feb 03, quick & dirty job on the conference room back

(L) [2006/02/03] [tbp] [feb 03, quick & dirty job on the conference room] Wayback!

This scene is referenced in a bunch of papers (MLRTA is among them), and i've found a so-so nff version on the web thanks to google.


Notes:


 hacked up nff partial parser (read bogus).

 materials are kinda wrong (darkish).

 there was no textures bundled with the scene.

 i've quickly thrown a lonely light into the room.

 my kd-tree compiler takes his sweet time to massage this one, i wonder what is annoying him.



[IMG #1 ]
[IMG #1]:Not scraped: https://web.archive.org/web/20061004012832im_/http://ompf.org/ray/wip/pix/20060203-01-conference.jpg
(L) [2006/02/03] [tbp] [feb 03, quick & dirty job on the conference room] Wayback!

Ah! Spotted that thing and couldn't resist... gave it a quick & dirty try while cooking...


It comes from: [LINK http://hdri.cgtechniques.com/~sponza/]

It's really nice, and is about to become a standard, so i suppose we should put it into our scene kit [SMILEY Smile]


[IMG #1 ]
[IMG #1]:[IMG:#0]
(L) [2006/02/03] [Phantom] [feb 03, quick & dirty job on the conference room] Wayback!

I know that scene. It's really nice, and I'll start using it right away as a default scene. Nice amount of triangles btw. Scene setup is a kd-tree traversal nightmare though, with all those pilars.
(L) [2006/02/03] [tbp] [feb 03, quick & dirty job on the conference room] Wayback!

Surprisingly it doesn't seem to be that horrible to traverse (tho i'll admit i haven't done mucho testing obviously).

And what's even more surprising is that it compiles real fast. I would have expected something much much worse.


The scene would need some 'fixing' for non-GI rendering, some materials are a bit off.

That or my export to wavefront went bogus.

Anyway, it's a good test scene and with proper lighting it should be quite pretty. A pretty test scene, what can you ask for, eh? [SMILEY Smile]


Hmm, those large screenshot really screw the layout. I'm gonna send a flame mail to the admin.
(L) [2006/02/03] [Phantom] [feb 03, quick & dirty job on the conference room] Wayback!

Not horrible to traverse? You're doing 1fps, while that truck you posted earlier did 1.5fps, at double the amount of triangles.
(L) [2006/02/03] [tbp] [feb 03, quick & dirty job on the conference room] Wayback!

True true, but as with every lonely-model-in-a-scene rendering you cheat on coverage.


So, for the sake of argument let's say there's a coverage of 100%; the max tree depth was hmm 25, and i get an avg of 33.3 node inspected per ray (i don't remember precisely if that's ray that hit something or all... anyway doesn't matter here).

Of course it's nothing to write home about, but iven the complexity of the topology, i mean all those arches/pillars etc... it's not _bad_ [SMILEY Wink]


In fact i find that a bit suspicious. I really wish i had a good visualisation tool to inspect that damn kd-tree.

I plan to spend more time looking into this, cuz i'm sure there's some clues for a better compiler in there.
(L) [2006/02/03] [Phantom] [feb 03, quick & dirty job on the conference room] Wayback!

We should colaborate on that kd-tree thing. I can easily isolate my kd-tree and I am willing to 'donate it to the community'. I would document it's inputs and outputs, and everyone could try to use it. If you could do the same, we could share a common kd-tree data format, so we can plug in your compiler in my tracer. Would be interesting.
(L) [2006/02/03] [tbp] [feb 03, quick & dirty job on the conference room] Wayback!

Yeah, that would really be a plus (btw it's not really the first time we try to address that point).

There shouldn't be much divergence in the final binary form, if that's what you're talking about.


In fact, while ruminating over yet another complete rewrital of my compiler, i've concluded again that the compiler has nothing to do within the renderer. That damn things keeps creeping back into tho.


So, my goal this time would be to keep a strict separation.

That would play really nice with your proposal.

And this time i want proper analysis/diagnostic tools.


The only problem is how efficiently you handle data in/out, but that's not that critical.

Well, that's not the only problem, the other being that tackling that MRLTA thing is much sexier than dealing with the kd-tree crap [SMILEY Smile]
(L) [2006/02/03] [tbp] [feb 03, quick & dirty job on the conference room] Wayback!

I'd need to check what's really going on but from the top of my head, interfacing that compiler as a black box should be really easy.


I mean it only needs to fed a triangle soup and it dumps a kd-tree in finalized form + some clipped triangles.

And it doesn't share much code with the rest (not really sure about that, would need to audit it).

Hmm.

back