Rotational Motion Blur and Overlapping Geometry Samples...

Does anyone know how to over come the inconsistent and overlapping of geometry samples in the search of smooth rotational 3d motion blur? As seen in the attached images the more geometry samples you apply the more overlapping artifacts you get. It is only at 20 geometry samples that it is 100% clear (and that only applies to this one scene), but then you get the overlapping artifacts again once you go above 20 samples. What is the magic calculation to know how many samples that are needed to achieve smooth non-overlapping rotational motion blur? The other concerns are the render time explosions, the missed bucket from the DMC sampler (more easily noticed as you increase the gamma), and the fact that the required samples for smooth results is different in every scene I have tested. This issue was also outlined by jbeau3d here, but no help was provided to work out the issue.

All images are rendered with the same settings and the only thing changed is the geometry samples in the Camera Rollout of the rendered.

-Eric





Is this being investigated or is there a known work around? This is a serious issue which prevents me from using your product in production.

-Eric

It works fine for me here on a simple test. If you have a scene that you can either post here or send it to me at vlado@chaosgroup.com I will be able to tell you more.

Best regards,
Vlado

Files sent. If you have any questions, workarounds, or solutions please post back here so anyone else who may encounter this can read the findings.

Thanks,
-Eric

oddly enough I was able to reproduce this problem too. But only with having 24 geo samples, with 10 samples it renders fine. It seems as soon as you go over 10, you start seeing this issue. In regards to render time, These renders only took seconds.


If I could attach a .gif you could see my problem a little easier. It starts at 4 samples (before 4 samples the shape of the motion blur is awkward and incorrect), and as I said it really depends on the scene setup. As you add more geo samples it becomes more pronounced as you have more overlapping geo samples. If you alter your angle to a lower angle it will probably be more pronounced as well, because you will have more geo samples overlapping. Also, if you add more AA samples it will be more noticeable. As it is now the noise of the AA sampling may lead you to believe it isn’t there, even though it is.

On the render times, are you using DMC or Adaptive Sampling? What happens if you increase the AA to get a smooth non-noisy image? Also, my scenes consist of millions of triangles for the complete scene so it gets rather complex real quick.

-Eric

what you might want to look into is per object motion blur geometry samples. You don’t need to blur everything in your scene with 20 geo samples. For that, in vray object properties there is a section for per object geo samples.

Forgot to say, for the render time I use adaptive DMC 2/16 typically. But you need to increase the motion blur samples as well as work with the adaptivity (lower is better) for quality.

I also find that adaptive subidivision produces smoother motion blur, though its not as accurate as aDMC.

Yes, I know but a lot of the detail is in the rotor system. So you can only adjust it so much for high speed portions. Also, I think the speed problem comes from Bugs in the DMC engine. There is no reason for the render time jumps (25%-75% increase shouldn’t be normal):

Geometry Samples - Rendertime (hh:mm:ss.s) | Percentage of rendertime increase

02 - 00:00:39.7 | 0.00%
03 - 00:00:38.2 |- 3.93%
04 - 00:00:37.7 |- 1.33%
05 - 00:00:43.0 | 12.33%
06 - 00:00:46.5 | 7.53%
07 - 00:01:02.2 | 25.24%
08 - 00:01:11.9 | 13.49%
09 - 00:01:31.4 | 21.33%
10 - 00:02:02.7 | 25.51%
11 - 00:03:03.3 | 33.06%
12 - 00:03:24.5 | 10.37%
13 - 00:04:38.3 | 26.52%
14 - 00:09:10.0 | 49.40%
15 - 00:10:45.6 | 14.81%
16 - 00:17:12.3 | 37.46%
17 - 00:23:34.5 | 27.02%
18 - 00:28:10.3 | 16.32%
19 - 00:43:07.0 | 34.66%
20 - 02:35:42.4 | 72.31%
21 - 02:19:02.8 |-11.98%
22 - 02:34:12.7 | 9.83%
23 - 02:47:31.4 | 7.95%
24 - 02:41:15.5 |- 3.89%

-Eric

when I did several shots with helicopter in them I faced similar issues. What I ended up doing is only applying multiple geo samples to the actual rotor blades them selves. Not ideal, but the rest of the detail looked ok with just the two samples.

Nope, it doesn’t.

There is no reason for the render time jumps (25%-75% increase shouldn’t be normal)Actually there is :slight_smile: It has nothing to do with the DMC sampling engine though.

I’m still looking into this; it looks more like an issue with getting the geometry from 3ds Max at sub-frame times.

Best regards,
Vlado

Good to know the issue isn’t related to DMC.

What about the missing bucket though? It is noticeable in all the renders I sent you, but more noticeable if you set the image to gamma 2.2 (like in the attached image).

-Eric

The motion is very fast and the initial 1 sample per pixel for the bucket does not hit any part of the object, so V-Ray does not place any more samples there. To fix this, increase the min. subdivs for the image sampler (2 seemed to work well for me) or increase the motion blur subdivs above the max. subdivs value for the AA.

Best regards,
Vlado

Thanks for the info still trying to get to DMC and what are the real values being used and the adjustments that have to be made in various areas to try and get the best results. Even after reading the Demystifying V-ray DMC Sampler I am not sure what more samples should go and what the best values are to start with. If I am using motion blur in most of my work do I put more samples into the DMC Sampler or the Motion Blur subdivs? Or should I start with the same DMC samples for static and motion blur images, and increase the values for the subdivs as needed?

One other thing while we are on the topic of Motion Blur, does Vray RT in 3ds Max support motion blur? If it does how do you enable it? If it doesn’t why isn’t it supported and are there any plans to add support for it like you have in Maya Vray RT?

-Eric

Quick one for you vlado - how does vray calculate motion blur per object? Does it kind of make a copy of the object for each of the time samples you’ve set and then trace vectors for each vertex to calculate their trajectory and blur along that? Or does it make copies of the object for each time sample, render it and make a sample set out of the combined data for that object? I’m writing a small thing on motion blur and I wanted to get a clear understanding of what it’s doing and it’s memory implications - for example if you’ve got four subdivs is it making four copies of the geometry or using another method?

Cheers!

John

It is mostly a matter of preference; I myself prefer to start from adjusting the image sampler alone and only play with the other settings if there is any need for it, e.g. to reduce render times or bucket artifacts.

One other thing while we are on the topic of Motion Blur, does Vray RT in 3ds Max support motion blur?Inside 3ds Max - no (we don’t do this inside Maya either). The reason being that jumping back and forth through the timeline while the ActiveShade is open would be a) slow and b) very error-prone. However you can still export an animated .vrscene file and render it with the standalone version of V-Ray. How successful this is going to be, I’m not very certain - the Maya .vrscene exporter is way more advanced.

Best regards,
Vlado

This is annoying.

Lack of MB is one thing that makes me use RT less.

Any word on the initial issue of stepping in motion blur? This is preventing me from using Vray.

-Eric

I looked at it; initially I did not find anything wrong on the V-Ray side except that the object rotates somewhat too fast for the 3ds Max internal time resolution (3ds Max has only 160 ticks in a frame; they are not enough to sample accurately sub-frame positions for quickly moving objects). I came to the conclusion that we may need to do some internal interpolation between the ticks; hopefully we’ll manage to implement this for the next service pack.

Best regards,
Vlado

It is using real world rotation RPM values. The real world system has a max speed of 395 RPM or 2370 degrees/second (calculated by a conversion tools such as this). And 2370 degrees / 24 frames results in 98.75 degrees/frame or 0.6171875 degrees/tick. So I guess the question is how many degrees/tick is the max to get non-artifacts?

-Eric

So this now fixed for the service pack in December. If you would like to get the fix before that, please email me to vlado@chaosgroup.com

Best regards,
Vlado