Welcome to Fractal Forums

All my renders so far have only used direct illumination, but for ages I've wanted to implement Global Illumination. It turns out not to be quite so tricky as I thought, but rendering times certainly suffer - something around 3 to 10x slower! My main immediate incentive for adding GI was to improve my Xmas Treehouse (http://dspwhite.deviantart.com/art/Mandelbulb-Xmas-Treehouse-148822925) render. I tried to make the light bulbs in that picture as large as possible to minimize the effects of 'hard' shadows, where parts of the material inside the little 'Mandelbulb pockets' are in permanent black, but it still wouldn't compare to GI. In any case, I've been somewhat surprised at how GI can improve aesthetics even for the standard sky-lighting I have used for all my pics.

GI imitates real life more closely than usual, because everything (including the object) essentially becomes a light source as light is bounced around through all the nooks and crannies. The material seems to have a 'warm glow' feel, and there are less 'inky blotches' all over the place.

Below is a preliminary comparison I made between the two approaches. To obtain the same brightness and feel of the global illumination version, I have attempted to adjust the contrast, gamma and saturation of the direct illumination picture. But as you can see, it's all to no avail, as global illumination just does stuff that DI can't hope to achieve.

(http://www.skytopia.com/project/fractal/new/other/gi.jpg)

Here are some other GI renders I made last night. I'm surprised at how badly JPEG compresses the deep red picture (even at 95% quality!), so I used PNG for that one.
(http://www.skytopia.com/project/fractal/new/other/gi-red.png)
(http://www.skytopia.com/project/fractal/new/other/gi-chinaware.jpg)

Lycium gave me a tip for the red jpg problem (at least on a PC) - download and use IrfanView to save jpgs - it allows you to enable/disable the colour sub-sampling on jpgs, an option not normally available.

woaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa  :o


edit : btw, i have a fortran GI raytracer in my TODO list. Someday, i may be able to general fractal and GI in the same program :)

I still like your renderings over others Twinbee, maybe because they were the first I saw.
I like the old “clay” or “plaster” feeling in your Direct illumination image, the global ill gives more of a marble feeling...

...but it is great, getting even better!
Yes, I just need to get used to it...
They are even better than the old ones!  O0

the images look great but due to the GI, they lost a lot of sharpness...
Do you have some filter settings? Then you could set them sharper... :)

Aha, thanks!


Heh, you're like my brother who also said there was a 'small something' he preferred in the old one too. The glossy material took me a bit by surprise actually. There's almost a sub-surface-scattering look to it, and I definitely didn't program that! :)


That would be cheating though ;) You're right, there's some blurring which is a side effect caused due to the way light is transferred between pixels. I'd like to solve it somehow internally if poss, rather than use a filter afterwards.

That's not the only problem either, you might see some dark outlines, especially near the top left of the first GI rendering. Also, unlike most other renderers, light still doesn't travel under 'bridges' etc., which is great for speed, but of course not how things should be. Until I implement that, the GI won't be perfect. Until then, I need to end up estimating the brightness of certain unseen surfaces. It's a bit of a kludge (albeit a nice one :) )

Here's a picture where unseen surfaces are estimated at being brighter for the GI bouncing stages. Gives a nice snow effect:

nice indeed :)

I'm not sure how your GI works and just naively assumed it to work like a "physically based raytracer"...
Then I remembered my time with Indigo (didn't use it in a while by now^^), where you could set different filter options:
box
gauss
mitchell netravalli (not sure if it's spelled that way, always have troubles with that name xD)
as filter type and
radius
ring
as filter settings. Default would be mn ring=1/3, radius=1/3 but you could get sharper results without really cheating on the physicalness, so to speak, if you set, I think it was ring to .5 and radius to .25 or something. Sharper than that would have introduced artifacts.
Maybe I'm totally mixing things up here, though :)

Well anyway: Very nice shading :D

Cool ! On the last picture it really starts to be difficult to distinguish it from a real ceramics molded object !

Nice! Will your renderer be available for us to try out?

I was wondering, is there some point at which it would be easier to export fractal shots as high res meshes and import them into a rendering package?
Then you could use all sorts of techniques, like depth of field, reflections, refraction, high dynamic range, multiple light sources etc.

Or would it maybe be too hard to decide which parts of the fractal to export?

nice one twinbee :) since i work on indigo now, i should probably try adding support for intersecting the mandelbulb sometime! (though i don't have a lot of free time for fractals these days, as you guys have probably noticed)

Personally, I do not care for what I have seen so far with the "global illumination".  The "direct illumination" is much better.  The global ones appear as if the lens on a camera is way out of focus (or I have forgotten to put on my glasses).

It even messes with the perception of the image, for example:
   (http://www.nahee.com/Stuff/gi_partial.png)
these two portions from your original comparison images shows what I mean.  In one it is clear that there are layers one on top of another.  But with the "global illumination", it looks more like a double or triple exposure, where somebody forgot to advance the film before snapping another image, overlapping the previous one.  Like ghost images.
 

simulating reality isn't for everyone, there will always be a place for abstract CG :)

I prefer my simulated reality to be as if I was wearing glasses and nothing appeared blurry, or that somebody knew how to properly use a camera.

Why do some fanatics prefer blurry out of focus digital images, instead of something with visible details??  Is it because they have poor eye-sight and refuse to get corrective lenses or surgery, and then feel that is the way images are supposed to look like??
 

I always wondered how you actually create your images, they are fantastic! For an unknown reason I cannot recreate such images, mine keep having much less "depth-effect". Your images have "holes" everywhere, mine have just flat surfaces. If I increase the number of iterations in order to get "holes", a few appear, but then I need to apply a heavy anti aliasing in order to remove coarseness.

And regarding global illumination: Although at least currently I like more the additional details visible with the local illumination it looks excellent, too, and until now you seem to be the first person who has a running algorithm applied for the rendering of a mandel bulb. And you can do both :-)

If one now compares images with different lighting models one can see that it's not only the formula which creates nice looking images: Especially for 3D fractals proper light and an excellent lighting algorithm is essential for nice renderings.

Kind regards,
Martin

Over in "3D Land" the programs use Global Illumination too (I'm a big fan / user of Carrara, Bryce, and other 3D Software) and the same issues of Obnoxious render times arise with Global Illumination.

There are ways around it called "Fake GI" and the techniques are often unique to each program, but one Universal way is to use HDRI Lighting (High Dynamic Range Image).  Bryce calls it IBL (Image Based Lighting) but whatever it's called, it produces almost indistinguishable effects from GI and is something you may wish to look into.

I'll find some examples and post them later today...

But they don't eat into the render time nearly as much as GI does, and they have other advantages too  (they can show as a background if desired!).  Keeping details and shadows is also possible with HDRI Lighting.

I wonder how manageable an obj file (3D Model Format) of a Mandelbulb would be??  Fractal obj's can be quite large files, but Fun to play with!

Is that the one in my first post, or the 'snow' one?


I'd still like to release software with a full GUI at some point. But before then, everything needs unifying more, and I'd like to port it to the GPU for speed...


Yeah, the whole thing is so tricky, not least because of the resolution. Programs which convert voxels to polygons seem few and far between, and can only seem to handle a limited number of polygons. In principle though, it shouldn't be too hard program really (marching cubes algorithm), and understanding a polygon format to convert to. I'm still a little surprised there's no real standard 'voxel format' in place.


The blurring present wasn't by choice. As I said in my previous post, it is a bug in the algorithm. I think I know what's causing the problem, and in fact I even managed to 'fix' it, but something else broke, so I need to investigate further. It's also what's causing the dark outlines that are appearing.

Apart from that, one can always increase the resolution, and then resize to make smaller, but obviously I would like to solve the problem.

Anyway, hopefully I can change your opinion about GI once I fix the bug ;)


I think given the choice, anyone would prefer decent anti-aliasing/downsampling over a blurred image. However, pictures with a shallow depth of field, like Lycium's amazing 'Rise' certainly seem to benefit from the added 'blur' without losing anything:
http://lyc.deviantart.com/art/rise-64751071


Going to the cinema the other night, the entrance to the seating area was a a long dark corridor (felt material?) with lovely mini lights dotted all along. It was quite surreal, and quite CGI-ish. I think given the right materials (some may be impossible to create) and lighting, real life can be incredibly CGI-like :)

I'm beginning to think anyway that the 'DI look' can be achieved with GI simply by using smaller lights rather than an entire sky light. Perhaps something closer to the red picture for instance.


Lack of time... sigh, yes, I know how that feels, though I'd still like to see more renders from you these days! :P I bet you'd do some great stuff with the Mandelbox too.


One idea to reduce the 'flatness' which I have sometimes used in the past is to have the z ray travel twice as slow (or less) as it should, relative to the x and y light rays. It's an artificial fix, but it may help...

Sockratease, that HDRI technique sounds interesting - I wonder what lighting effects can't be simulated with that...

Here's another render from last night. I think it has a 'pearl'-like sub-surface scatter look, but heh, I didn't really intend that. Just changing one or two numbers in the algorithm can produce very different results.

(http://www.skytopia.com/project/fractal/new/other/gi-crystal.jpg)

Great latest image :D
Already got back a lot of its sharpness :)

The problem now is the inverting colours in too bright spots, right? (I see some cyan specular highlights which surely are supposed to be white or orangeish...)

@Lycium:
Natively supporting fractals (or algoritmic geometries) would be an excellent test for Indigos algorithms, right?
You said yourself already, fully raytracing a fractal isn't exactly an easy thing to do, as a lot of rays are gone for good in the extremely fine (infinite) details...

Did you see the thread on "colouring fractals in their true colour"?
(Something like that at least)
Do you think, you could do that? - Using the details of a fractal and interpretting them as thin soap-bubble-like walls which cause reflection and refraction in different colours due to different path-lenghts?

Sockratease, I don't understand how HDRI is in any way connected to GI. As I know it high dynamic range just means you store pixels with much bigger range than 0 to 255. This allows you to record really bright things in the image which helps with adding glare or changing the exposure of the image, e.g. when you look at the sun you can darken the rest of the image and the sun remains bright.
Twinbee, yes you would think someone would have made a standard oct-tree voxel format that rendering packages could use, or perhaps even some sort of 3d jpeg. But it wouldn't be easy to select the voxels if you're in a deep zoom, you might be looking off into the far distance, or the voxels on screen may be close but a shadow is coming from a far distance. And the data size could be very large.
myrtlesoftware.com is the only voxel renderer I found after a quick google, but it deals with transparent voxels mainly, and talks about 4k data sets, ha!

@tglad hdr is for getting more detail out of large ranges, especially high and low e.g. the sun and a shadowed place
i ised the formula n/(n+1) as a map for my values, getting a more  smooth behaviour in areas with great values :fiery:

Such images (hdr) can be used by software as a light source. They superimpose the image on a virtual sphere surrounding the scene, and it sort of "projects" everywhere.

They give results at least as good as GI and do not suffer the render time impact that is associated with Global Illumination.

It's not related to GI - it's a substitute for it.

I can find articles on it that can explain it better than I can if you want to know how it works.  It's a mainstay in most 3D Software I know.

From what I heard about it at a 3D forum;

The general  idea is mainly intended to be used along side normal lights. They're there to provide over all hue and subtle lighting, filling in spots regular lighting often can't... And it does it in a way "Ambient Lighting" can't live up to.

HDRIs are actually meant to be used for capturing a scene's lighting for putting CG inside of live action pictures or video. A lot of times you'll see in "making of" videos for movies which use a lot of CG+live action that they have a diffuse ball painted a certain value of gray in the scene. This is for calibrating an HDRI they made of that scene to the same value of gray used on a sphere in their 3D program. Then whatever 3D content they render out will have the same (or close to) lighting of the live action place they filmed.

People eventually branched off from the diffuse balls and started using HDRIs just as a simple, easy way to get high quality lighting. You don't have to fumble around with numerous light sources. You just load an HDRI image, maybe add one spot light for shadows and render.

Results comparable to GI with very little hit on the render time.

Sounds like it may help here.  After all - be it a Mandelbulb, Maya, or Bryce - It's all 3D Rendering using mostly the same basic principles for making the final picture.

HDRI isn't really a substitute for GI, it's another source of illumination (for environment lighting) as you mentioned, unrelated to the rendering algorithm in use. For example Twinbee's earlier renderings used a non-constant sky colour, probably with high dynamic range - he's still using that now it seems, but he's also considering secondary light bounces. My images on his page (also in the first article) rendered with GI as well.

Well...  I use it as a substitute for GI all the time in 3D work.  So I have no issues calling it such.

Plus, it has the added advantage of supplying reflective surfaces something to reflect - a nice trick if put to abstract use   O0

Hi!
regarding meshes..
I did a few tests using mandelbulb and mandelbox's slices (1 pixel) producing 3d objects using imageJ plus mandelbulb3D (thanx to Jesse).
The result is not so impressive according to me but it's ok for a test.
The very first time I used images produced by Daniel White (thank you!).
I think it's better to produce a 3d mesh of the bulb zooming a little bit inside a corner.. I will try it.
The meshes are usually 10-12 MB; so if I put together 4 meshes to obtain a better bulb would be 4 times bigger!
Morphing from power to power would be a problem; how many objects are necessary to morph a bulb from pow2 to pow3?? (this thing I could try as well).
the first one (Daniel's) has Sub Surface Scattering on, the other two are created using Jesse's mandelbulb3D.
(http://th04.deviantart.net/fs70/300W/i/2010/052/e/7/cinema4d_mandelbulb_by_bermarte.jpg)(http://th07.deviantart.net/fs70/300W/i/2010/071/8/4/pow2_mesh_by_bermarte.jpg)(http://th07.deviantart.net/fs71/300W/i/2010/073/1/a/mandelbox_mesh_obj_by_bermarte.jpg)

Here's the GI version that I should have posted first time. It regains the crispness, and the old clay/plaster look that was present in the old renders (the glossy look was actually a bug, but it's easy to add even now if need be).

(http://www.skytopia.com/project/fractal/new/other/gi-no-blur.jpg)

Hopefully, this one is more to Paul's tastes! ;-) I still have a feeling though that it's not quite how real global illumination with a perfectly diffuse, plain white surface, and a full-cover skylight, would appear. The reason being is I still haven't implemented 'underside' surface reflection, so overall brightness/contrast is almost a guessing game. I have a feeling that real GI would instead look as if covered in white paint, perhaps a bit more like this next picture (but less snowy, and without the 'black bits'):

(http://www.skytopia.com/project/fractal/new/other/snowy-bulb.jpg)

It definitely is better than the first GI version you posted.    :D   But I would not go so far as to state "(no blurring)".  Maybe very little blurring is a bit more accurate.  (My tastes are bit more discriminating.)    :dink:

If you look at the same area I pointed out before (below):
   (http://www.nahee.com/Stuff/gi_partial.png)
You can see in your latest GI version posted, that there are still areas which are unclear in detail.  The DI version is quite sharp in showing what is actually there, even down to the smallest pixel level.  To me that is what I call "normal", because I have very good near-vision, and expect images to contain the same level of detail.

I guess it does come down to personal tastes (which is how I will rate and vote on images and videos).
 

Thanks for giving your honest opinion.

But hang on, I think FF resized the image in your browser. You need to copy to the clipboard (or save the picture directly), and then paste/load into a picture viewer. If you just 'grab' the screen, you're getting the more blurred version (at least that's the only thing I can think of).

You should have gotten this:

You are most welcome.  And I always give such.  The problem is that most people can not handle that much honesty.  Most people state that if one cannot say something nice and positive, then they should not say anything at all.  I would rather give (and receive) the blunt facts and truth.


As I already stated:  "It definitely is better than the first GI version you posted."  But:  "....there are still areas which are unclear in detail."

And this has me wondering.  I have already stated I have Near-Vision, I can see things extremely well when they are close.  But I need corrective lenses to see things far away, otherwise everything looks like your first GI image.  Are you Near or Far Sighted ??
 

I think what I was trying to say was that with big pictures, Fractalforums.com resizes the picture to make it smaller (and therefore, makes it seem a bit more blurred than it really is). I'll post my latest comparison again, but this time zoomed in. Although parts are less contrasted in the GI version (as a natural consequence of global illumination), there's not even any anti-aliasing, let alone blurring on the GI version. See the parts circled by green circles - the pixels are fully there and precise. The jpeg artifacts cause the GI version to be more blurred in the red circle than it should be - I'll post the original PNGs if I can find the DI PNG version. So what do you think of this below:

(http://www.skytopia.com/project/fractal/new/other/gi.png)

I'm near-sighted anyway for the record, but apart from jpeg artifacts like the section in the red circle, it seems clear even upon zooming in, that there's not any more blurring than with direct illumination. Only less contrast (which I guess may look as though it's more blurred in a way), so I see what you mean up to a point.

** EDITED TO ADD CIRCLES **

Then I guess, in this case, it must be both the contrast and the JPEG compression that causes what I call areas of blurring.  It would be interesting to see the original .PNG outputs for the DI and GI versions.

I still feel that some of what the GI is doing tends to soften "edges" too much and make them appear as if they are rounded instead of maybe sharp.  See the area just to the left of the red circle??  It looks like a depression in the surface.  In the DI version, it almost appears to have a "lip" on the far left side of that sunken area.  But the GI version is more like a soft rounding over into the depression, no "lip".

Oh well, just a matter of taste as to what looks best to different people.    :dink:
 

I'll keep looking for the other DI PNG.

I agree with you about how the 'lip' part looks sort of better in the DI version.

There are a few things to say however. One is that this I might need to render at higher resolution to obtain the same effect in GI. And two, that that 'lip' effect shouldn't be there, and that's an artifact that DI produces, but which shouldn't. In other words, GI can create that too, but then the apparent lip would have to be actually there in the first place.

In any case, a proper implementation of GI with underside surfaces/reflections might also fix it. My current implementation is only a kludge so far really. And the final thing and most probable thing is, that lip might also be far more noticable if I used a single smaller directional light source, (instead of a full skylight which tends to produce less contrast, and reduce shadows generally). In other words, you would tend to prefer smaller light sources rather than DI over GI per se.

I might as well post the PNG of the GI version. It may not be a fair comparison admittedly though, until I find the DI PNG too:

@ Twinbee

I would love to see a combination of a single "heavy spotlight" combined with some of that global illumination.

I think that would result in a less "blurry" (wich I don't think it is) image.

Kind of what you should get from a nice photo-shoot IRL.
(I used to work as a photo assistant many years ago...)  O0

J

Hi,  maybe is better to have a mandelbulb inside a complete scene (for example a floor, various props, shiny objects, tiles, etc..) to enjoy the beauty of GI.

Just a test using .obj file.
(http://fc03.deviantart.net/fs70/f/2010/112/8/8/GI_test_by_bermarte.jpg)

Yes, I agree - a combination of local and global would be ace. In a way, the sun sort of acts a bit like a local light source since it's so small compared to the rest of the sky. Like Visual says, a scene will add more realism further (reflection mapping from a 'sky image' would be effective). Got to fix my GI alg though first...

Speaking of HDRI, I just wanted to show you a mandelbulb rendering result I get with my GI renderer using the technique of high dynamic range image based lighting.

http://www.fractalforums.com/index.php?action=gallery;sa=view;id=2383 (http://www.fractalforums.com/index.php?action=gallery;sa=view;id=2383)

I see some beautiful rendering here.  I do not wish to be too much off topic, but here goes.  I use Terragen 2 to render and Carrara.  I recently read about Mandelbulbs and I was hoping to find a program that could output obj files and import them into my favorite rendering programs. 

However, I think that I've been thinking about it all backwards.  A super detailed fractal would be impossibly large.  And of course the closer you zoom, the more coarse such an object would be.  To place objects and fractals into the same world, the proper solution would be to have the rendering program capable of generating the fractal.

I think you or someone like yourself should contact the Terragen programming team or the Carrara one and offer to add that functionality.  The Terragen rendering is awesome by the way.  I'm sure a terrain and atmosphere rendering program would love to have integrated fractal generation abilities.

And then end users like me would be able to render fractals and integrate objects and textures and control lighting, etc, etc, etc.

Hi.

I'm a big Carrara Fan  (you may know me from the Daz3D Forums!).

It seems that there won't likely be such a collaboration because these fractals are ray-traced things using Distance Estimates and not true 3D Objects at all.  There is a way to generate stl files from a group of slices of a fractal, but I never had any luck trying it.  In theory, the stl files could then be converted to obj or 3ds or whatever.

I find the best way to combine these fractals with my 3D Scenes is in postwork. 

(http://i106.photobucket.com/albums/m278/sockratease/th_z.jpg)
http://s106.photobucket.com/albums/m278/sockratease/?action=view&current=z.mp4 (http://s106.photobucket.com/albums/m278/sockratease/?action=view&current=z.mp4)

Sorry, never upped that one to yootoobe or vimeo, so there it is on photobucket.  You just have to click the image or link to see the video!

AFAIK  there are two main programs that output 3D+ Mandelbulb fractals as obj/stl files - these are Xenodream (latest version, Julias only) http://www.xenodream.com/ (http://www.xenodream.com/) and the latest version of QuaSZ by Terry Gintz http://www.mysticfractal.com/QuaSZ64.html (http://www.mysticfractal.com/QuaSZ64.html) .
I don't think either do Mandelbox or mixed types yet but I could be wrong.

As has been stated to get decent detail does create *very* large stl/obj files but if you have a 64-bit system this shouldn't really be a problem - folks have been exporting quaternions and similar to Bryce/3D Studio etc. using QuaSZ for many years even on 32-bit systems.

Hi Sockratease and David,

Sockratease, I always read your posts at Daz3D with interest!

I agree about postwork integration - or taking a fractal image and placing it on a plane within the render program, but the collaboration I hope for is not for object production (since good objects are so large), it is for integration and generation of mandelbulb fractals into existing rendering programs (like Carrara or Terragen).  They ray-trace, they texture, etc. 

Is it possible to save the mandelbulb as a obj? That would me so cool, always wanted to play around with 3d fractals in maya!

Yeah, I post a lot of 3D Cows at the Daz forums, and fractallized photos Lissa here!


But texturing is handled via UV Mapping a 3D Object, or for procedural textures - following the geometry.  Neither is possible with an object lacking mesh  (that's why volumetric clouds and fire and such (in software like Carrara and Bryce) can only have color gradients and not image textures)  (just like Mandelbulbs!).

The closest we came was in an early version of Mandelbulb 3D which had a feature called "Bitmap Lighting" - which used an image projected as a light source.  Just as I got the feature under control, and learned how to focus the image, it went away   :headbatting:


(http://www.fractalforums.com/gallery/2/162_04_04_10_9_02_14.jpeg)

http://www.fractalforums.com/index.php?action=gallery;sa=view;id=2010

I'd love for that feature to come back, but Jesse said it was impractical for some reason I forget...
It is possible, but only in limited ways thus far as the obj files tend to be half a gigabyte for decent detail.  See the software mentioned above by David.

about the texturing...  I see what you mean.  The color then must come from a lighting strategy (as you mentioned) or be procedurally integrated with the formula that generates the fractal itself.  I don't know if it could know slope for instance, but it might know distance or location or what iteration of a formula it came from.  Any way though, I hear what you are saying...