CoolColJ's test pics thread

BbB · Post by **BbB** » Wed Sep 26, 2007 10:33 am

Looks nice though. Shame about thoses caustics

CoolColJ · Post by **CoolColJ** » Wed Sep 26, 2007 6:20 pm

this is with default settings and you can clearly see the sun/caustics go through the glass

what makes this situation any different from the water/pool?

BbB · Post by **BbB** » Wed Sep 26, 2007 7:27 pm

I would say the glass is much thinner (Your water pool is a closed volume, right? Not a single plane...?)

CoolColJ · Post by **CoolColJ** » Wed Sep 26, 2007 8:53 pm

yeah it's a single mesh, and it's wedged inside the sides/bottom of the pool
Otherwise it doesn't look like water, the refraction won't work properly

that glass above is the same as well, single mesh with thickness and volume.

BbB · Post by **BbB** » Wed Sep 26, 2007 8:57 pm

And it can't be a general problem with caustics showing inside a volume as you see them when the camera is in the liquid... Mmmmh. Puzzling

CoolColJ · Post by **CoolColJ** » Wed Sep 26, 2007 9:04 pm

here is Ian's solution with Radium recently after I brought this up - hope Ono can take something from this

Right, I think I'm onto something now. A few extra lines of code and winching the MCR (MNCR) up to 50,000 has resulted in this after only 30 mins:

http://www.trackvids.co.uk/radium/galle ... water3.jpg

It's still a bit patchy but I think I can refine it even further.

The way it works is fairly simple... if the path generator detects a sun caustic (direct or indirect) it instructs the next mutation to set MLT_MAX_CHANGE to zero for mutations at diffuse surfaces. This exploits the fact that the sun is omnidirectional and almost guarantees that many of the subsequent mutations find the sun again. It's unbiased too

Ian.

Unfortunately the above render died overnight (the laptop hibernated and killed all the running processes... oops)

But... I resurrected the original pool scene (provided months ago by clipi based on an Evermotion scene). I had to import into Blender from .3DS, copy and paste Colin's water into it and re-export as .OBJ.

Seems this optimisation works extremely well for the underwater caustics... they were nearly perfect after 30 mins rendering across 12 cores but I left it for a total of around 2 hours just to see if the directly-viewed caustics on the walls would resolve a little better (they didn't but I've not implemented the full optimisation for these yet):

http://www.trackvids.co.uk/radium/gallery/pool2.png

The MCNR was set to 500,000, max_change (default) at 0.003 with diffuse bounce max_change dynamically changed to 0.0 for certain paths (as described earlier).

Just a sidenote ... no-one should ever be afraid to set MCNR way high... it only affects the complex paths and won't introduce any fireflies on well-lit diffuse surfaces. I've previously rendered a diffuse Cornell Box scene with MCNR set to 2,000,000 and it's indistinguishable from a 100 MCNR render after the same amount of time

CoolColJ · Post by **CoolColJ** » Wed Sep 26, 2007 9:07 pm

maybe I can "cheat" by putting a thinner water mesh inside another with greater precendence or something

BbB · Post by **BbB** » Wed Sep 26, 2007 9:15 pm

Yes. Try it.

CoolColJ · Post by **CoolColJ** » Thu Sep 27, 2007 1:54 pm

is someone playing Jedi mindtricks on me?! I do believe I see some underwater caustics

all default settings too....

new version of Cindigo scales ray_origin_nudge_distance with the global scale multiplier of 0.01 to this <ray_origin_nudge_distance>0.0000010000</ray_origin_nudge_distance>
default is 0.0001

hmmmm

pic to come..

CoolColJ · Post by **CoolColJ** » Thu Sep 27, 2007 6:44 pm

hmmm maybe I was wrong, at least with the default settings

CoolColJ · Post by **CoolColJ** » Thu Sep 27, 2007 10:11 pm

progress so far at 800 samples, default settings but with MNCR jacked up to 50,000. Those Water surface highlights show up really quickly with MNCR this high. I have another render going with the default MNCR of 1000 and I don't see any surface specular highlights yet, but there looks to be some dots so they may show much later

Still, they are building up dot by dot... slow... but they appear to be in a wave type shape this time.
I removed the smoothing off the water, which got rid of a few shading/smoothing errors and the ripples in the water surface show up better. Water has 66,000 polys so probably doesn't need smoothing anyway, at least from above. Underwater is a different story

This is gonna look real nice with bloom and glare once it's all finished up and clean. Just hope those caustics show up, however slim a chance that may be....

water really needs specular highlights and caustics to look sunny. I have the sun almost directly above, ie Noon

Kram1032 · Post by **Kram1032** » Fri Sep 28, 2007 3:20 am

looks great

I've an idea for smoothing

what about using an edge split modifier on very low angle? - some flat stuff will stay smoothed, while most will look wavy

I wonder, how long it will need to finish, with that hight MNCR?

CoolColJ · Post by **CoolColJ** » Fri Sep 28, 2007 8:39 pm

well I didn't use Blender for this scene though

But that would definitely help, water is pretty tricky smoothing wise

A long render time even with lower MNCR. Most of the renders I have done with any amount of reflective surfaces and caustics needed to be rendered into the 7000-8000 samples range to be even considered near finished

I resumed the render with MNCR cranked up to 500,000 and this is what it looks like at just under 3000 samples. I think it's about 40% near finished.
I'm confident those white things are caustics to be

Once they smooth out and blend in a bit more. I just wished they would show up quicker like when the camera is underwater!

For one you only see them in the non shaded areas of the water and only in places where you can see the refracted floor. There is none of the right hand side where you can't see the floor due to the reflections.
Keeping my finsgers crossed

I also did a render with stock settings of MNCR of 1000, Max change 0.01, Large Mutation probability of 0.4. That is what it looks like at 2500 samples and it hasn't changed much in the last 1500 samples

I think MNCR of 1000 is definitely insufficient for complex light paths...
Definitely not unbiased. I'm going to set my general default MNCR to 1 million and leave it at that

This pretty much proves that fireflies with MNCR 1000 is just incompletely rendered caustics/reflections in the MLT pathways. Low MNCR renders don't get this because the paths are cut short!

Kram1032 · Post by **Kram1032** » Sat Sep 29, 2007 6:57 am

very nice

waiting for finished version

I wonder, if it was possible to make the MNCR set to "false", for test purpose, so to say, an infinite MNCR

(there would definitely be fireflies, then, though, I guess...)

CoolColJ · Post by **CoolColJ** » Sun Sep 30, 2007 12:40 pm

MNCR 500,000 is too high, but 50,000 works well
It puts more emphasis on the important light paths in your scene without neglecting the rest. 25,000 to 50,000 is what I like so far after more tetsing

I'm up to 6500 samples and still no end in sight!

this one is gonna take till 20000+ samples to finish! Me thinks Ono needs to do some paths tweaking ala Ian with Radium