Author Topic: How does Maxwell solve the problem of jagged edges (clamp)?  (Read 6080 times)

2017-01-10, 14:22:33

cecofuli

  • Active Users
  • **
  • Posts: 1578
    • View Profile
    • www.francescolegrenzi.com
Hello,

as we know, Corona, V-Ray, MR solve the problem of jagged edge with the Clamping option.
But, in Maxwell we havn't  this option.
Unfortunately, I'm no a MW expert. So I ask you how it solves this situztion?

(*) Special AA filter?
(*) under-the-hood clamping?
(*) ??


Thanks!

2017-01-18, 16:44:30
Reply #1

cecofuli

  • Active Users
  • **
  • Posts: 1578
    • View Profile
    • www.francescolegrenzi.com
Ok, I did a simple test.
I have ZERO knowledge of Maxwell. It's my first time with MW. So, maybe I did some errors ^__^

I create in 3d max a simple scene, with plane, teapot, and two bright arealights. Background is black.
In attachment you will have both Maxwell and Corona Scene.
In Maxwell render, you will see that the AA is ok...



But, try to save the file in 32bit (exr) and 16bit (png) and open it Photoshop.



It seems that the Maxwell frame buffer will do some AA post-processing or clamping.
And, I don't understand why the 32bit version is so bad... especially in the red lights
So, it isn't true that Maxwell AA is the better that Corona.

Now, I want to see Corona in action (PT+PT).
After 1min30 seconds we have a much better noise. Maxwell version is very noisy.



Obviously, we have AA problems (but not as much as the 32bit in Maxwell), because we didn't use any clamping.
Now, some two renderings with clamp = 10 and 5.



Something to say? Is the Maxwell VFB only 16 or 8 bit?
Isn't Maxwell able to save a 32bit exr with good AA?
I'm only curious...





« Last Edit: 2017-03-17, 16:23:02 by cecofuli »

2017-01-24, 19:30:52
Reply #2

SharpEars

  • Active Users
  • **
  • Posts: 103
    • View Profile
In all fairness, it certainly would be nice to see the quoted comparison given more rendering time (e.g., an hour in each of Maxwell and Corona) or at the same (low) noise level (which may mean giving Maxwell more time). This would help in determining which produces a more realistic and anti-aliased render.

Subjectively, it looks to me like Maxwell is going for a more realistic look (e.g., look at the contrast present in the two images and especially the dark tones) which of course means that it takes Maxwell longer to clear the noise (exaggerated by the higher contrast).

I think that this scene offer excellent composition for a comparison of quality and AA in a stark (i.e., high contrast) scene. Off the top of my head, you can play with the Sharpness setting under Tone Mapping in Maxwell to get more or less AA. There may be other ways as well.

We all know that Corona gives much lower noise for a particular render time vs Maxwell (and V-Ray). But, with regard to comparing with Maxwell, which is always slower to render and clear noise, I think that the quality and realism of the production render, given an adequate and reasonable amount of time, should be factored in to the equation. That is why I recommended giving each at least an hour (or four!) and then making the same comparison. I am fairly confident that the Maxwell version, albeit having more noise, will look (far) more realistic given a "multi-hour" render time.

BTW, it would also be interesting to compare the red reflection caustics on the other side of the teapot, although this has nothing to do with AA.


Now, I want to see Corona in action (PT+PT).
After 1min30 seconds we have a much better noise. Maxwell version is very noisy.




« Last Edit: 2017-01-24, 19:52:20 by SharpEars »

2017-01-24, 23:44:59
Reply #3

cecofuli

  • Active Users
  • **
  • Posts: 1578
    • View Profile
    • www.francescolegrenzi.com
It's the first time that I used Maxwell, so...
My goal was to understand how MW handles the clamping and the AA when we have a high contrast pixels.
In this scene, MW is more noisy. I don't want to make a noise/quality/time comparison. I have not enough knowledge in MW.
Bye

2017-01-25, 03:50:48
Reply #4

SharpEars

  • Active Users
  • **
  • Posts: 103
    • View Profile
It's the first time that I used Maxwell, so...
My goal was to understand how MW handles the clamping and the AA when we have a high contrast pixels.
In this scene, MW is more noisy. I don't want to make a noise/quality/time comparison. I have not enough knowledge in MW.
Bye

Just friendly words of advice:

1. Use the material assistants in Maxwell to generate the materials. Then tweak them.
2. If the AA looks bad, you can increase the Roughness of the material a little (very small values of 0.1-1.5) to dither the reflected hard edges. The reflection will take longer to clear (i.e., more noise), but the edges will not be as aliased (or aliased at all with enough roughness).
3. Don't overexpose. Perfect mirrors reflecting very bright direct light sources with hard diagonal edges would be hard for any renderer to anti-alias. Don't forget about the Burn setting also to help control overexposure.
« Last Edit: 2017-01-25, 03:57:06 by SharpEars »