admit that I don't really know the correct answer to this. I own the "realistic image synthesis" book but I've never read the whole thing. I always thought that a "Biased" renderer was one that was going to try and find the areas of light in your scene and "Bias" the rays in those directions while sampling for GI. Unbiased would then mean that the rays are just shot out in random directions without paying any attention to where the light was coming from thus giving you an "Unbiased" GI evaluation. But I admit that I might be wrong about that...

A more technical explanation would go something like this: a ray is traced from the camera into the scene. When it hits an object, An imaginary polygon hemisphere is placed around that point. Then additional rays are shot from that point say, 8 to start with, through the vertices of hemisphere. When those rays encounter other objects, they analyze the light value at that point on the surface of that object. If that value is above a certain threshold, then we go back to the original sample point and shoot even more rays in that direction (I think this is what is referred to as "Importance Sampling"). Otherwise, we move on to the secondary GI bounce calculations where this process is repeated a number of times for each of those biased rays moving from surface to surface. Each time we sample a point on a surface, that value is collected and averaged with all the other samples and that becomes the "GI contribution" value for that original point that we first sampled. This is done for every pixel in the image. The importance sampling part of this allows the renderer to save time by only concentrating on the areas of the scene that are going to contribute the most to the GI calc. Otherwise, to an unbiased renderer, every point sampled in the scene is important. If you continue shooting rays through every point in the original hemisphere, you will eventually be getting a correct evaluation of the lighting at that point. Of course, this causes a problem for time crucial endeavors. ;)

Now, I have to admit that I'm starting to have my doubts about whether this actually constitutes a biased or unbiased render. I'v seen several renderers that claim to be "Unbiased" yet are clearly doing some sort of importance sampling. Especially those that use Metropolis Light Transport. This seems to suggest that maybe using irradience cacheing or some other type of interpolation is what caused a renderer to be unbiased.

Basically the way I understand Irradience caching to work is that instead of doing the above process for every pixel in the image, it's done for every 4th or 8th pixel (or however you want to divid up the image) in a pre-pass. Then when we start to render the actual image, each ray per pixel in the image looks at the nearby surrounding samples and averages them together to form the GI contribution at that point. That definitely sounds like a biased method. But, if that's the case, then any renderer that doesn't use sample averaging could potentially call themselves "Unbiased"? Is that right?

CaptainObvious
03-29-2009, 12:44 AM
I always thought that a "Biased" renderer was one that was going to try and find the areas of light in your scene and "Bias" the rays in those directions while sampling for GI. Unbiased would then mean that the rays are just shot out in random directions without paying any attention to where the light was coming from thus giving you an "Unbiased" GI evaluation. But I admit that I might be wrong about that...
Well, you're not exactly wrong. Contrast-based supersampling of GI sampling would indeed be a biased method (modo has a checkbox for whether or not you want this, actually). It typically decreases splotches, but it generally biases the result towards darker. In fact, almost most adaptive sampling will generally bias the result towards whatever is more common in your image. Adaptive anti-aliasing of an image that contains mostly dark pixels will generally make the entire thing darker on average.

Imagine that you have a class of students taking a test, getting grades from 1 to 5, and you set the difficulty so that the average result is approximately 3. If you then make everyone who got a 5 retake the test, the average result is likely to be lower than it was before. The same thing is true with adaptive sampling. By finding pixels (or cells in a hemisphere for GI) that are very different from the rest and sending more rays in that particular direction, the most likely outcome is that the general result will be closer to the average of the rest of the cells. Generally speaking, this means darker, which would then be biased.

Supposedly, MLT is unbiased, but I don't know enough about the algorithm to comment on it and I'm too tired to read the paper on it at the moment.

ngrava
03-29-2009, 07:14 AM
I removed my original post because it was just plane wrong.

I looked over MLT today and discovered that it's based on statistical analysis which is unbiased and not biased as I had originally thought. Sorry, you're right.

Navstar
04-10-2009, 03:21 PM
Is modo 64-bit aware?

ngrava
04-10-2009, 04:23 PM
The PC version will be when 401 comes out. the 64 bit version for Mac will follow some time latter due to having to port everything to Cocoa.

Navstar
04-10-2009, 07:25 PM
That's surprising about the Mac version. It "feels" like a next-gen 3D app. I assumed it was already Cocoa.

CGTalk Moderation
04-10-2009, 07:25 PM
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.