How To Avoid Moire Effects

[echelon1010000]

Hi

I have a Cube with a High-Frequency Noise and a camera wich is moving towards it. As expected Im getting awful moire effecst when rendering this animation. Has anyone an Idea how to avoid this without modifying my shader or the animation?

mayaversion: 7.0
renderer: mentalray

Greetings,
Echelon

[failure]

In general - use better sampling and more post filtering. What are you current sampling/filtering settings? Is texture procedural or image? It would be useful to post a screenshot.

[joojaa]

Oh yeah, this is the classic i want to do stuff better then worlds best signal porescesser cant do

Basicaly signalprocessing has a law that states:

To sample a signal the sampling frequency need to be atleast the frequency of the signal.

read

http://www.renderman.org/RMR/Books/sig02.course16.pdf.gz

1.1.6

explains the problematic quite clearly and it aslo clarifies how the filtering helps

Anyway to quote the paper

"Doing antialiasing well is what separates the men from the boys"

[failure]

QUOTE
"Doing antialiasing well is what separates the men from the boys"

[echelon1010000]

Hi

Thanks for your replies. I have read the specified chapter of the pdf and will read the rest of it later this day. I have also included a screenshot of my surface:

My sampling settings are here:

CODE
AA Quality:
Min Sample Level: 2
Max Sample Level: 8

MMP Filtering:
Filter: Gauss
Width: 3.0
Height: 3.0

Contrast Options:
R: 0.1
G: 0.1
B: 0.1
A: 0.1

Sample Options:
[\_] Sample Lock
[X] Jitter

Defaults:
Min Obj. Samples: -128
Max Obj. Samples: 127
Volume Samples: 1

At current, in the screenshot the effect is hardly visible. It occurs when rendering the animation and playing it. If neccessary, i will upload a short excerpt of it, so you can see it. The Shader itself is procedural: A blinn material with a noise in its color channel and some transparency.

The problem is: I have this Object with a very fine structure and the camera has to show first the whole object, and then, after moving very close to it, only a small part of the surface. It is important to show that the surface of my object is some kind of "grainy". So I'm stuck with this flickering patterns and have to find a way to make it "look good".

Greetings, Echelon

[failure]

For start, your min max samples are insanely redundant. Sample level of 8 means that in worst case scenario 4^8 = 65536 sub-pixels are calculated per one pixel. For everything you'll ever produce 0 2 is more than enough. Going over 0 for min makes no real sense because in some high contrast area of the image that's exactly what would suffice. max of 2 means 4^2 = 16 sub-pixels are calculated and averaged per pixel. Normally the number of samples that gets calculated is somewhere in between and depends on your contrast settings. Lower the contrast - more samples are taken.

Next, you should try to increase your multi-pixel filtering values. It will make your image appear softer and can kill some moire. For animations, a bit softer image is usually better, more so if the final output is for tv screen.

There is more to filtering/antialiasing, of course.

Also moire could appear because of the two overlaping transparent surfaces.

If this doesn't help try to post a video and shader perhaps.

Btw if you need to show that surface is grainy, why don't you use a bump map?

[echelon1010000]

Hi again

I will reduce the sample levels, increase my MMP Values and do a render again. After this I will post Clip and shader. I think this will last until tomorrow as I cannot do this at home (Its 6:00pm local time here and I have no access from home). As you assumed, the animation is for TV so a bit of softening the video couldn't do any harm. Until this is done I can read some more about AA and those effects, as I'm way to inexperienced to these.

What did you mean with overlaping transparent surfaces? I think there are some of those: My noisy shader sits on the front face of my cube, the rest is shaded with a semi-transparent lambert shader. This was neccessary as my object is made of glass. The side with the noise shader has a specific structure, wich allows it to generete prictures if a laserbeam is directed on it. You cannot recognize the structure itself by looking on it with your eyes ore a standard camera, but you can see that there is "something". It looks a bit like "milkglass". These thingies are called DOE ("Diffractive Optical Element"). This is also the reason why I didn't used a bumpmap. Apart from the fact that a bumpmap makes even more moirepatterns, you cannot see if the surface is unevenly or just if there are just differences in transparency or color.

Greetings Echelon

[echelon1010000]

Hi

I finally have uploaded a short clip: Clip (xvid encoded) for you to see whats my problem is. This is rendered with Multipixelsampling 4, I tried 8 also but this was way to much blurred.

Have you any idea what to do?

Greetings, Echelon

[failure]

Are you by any chance using 2d noise with repeat uv values larger than 1? If so either use repeat uv 1 1 or 3d noise.
Not directly related but for fast motion like this you'd want to use motion blur.

[knowyouremeny]

I'm a bit new, but to me, the problem appears more to do with the texturing than the AA. It goes from the texture being very small (so you can't see the repetition) to the texture being displayed so that the repetition is noticable.

[joojaa]

What you clearly need is a filter that filters out the huigher frequencies based on sample size better. Since you have a absolutey insane frequency on the images. Filtering is teh rolsroyce of AA but AA in terms of render technology tends to point elewhere.

Filtering is texture AA! And yes it has more to do with texturing like i said.

The renderer does not care what the mathematicaly correct interpretation is all it cares of is juts samples. Very often i end up filtering most of the real pattern away because at the sample ranges it IS the mathematicaly correct thing to do. Usualy what you end up doing is

THis is actualy very advanced 3d stuff if you know how to do thsi well you can allways get a job rendering if not then you have no carreer in thatkind of stuff.

Hiowever BE suyre you undertand taht NEVER EVER IN AN CASE can two polygons overlap eachother in same space. This is called coplanarity and it will allways behave BADLY. So if you copy a surface on top of another be sure to move it away from atop of the other sufficerntly far away! That or use layered textures.

Im afraid tghius is actuualy as far as we can take it without actualy fiddling the file. And yes even a BW chekkerboard should become GRAY whan viewed far away. Sufficently to say the rpoblem is NOT in your render globals.

Also no amount of sampling helps if you push a too high frequency texture into place, you will need to undertsnd fltering and theres no way around that.

PS renderes dont work like real world! so saying you must do something a way because reality works thatway is like saying fish need bicycles. It wont work untill we have compuyters that are much faster than the entire computing capacity of earth combined. Atleast the way it currently is being done.

[echelon1010000]

Hi

thanks for your reply. First let me state thate there is no surface "within" another. The object with the noise texture on in is a simple poly cube. So i think there is no problem with coplanarity.

Second thing is, maybe you didn't understand me correct because of my bad english: I'm not trying to rebuild the above described DOE structure. I'm just trying to create something wich looks like the doe structure. lighteffects and behaviour of laserbeams are animated manually. The thing I'm working on is some kind of technical visualization.

Now I will try with texture filtering. As you may have noticed I'm quite a beginner with this, so i didn't understand the deeper internals of rendering properly. Thank your for your reference to texture AA.

Greetings, Echelon

[echelon1010000]

QUOTE(failure @ 07/11/06, 06:27 AM) [snapback]242102[/snapback]
Are you by any chance using 2d noise with repeat uv values larger than 1? If so either use repeat uv 1 1 or 3d noise.
Not directly related but for fast motion like this you'd want to use motion blur.

No, I don't use the UV repeat feature. The structure is entirely generated by the noise pattern.
I already thought about motion blur, but wanted to resolve the problem with moire effects first. If I use mb until I do not get any moire effects, the entire scene would be blurred too much.

QUOTE(knowyouremeny @ 07/11/06, 04:23 PM) [snapback]242177[/snapback]
I'm a bit new, but to me, the problem appears more to do with the texturing than the AA. It goes from the texture being very small (so you can't see the repetition) to the texture being displayed so that the repetition is noticable.

This "repetition" is generated by the noisepattern, but its ok. I do not use any repetive features like uv repeat or something similar.

Greetings Echelon

[joojaa]

If you dont understand why lowering the frquence helps form the paper then this describes it well..

http://www.fundza.com/rman\_shaders/high\_fr...html#sampling\_1

However the site itself contains alot fo stuff that is totaly incorrect but thiones a good explanation

[failure]

Show us your noise settings

[echelon1010000]

QUOTE(Joojaa @ 07/12/06, 03:32 AM) [snapback]242228[/snapback]
If you dont understand why lowering the frquence helps form the paper then this describes it well..
http://www.fundza.com/rman\_shaders/high\_fr...html#sampling\_1

However the site itself contains alot fo stuff that is totaly incorrect but thiones a good explanation

Hi, it seems there is something wrong with the Webserver at this Site: My Browser (Konqueror and Firefox) tell me that it gives an invlid reply to my http-request.
However, I think I did understood that my feqency settings may be to high. (In fact they are insane high)
What I didn't understood is what to do: I don't know how to decrease the frequency and keeping a look like the high-frequency setting: How can I reproduce a fine grainy (or something) structure without the high frequency settings?

Greetings Echelon

[failure]

If grain is fine that means it's visible only when you look at it closely. When camera is far from object it doesn't need to appear grainy. Isn't the whole point of zooming in to show the fine grain which is not visible when zoomed out?

[echelon1010000]

Hi

You are right, this is my problem. I have to show the grainy surface, but only when the camera is close to the object.

I'm sorry for my late answer, yesterday we have had a power failure in my office and I was busy in reactivating our server and the network. Nothing critical bur needed some time. I will post the noise shader when I'm back in office today.

Greetings Echelon

[knowyouremeny]

Also, when you have a repeated pattern like you do, there may be irregularities in it. One area of it may be brighter or darker or more saturated or more of a certain color than the rest of the pattern, and when you have repetitions of the pattern, it CLEARLY stands out. This is very obviously happening in the last few frames of your animation. It goes from something that looks like sand to something that looks like a repeated texture.

If you're far away and a lot of aliasing occurs, you don't see this pattern as well, and it certainly isn't a perfect looking repetition. It takes samples, and those samples aren't always from the same spots on the texture. So from afar, while you should definitely get a washed-out gray like joojaa was describing, you also will get an effect of the image being slightly scattered. Anti-aliasing will take samples and blend them, making a slightly blurry reproduction that is also usually washed out (especially in the case of noise).

In my opinion, you have two problems here. One is that it's not washed out enough from a distance, and the other is that you have a wave-like effect going on. In fact, it's so bad that it almost looks like you're using Wave noise instead of something like Perlin or Space-Time noise. While I can't give any advice on the second problem, I would suggest trying to keyframe the amplitude of your noise (since you are using noise -- something that you can easily change). Not only will this give you more control over the level of contrast (going from washed out to sharp), it will cut down on render time as well (since you're making a very simple change instead of increasing filtering and/or AA).