Welcome to Fractal Forums

While refining my renderer a got a better understanding of some of the problems connected with rendering the mandelbulb, that is that the amount of detail varies greatly across the surface of the bulb. And often you will get a mix of well defined areas and overly smooth areas.

The reason for this can be seen when stepping up the iteration count one step at a time. Every step brings up the level of detail of a particular area one notch. This stepping effect can be particularly well observed when looking into the crevices.

But over a good nights sleep, I came up with a possible solution to this problem. In essence we do not wish to render details at a scale smaller than 1 pixel or so. In the 2D Mandelbrot, this problem takes care of itself. But on a 3D version an iterative approach can be taken.

In essence you start at say 4 iterations, the run a filter over the image which identifies smooth areas. In the smooth areas, you continue the ray marching for another iteration, and then run the filter again, (rinse and repeat). With this method it should be possible to avoid bitty regions.

I am still refining the method, but have attached an early test render:

(http://traxme.net/fractal/adaptive.png)

Don't forget to take care of eventual antialiasing :) Then you'll need to do the same for a certain amount of subpixels :)

You forgot to mention colour.  :D so far I am content with getting an accurate rendering of the surface boundaries. Here is another rendering:  Some artifacts of the method are still visible.

(http://traxme.net/fractal/adaptive-col.png)

Edit:  Colour coded the different Levels Of Detail used for clarity.

Cool idea, that's impressive.

I'm using a different technique for adaptive LOD that's not perfect but it seems to work pretty well. My raymarcher stops when the estimated distance to the fractal becomes smaller than a certain threshold, and this threshold dynamically depends on image resolution, camera field of view angle, and distance from the current point on the ray to the camera. I try to stop raymarching and render the surface when the estimated distance from the current point on the ray to the actual fractal becomes less than a pixel. Here's an example:

http://www.fractalforums.com/gallery/?sa=view;id=1192

With my method, sometimes certain areas still seem to be left too smooth and I'm guessing it's because my distance estimation is mis-estimating in those areas. I'm using a DE formula similar to the one David Makin posted on the main "true 3d" thread, but a little bit different. I haven't had a chance to try his formula yet; it might work better.

I suppose I should also try the DE algorithm. I use a home grown "minimum" distance estimation to the bailout surface:


here.x = bailout iteration count
here.y = escape distance squared after bailout (internal dot product of final Z)

I updated the image with colour coding, and it does appear very similar to other images coloured by iteration count, but having the surface determined by DE alone. It seems that DE may be the path to take.