0.6 test 2... half baked or fully baked? :)

psor · Post by **psor** » Sun Sep 03, 2006 2:46 am

Thanx Nick, awesome!

FYI ... there is a folder missing ==> tree_cache

So everybody should create this folder to take advantage of
the new tree_cache option!

edit: Example: ...\indigo_06_test2\tree_cache

take care
psor

Post by **OnoSendai** » Sun Sep 03, 2006 2:52 am

Ah yes, thanks Psor.
And on a related note, watch the size of that folder (tree_cache), it can get pretty big!
You can always delete stuff in it if you need to.

psor · Post by **psor** » Sun Sep 03, 2006 3:01 am

You are very welcome! btw. it would be superb if we could specify
the path for the tree_cache in a later release so that it will be not
to confusing. For example if I have 2 scenes I like and would like
to play with it longer, on the other hand I have 10 testscenes ...
I dunno which tree file I can delete or not. If I could use the CLI
to give indigo another path for the cache for those files I wanna
keep this would be helpful. ;o))

But this is just something for a later release, don't hurry, just keep
it in mind. Thanx!

take care
psor

Post by **OnoSendai** » Sun Sep 03, 2006 3:08 am

psor:
yeah i think the tree cache will need a lot of tweaks and stuff before it is complete.
btw you can turn it off in the inifile.

zuegs · Post by **zuegs** » Sun Sep 03, 2006 3:20 am

Hi Nick,
my cornell-box with v0.6t2 gets 50kmutations/s only (with v0.6t1 it was 100kmutations/s in bi-dir-MLT). But the v0.6t2 picture looks smoother (less noise) after the same time than v0.6t1. Is it correct, that the new bi-dir code gets less mutations but better ones???

Post by **OnoSendai** » Sun Sep 03, 2006 3:26 am

Correct, mutations take longer now. But each mutation now involves many connecting paths, so the overall quality per time is higher, or perhaps i should say the variance after a certain time should be less.

zuegs · Post by **zuegs** » Sun Sep 03, 2006 4:08 am

Hi Nick,
yes, also i can see more color-bleeding on diffuse-materials (more interaction between surfaces) and less directional dependance from light-source (the cone of light on diffuse-walls is weaker). I feel I have to decrease the luminance of diffuse materials to match older renders?!?

Post by **OnoSendai** » Sun Sep 03, 2006 4:10 am

Zeugs: can you post a comparison or new render?
Sounds like a bug

Camox · Post by **Camox** » Sun Sep 03, 2006 4:23 am

wow, thx Ono-Sendai thats cool !

I must try out immediately .

arneoog · Post by **arneoog** » Sun Sep 03, 2006 4:26 am

Cool

I did this test to compare speed

(30 min)

It's less noise in the 0.6 test2.
I notised some normal smoothing problems have been fixed but now a new one have emerged...
The object inside looks like it has normal smoothing OFF, but it's exactly the same scene with normal smoothing ON!

Well well, it's not a complete version

zuegs · Post by **zuegs** » Sun Sep 03, 2006 4:27 am

Hi Nick,
I can't figure out if it's a bug, over all it looks very good too, i like the color-bleeding. The directional-independence of diffuse-materials produces less contrast, thats odd.
Here a picture, left v0.6t1, right v0.6t2 (the sphere is a mesh)

regards, zuegs

Post by **OnoSendai** » Sun Sep 03, 2006 4:46 am

Zeugs:
there's something really strange going on in that right picture, i'll have to do some tests on this.

Arneeog: normal smoothing is disabled right now.
Nice render btw, but we're going to get some mocks from the maxwell forum if we use that as the material test scene

arneoog · Post by **arneoog** » Sun Sep 03, 2006 4:51 am

ah, okay!
I did model it myself...

Hehe, I think it's okay as long as it's not used for Material Previews

Zom-B · Post by **Zom-B** » Sun Sep 03, 2006 8:31 am

Hey Nick,

Nice you put the cache_trees in!
Testing different Materials & setting on scenes now gets even faster

I got 2 Questions about the v0.62 inifile.txt

1) You reduzed the "max_num_consec_rejections" down to 100 from 1000 in 0.61.
What this is about? I even don't have a clue about this option ^^

2.) The "russian_roulette_live_prob" is tagged out using -- , why?

Thank you for the great work, I'll start some testscenes soon

psor · Post by **psor** » Sun Sep 03, 2006 9:18 am

I would like to try to answer your questions.

Zom-B wrote:1) You reduzed the "max_num_consec_rejections" down to 100 from 1000 in 0.61.
What this is about? I even don't have a clue about this option ^^

This option will introduce more biased calculations which will help to
get rid of white dots caused by caustics. Thats the pro ... the contra
is it is not unbiased anymore ... but you have the choice ... nice! ^^

Old example:

Indigo v0.5 Test3 (hardcoded max_num_consec_rejections = 1000)

-------------------------------------------- Click image to enlarge ---------------------------------------------------
resized to 1360x1020 (original image)
rendertime: 100h
box: AMD64 3000+ @2,3GHz; 2GB ram

Indigo v0.5 Test5 (max_num_consec_rejections = 100)

-------------------------------------------- Click image to enlarge ---------------------------------------------------
resized to 1360x1020 (original image)
rendertime: 100h
box: AMD64 3000+ @2,3GHz; 2GB ram

Zom-B wrote: 2.) The "russian_roulette_live_prob" is tagged out using -- , why?

This option is not needed for bidirectional MLT ...

take care
psor

0.6 test 2... half baked or fully baked? :)

Should i release 0.6 test 2 now?

inifile.txt

Re: inifile.txt

Who is online