Sphereflake level 9 (big image)

(L) [2007/04/13] [fpsunflower] [Sphereflake level 9 (big image)] Wayback!

I'll post images of my own sphereflakes here to avoid further hijacking of tbp's thread.

Like I said before, the code that renders this is totally procedural, no geometry is actually generated. I just apply the recursive sphere generation during ray intersection.

This image was originaly 8192^2, I rescaled to 4k for posting online. Two spherical area lights and a single bounce of path tracing were used. The render took 4 hours on a dual dual-core opteron.

[img width=999 height=999]http://img126.imageshack.us/img126/8478/outputxl4.jpg[/img]

tbp: tweaked due to thread enlargement

(L) [2007/04/13] [phresnel] [Sphereflake level 9 (big image)] Wayback!

Veeeery neat [SMILEY :D]

Make it transmissive, plant on bottom and add Preetham and show us some caustics (please) [SMILEY ;)]

Can you give us any measure of the Rays/Sec in your renderer?

(L) [2007/04/13] [fpsunflower] [Sphereflake level 9 (big image)] Wayback!

[SMILEY :)]

I've actually never implemented a ray/sec counter. Proper stats gathering from multiple render threads is one of those things thats been on my todo list for a long time.

I'd be scared to turn this into a refractive object. With so many tiny details, you would need a really high trace depth I think. My caustics code is also mostly foobared at this point in time (on the redo list [SMILEY ;)]). I'll give it a shot nonetheless. It's also a bit tricky to rest this primitive on a ground plane neatly because of the way the spheres are distributed (off axis and uneven). I'm currently trying to fix this - even the original SPD code generates a sub-optimal distribution IMHO.

Interestingly, while it is possible to go way past level 9 with this system with little impact on rendertime, it doesn't do much since you start running into the limits of floating point accuracy. You also would need a fairly ludicrous image resolution to capture the highest and lowest level of the sphereflake in the same image. Then again, the memory you save on geometry can be put towards more resolution [SMILEY :)]

(L) [2007/04/13] [tbp] [Sphereflake level 9 (big image)] Wayback!

Mighty impressive!
I've rendered lvl 9 but that took a 64bit system... and really, those details...

edit: forgot to say that if you come up with an interesting definition for a sphereflake lookalike, i'd like some more info [SMILEY :)]

(L) [2007/04/13] [beason] [Sphereflake level 9 (big image)] Wayback!

Awesome. After caustics, maybe run particle deposition (taking advantage of your instancing) to deposit millions of sphereflakes [SMILEY :P]

Edit: Making this comment after my suggestion to tbp I guess I really have a hankering for universe of sphereflakes. Complexity == cool

(L) [2007/04/16] [lycium] [Sphereflake level 9 (big image)] Wayback!

tbp: your tweak killed his aa :(

fp: hot damn! very smooth, very nice. i guess that stuff is represented implicitly? any clues on that? edit: man i suck, just read your description...

(L) [2007/04/16] [lycium] [Sphereflake level 9 (big image)] Wayback!

have done, punk, that's how i know you fudged it :P really the proper action is thumbnail + link, but since browsers in 2007 are still clueless about sampling issues, the hackfix is suckful.

(L) [2007/04/16] [fpsunflower] [Sphereflake level 9 (big image)] Wayback!

I've tweaked the distribution a little, and simplified the basis generation to avoid having to normalize too many vectors.

You can check out my code here:

[LINK http://sunflow.svn.sourceforge.net/viewvc/sunflow/trunk/src/org/sunflow/core/primitive/SphereFlake.java?view=log]

(L) [2007/04/16] [fpsunflower] [Sphereflake level 9 (big image)] Wayback!

Here's another version under skylight illumination.

[img width=960 height=540]http://img383.imageshack.us/img383/893/sphereflake02hd3.jpg[/img]

Click for full-res. Rendertime was roughly an hour on a dual-dual core.

(L) [2007/04/17] [fpsunflower] [Sphereflake level 9 (big image)] Wayback!

a) It took about an hour on a dual dual-opteron (read the post! [SMILEY ;)]). I think its only level 7 and there's still some noise left but its not particularly objectionable. You can pretend its a subtle surface texture [SMILEY ;)]

b) I did a glossy reflection test, but somehow it was less visually pleasing. Caustics are out of the question with my current code. Even just turning this into glass looks strange.

I'm working on a new cgsphere entry next ....

Sphereflake level 9 (big image) back

Board: Board index ‹ Ray tracing ‹ Visuals