makc
Strange Attractor
Posts: 272
|
|
« on: March 23, 2010, 08:36:18 PM » |
|
Check it out - ImageVis3D is opensource multiplatform software dealing with CT scans mostly, my initial interest in it was due to isosurface extractor: However, it also has built-in GPU renderer and primitive automation support that allows for
http://www.youtube.com/v/cHwQhR&rel=1&fs=1&hd=1. To render, for example, various Mandelbulb animations with this thing, you would only have to create data set in their format. The only shortcoming is that their raycaster does not work inside isosurface.
|
|
« Last Edit: March 23, 2010, 08:40:01 PM by makc, Reason: because I can »
|
Logged
|
|
|
|
makc
Strange Attractor
Posts: 272
|
|
« Reply #1 on: March 24, 2010, 05:41:13 PM » |
|
This actually gives rather shitty results for power 8 mandelbulb sampled at ~128x128x128 resolution (edit: turned out to be mainly due to lack of bits per voxel - see below) Same after extracting isosurface with over 300K polygons to blender: I don't see how increasing resolution could help, but will try, just to be sure.
|
|
« Last Edit: March 25, 2010, 07:03:45 AM by makc »
|
Logged
|
|
|
|
cKleinhuis
|
|
« Reply #2 on: March 24, 2010, 05:59:18 PM » |
|
isosurfaces produce extremely large files ... would you like to watch a 128x128 pixel mandelbulb ? !it is no more than an icon on your desktop fractal surfaces tend to infinity .... and the problem with 3d is, that if you double the axes, e.g. 256x256x256, you get a 8x increase in data so, to get reasonable results, you should start at experimenting with 512x512x512 resolution at least
|
|
|
Logged
|
---
divide and conquer - iterate and rule - chaos is No random!
|
|
|
hobold
Fractal Bachius
Posts: 573
|
|
« Reply #3 on: March 24, 2010, 06:24:06 PM » |
|
My own rendering experiments were based on volume data as well. All the "high definition" imagery I presented on these forums used sampling grids of at least 1000^3 voxels. Simply speaking, the voxels need to be roughly as small as the pixels of a 2d rendering.
Volume rendering fractals is a brute force approach, but it has the advantage of being very general. There is nothing that cannot be rendered. But it can be very difficult to get an accurate sampling of the volume in the first place.
|
|
|
Logged
|
|
|
|
makc
Strange Attractor
Posts: 272
|
|
« Reply #4 on: March 24, 2010, 06:36:58 PM » |
|
well, I don't mind low resolution (as in "absence of rich details") as long as isosurface comes out smooth, but it just doesn't I tried 256x256 and this gets me some pretty pictures in volume rendering mode, but isosurface still suck both in shading and smoothness: in addition, their software has bugs on my gpu, so this isn't going anywhere
|
|
|
Logged
|
|
|
|
cKleinhuis
|
|
« Reply #5 on: March 24, 2010, 06:59:22 PM » |
|
if you want "smooth" surfaces, you have to deal with lower iterations .... but hey, it is a fractal, what did you expect when increasing resolution !? the deeper you go the higher iterations you need, but for smooth look on the whole set, iteration count of 4 or 5 is enough!
|
|
|
Logged
|
---
divide and conquer - iterate and rule - chaos is No random!
|
|
|
makc
Strange Attractor
Posts: 272
|
|
« Reply #6 on: March 24, 2010, 07:05:36 PM » |
|
what did you expect when increasing resolution !? I expected some sort of smooth but bumpy surface like goose skin?
|
|
|
Logged
|
|
|
|
makc
Strange Attractor
Posts: 272
|
|
« Reply #7 on: March 24, 2010, 09:51:23 PM » |
|
I continue messing with it, log-log shading seem to give better results.
|
|
|
Logged
|
|
|
|
kram1032
|
|
« Reply #8 on: March 24, 2010, 09:58:51 PM » |
|
I expected some sort of smooth but bumpy surface like goose skin?
Nice contradiction
|
|
|
Logged
|
|
|
|
makc
Strange Attractor
Posts: 272
|
|
« Reply #9 on: March 24, 2010, 10:20:32 PM » |
|
Nice contradiction my feeling is that there is some subtle difference between "bumpy" and "absolute noise".
|
|
|
Logged
|
|
|
|
cKleinhuis
|
|
« Reply #10 on: March 24, 2010, 11:07:27 PM » |
|
Nice contradiction my feeling is that there is some subtle difference between "bumpy" and "absolute noise". hey, it is not "absolute noise" ... it is fractal noise ... this is an essential property of the objects when working with higher iterations, as i told before, try lowering you iteration depth!
|
|
|
Logged
|
---
divide and conquer - iterate and rule - chaos is No random!
|
|
|
kram1032
|
|
« Reply #11 on: March 24, 2010, 11:19:21 PM » |
|
as this is volume data, you could also try to make the density iteration-depth based. Higher iterations being more transparent. This way, you'd get the basic shape pretty clear but the increasing level of fuzzyness will be weakened. If I remember correctly, bib used soimething similar on some of his renders of the Mbulb....
|
|
|
Logged
|
|
|
|
|
makc
Strange Attractor
Posts: 272
|
|
« Reply #13 on: March 25, 2010, 02:21:12 PM » |
|
ok, so I would consider smoothness problem solved for isosurfaces. it persists, however, in volume rendering mode. why do I care? well, I was thinking about displaying several isosurfaces at the same time with some nice transparent shading but, even if you can do that by importing those in blender once, or twice, this is a problem when you need 100s of frames. and here comes volume rendering feature: they have this transfer function that essentially maps your voxel value to color/opacity. so I thought, if you leave everything transparent, and make some interval a...b slightly opaque, you would have nice colored shell in result. this is indeed true for outer shells (blue in the video below), but inner shells (yellow) are ripped into filaments, and innermost shells (not present) are dispersed in random dots that look like noise (this eventually happens to isosurfaces too, but they are able to resist rupture much longer).
http://www.youtube.com/v/nIx9AA8BDfg&rel=1&fs=1&hd=1obvious way to solve this would be to increase data resolution up to few GBs, but that defeats the purpose, because then it would be actually faster to raytrace the fractal over at every frame (which also does not have rupture problem and gives any resolution you want). another obvious solution is to use bigger a...b interval, but I personally find think shells just result in ugly blurry dirty large unnecessary spot. so the only option left is to make smooth voxel values gradient even smoother. I am currently using log-log method basically, and looking for ideas to modify it so that it would produce results similar to flood-fill near the set, gradually fading into normal shading at the distance: * makc is in a search of perfect vacuum plastic wrap
|
|
« Last Edit: March 25, 2010, 02:47:37 PM by makc, Reason: vimeo -> youtube »
|
Logged
|
|
|
|
Sockratease
|
|
« Reply #14 on: March 25, 2010, 05:04:32 PM » |
|
I actually like the translucency in that animation I'll have to look into this software when I get some spare time (in about a year or 2!).
|
|
|
Logged
|
Life is complex - It has real and imaginary components. The All New Fractal Forums is now in Public Beta Testing! Visit FractalForums.org and check it out!
|
|
|
|