Need help: trying to get rid of rendering artefact ...

[cytotox]

Hi

I am new to this forum and still fairly new to 3D fractal rendering and this impressive software (quite addictive - I wish I had more time). I have have been rendering fractal images on and of for about 20 years (starting with fractint and later using ultrafractal), and I also have a basic understanding concerning complex number math. However, I don't understand much about the concepts of raytracing.

Having said that, is somebody here able to help me getting rid of rendering artefacts? I have tried a lot of different settings for the diverse calculation parameters, but generally the artefacts simply move to other parts of the image. My best bet so far was to disable the "Vary DE stop on FOV" button 

The artefacts are solid-looking planes with fragmented edges as shown in the example image (partial zoom)

Mandelbulb3Dv16{
J.....K0...r3...O/....M....AzOG9zdQ6.1.......U/EQZKDk8u6U.gBm85Fcf/6.LTDw9pX840k
................................Wc7zSk0Dy.2........Y./..................I/2.....
.................QE/3....6k69...12....E/.....Inf3a66MvjD.QEq0l4E...s.pAnAx1....Q
/7....kD....1..........wz.......................................................
........jg5wKljtJwXP7I0JVrvwyG4vvrOJANjDmEfbt2Hdxv1q1Tf47i/2zs1UhL6Krqkjoh7INuIl
4vP4J9XysHv1zYwywBJuI1lD......YdE...I3....................E.....................
..............................nxDC.krzs.EVzX1.CyDC.Qtzs..czX1.dyDC..............
................/..........U..6.B/...U0...UB....h....6/...EA....5/...I1....UM016
...U.0kEnOvwcqlx.gYmw9eO0.6.0c..U1Cs.El18XGQeGyDjvIRhrVAkz1............28.kFrA0.
FWb96aAIVzvh1se7Umvxz0...........6U0.wzzz1................................E.0c..
zzzz.................................2U.8.kzzzD.................................
/6U0.wzzz1...................................2CcN/UvNP6.eeWCNq0.yRii.EJJUk1f..XR
S1.mx3CcN/UvNP6.QsLsUa3.ibhV..bTV1OK.sSq40.ly3CcN/UvNP6.MwLsUa3.ibhV.kqTV1OK.sSq
40.kz3CcN/UvNP6...EsUa3.eeWCNq0.IJ36wk8.wyLsUa3.................................
E....6....E.....2....A....EEh3aSdtqNU6oPs/UQ..........................k/9.......
.....................NaNaNaNaNyD........kz1.....................................
................................................................................
.....................2...../....1....2IPVdLOiR460x4S............................
.Qk0.........................UzjOaNaNaNaty1........../..........................
................................................................................
................................}

[bib]

I would try to reduce the Raystep multplier, but as you have surfaces in this fractal that get thinner and thinner, you will probably see more of the surface, but still with artifacts on the edges...

[Jesse]

Bib was faster    and right, the problem here are the thin slides, and to not "overstep" you have to reduce the Raystep multiplier.
Maybe to down to 0.2, depends on when you are lucky with the result and how much time you have to render.
And i recommend to keep the "Vary DE stop on FOV" option and leave the DEstop value at a common amount.

[bib]

hehe
In fact, it's Prokofiev who explained me what "overstepping" means, just 2 weeks ago

[cytotox]

Thank you for the quick responses!

Bib was faster    and right, the problem here are the thin slides, and to not "overstep" you have to reduce the Raystep multiplier.
Maybe to down to 0.2, depends on when you are lucky with the result and how much time you have to render.
And i recommend to keep the "Vary DE stop on FOV" option and leave the DEstop value at a common amount.

I re-rendered as suggested ("Vary DE stop on FOV" option active; DEstop value set to 0.5), setting the Raystep Multiplier to 0.05. However, the amount of artefacts stayed the same for the overall image (only shifted to other parts; you may compare the new zoom-in with the one posted previously).

So, the bottom line is that these artefacts are unavoidable (and not a bug in the program)? And do I understand it correctly that they are resulting from the way the surface of the fractal is defined / calculated using the distance estimation in conjunction (?) with the way the raymarching algorithm approaches this surface and decides where the surface is 'hit'? 

(as I stated above, I have used distance estimation in UF on 2D M-set fractals without observing any artefacts. So, if I understand this right, either the 3D nature of the Mandelbulber-coded fractals and / or the raymarching process seem to necessitate somewhat arbitrary decisions, resulting in discontinuities ...)

[Jesse]

The 3d fractal generation differs here from 2d in the way the surface must be found.
Normally you compute for 2d only every pixel to find if it belongs to the set or not, in 3d a common way is
to find for each pixel where the viewing ray hits the surface.  So you step from the camera position with
distance estimates until you hit a surface or until the maximum defined raylength is reached.

But because the distance estimates (DE's) are not absolute accurate, dependend also on the formula and its
parameters, thin slides can be just overstepped because the DE was to big in this case.

Another point is especially for the Box and IFS with high scales equally or above 2, that the real shape is like
dust and when you approach the surface, the object seems to vanish more and more because the iteration
count increases and the object get holes where it seemed to be solid before.

So there are some effects and these thin structures are difficult for raycasting algorithms to render, at least
for the algorithm m3d uses, there might be other ways that have not that problems.

Another effect is that some formulas have steps when calculating the DE function with the 4 point method,
what could lead to noise in the image where no object is.  That can be avoided by settin the minimum
iteration count to higher values, but lower than the average iteration counts so that the object itself is not
affected.

[hobold]

There might be a way to to change the nature of such artifacts, and turn them into noise rather than stripes/moire/wave interference. This is not something the users can do, but programmers can try it.

When you start marching along a ray, treat the very first step specially. Rather than using the DE as usual, multiply it with a random number between 0.0 and 1.0; use different random numbers for every ray. That way, the rays are no longer traced in lockstep, because each one is slightly out of sync with the others. When they finally hit the fractal surface, most potential artifacts should be distributed randomly across all rays.

Noise is an artifact, too, of course. But often the human visual system is much more tolerant of noise than it is of very regular artifacts.

[cytotox]

<...>
Noise is an artifact, too, of course. But often the human visual system is much more tolerant of noise than it is of very regular artifacts.

If this is possible, it would be nice to have this option. As most people are rendering their fractal images at higher resolution anyway before downsampling, the use of anti-aliasing during this step would also contribute to smoothing out this noise. I'm not sure, but the result should give a less frayed appearance to the sheets' edges; instead they should look more like fading gradually ...

(possibly the edges would appear to be partially translucent)

[Jesse]

That is true, noise is very natural, though not often seen as a object structure itself.

A little problem might be that it becomes more obvious by the darkening from the ambient shadow,
like seen in the attached examples.
Edit: this could also be a result from the corrupted normals and/or roughness estimates on the thin slide.

A sideeffect seen in the right upper part is that the artifacts from the low binary search count is
also reduced, but these are preview settings, a value of 8 would do it better.
 

[cytotox]

... this could also be a result from the corrupted normals and/or roughness estimates on the thin slide.

A sideeffect seen in the right upper part is that the artifacts from the low binary search count is
also reduced, but these are preview settings, a value of 8 would do it better.
 

Hello Jesse

I'm sorry but I do not understand what you mean here. Are you suggesting to raise the value of the 'Stepcount for binary search' parameter from 5 to 8 (or more) to get the less frayed appearance seen in the lower picture? I tried that, but could not reproduce your result (and what about the 'Smooth normals' parameter?) ...

Can you post the parameter file or tell me which settings you changed to obtain the result shown in the lower picture?

[Jesse]

Yes, the 'Stepcount for binary search' parameter should be 8 to 12 for the final image, but this don't changes the overstepping artifacts seen in this images.
The last image is made with hobold's suggestion to give the artifacts a random look and is available in the next program version, wait some days. 
And the smooth normals parameter is affecting the phong shading and the main effect is that it reduces the specular light amount at edges to reduce moire like effects.  More than 3 does not change very much and does not change the overstepping too, unfortunately.

[cytotox]

...
The last image is made with hobold's suggestion to give the artifacts a random look and is available in the next program version, wait some days
...

Hey, that's cool, man!