CoolColJ's test pics thread

[CoolColJ]

hmmm I think we have a winner.....I may have found the best overall settings for rendering the underwater stuff, for this scene at least

Bidirectional Hybrid
Large Mutation probability 0.5
Max change 0.01
MNCR 10,000

I haven't seen such orderly and uniform convergence of the water and caustics and this quickly. All the other modes and settings I've tried so far have been fairly patchy and rather random
Only at 260 samples, and caustics started to show within the first few samples! And they have the same smoothness with max change of 0.005 with Bidirectional MLT. Plus the sky is super smooth and clean already as per the usual with path tracing

Surprised me as I've never been a fan of Hybrid mode!

update later with more rendering time

[CoolColJ]

perhaps I was wrong, a few hours more and Hybrid mode isn't working as well as I thought...your sample rate drops dramaticly, so while your directly lit parts are uber clean, your water is still not converging all that well...

MNCR 10,000 is not suffcient it seems, back to MNCR 30,000 and more with Bidirectional MLT!

[CoolColJ]

bump map vs none rendered to 200 samples each - why the difference?!
All diffuse materials except water - in V0.9 test 8

samplerate is also slightly higher without the bump mapping on the pool walls and floor

also the yellow cyclinder has smoothing, but I can still see the segments

[zsouthboy]

Bump mapping "throws away" samples - hence the difference.

[CoolColJ]

it must be throwing away 90% of the samples! At this rate it will take till 20000 samples to render the scene to full convergence!
Like my last test with this scene in V0.9 test 6

I guess Bump maps fall inline with exit portals then, don't use em unless you really need em

[CoolColJ]

The non bump mapped version is huming along nicely up to 2000+ samples now. MNCR of 30,000, Max change of 0.005, Large mutation probability 0.5

just imagine how long it would take the bump mapped version to reach this level?

[WytRaven]

Looking nice Cool

[CoolColJ]

pity I can't use Phong material with Test 8

[WytRaven]

What exactly about phong is screwed? I'm mixing phong and specular in my crstal test scene, and it's rendering just fine...friggen slow...but fine

[OnoSendai]

There are issues with phong inside a medium with IOR > 1

[WytRaven]

ahh ok

[BbB]

CoolColj. I haven't tried it, but have you tried to blur your bump map and see whether it speeds things up? Just an idea...

[CoolColJ]

CoolColj. I haven't tried it, but have you tried to blur your bump map and see whether it speeds things up? Just an idea...

yeah the bump map is already filtered with a median filter

There is a big difference in this scene since the water lights up so quickly for some reason, but in the other scenes it doesn't light up as fast, so perhaps there isn't much difference. I will try a few more comparisons and see

I think MNCR 1000 in v0.9 test 8 may show the same level of water detail and caustics as MNCR 30000, when you get to the 5,000+ or higher sample range. It just does it in a different way - MNCR 1000 renders little bits and pieces evenly spread over the whole scene gradually until the full image is formed. Whereas MNCR 30,000 or higher renders a big chunk detail, ignores some other parts and then proceeds to the next bit, and so on. So it shows the main scene details much earlier on.

The pic below is using the default settings with MNCR 1000, at around 500 samples now and you can see some of the caustic and sunlit water details, but its spread out all over the scene in small bits

[CoolColJ]

I'm also trying a different approach for water in a modified scene.
I have a 68,000 poly water forming the basic shape with gentle waves and swells. While I have a bump map on top providing the high frequency detail. It should give it a similar look to the water in that caustic photo I linked earlier http://www.lysator.liu.se/~kand/caustic ... ustics.jpg
if you look at the water reflections in the upper part, you can see a gentle shape plus higher frequency patterns overlayed ontop.

This should avoid any shading errors while providing the same sharp detail as before. Hopefully it doesn't increase the rendering time of the caustics by a lot!
Currently it's at 220+ samples and only shows some small bits and pieces with MNCR 30,0000

[CoolColJ]

at 4500 samples, I would say it's still not fully converged yet

again I'm puzzled by the segments on the yellow cyclinder - it does have smoothing on