CoolColJ's test pics thread

[CoolColJ]

Nice!

That's pretty smooth and blotch colour free for such a low SPP. Currently it takes well over 2000 samples to clear out out all the artifacts, and some bits never go away completely.
It always puzzled me why Indigo would splat some random colour in some area where there is no caustics.
It can get pretty bad with plain MLT, without bidrection

So what were the changes that you added?

Hope you guys roll out the new build ASAP

Guideline for max change: if you have large areas of smooth illumination, maybe use 0.01, for lots of bright small lights use maybe 0.005, and consider not using MLT at all (except if you have SSS or something).

Yeah, I've done a lot of experiments with these MLT values, but it's nice to get some concrete values from people who know the innards

Indirectly seen caustics without MLT? Not a chance, believe me I've tried BiPT, but it always tends to look far grainier on caustics and has no hope of rendering refracted caustics and other things.

[CoolColJ]

If I remember correctly you are on C4D J ?!
In Cindigo per default Sun renders in 2 layers, sun + sky! Disable this in Advanced Indigo Settings for a speedup!

Zomb-B

I tried this, and it makes no difference. Indigo is still crazy slow with sunlight refracted caustics

It renders a small patch and then nothing for a while, and then another small bit and so on. Even with a simple flat water plane/glass and no bump mapping etc

Do this in Luxrender, and it finds the caustics within 30secs, and starts to lay down caustics continously until the whole area is filled and starts to refine it, so that by 100-200spp you can see it properly

[CoolColJ]

Well displacement does seem to work fine here, it just doesn't like my pool model for some reason...
As long as smoothing is turned off!

Pretty chuffed with the results - sweet!

I did some of these in plain path tracing and couldn't believe how fast it was going!

path tracing

path tracing

Bidirectional path tracing

[lycium]

You should see how fast plain path tracing will be in the next build...

[CoolColJ]

You should see how fast plain path tracing will be in the next build...

Never thought I'd see over 100k samples per second on my PC, but I did! It used to be that 40-50k on a simple scene was fast for me...

anyway I found with the above scene, because the floor is so large, it really slows down the appearance of caustics!
I sometimes wish unbiased render engines weren't so random. Logically there would be no point in shooting much rays at the edge of the scene floor, since it's a distant horizon

[Borgleader]

dude that render is sooooo cool how dyu un wrap the cube for the texture to be seamless like that?

[CoolColJ]

dude that render is sooooo cool how dyu un wrap the cube for the texture to be seamless like that?

I didn't do anything special, just a basic UV map with a seamless tiled texture

here is what the texture does, displacing a flat plane. With an offset, and a negative quadratic(b) of the same value as the displacement

Now the cube has rounded edges which probably gives the effect of wraping around - pretty cool it is too

[CoolColJ]

Ok this is puzzling, after all the slow renders I've had with water caustics, I do a new test and the caustics showed quite a lot quicker, and there was bumpmap on the water surface!

The only thing I can conclude is that the pool model makes a big difference... maybe.
In the other pool scene I tried, that was slower, the pool was proper 3D enclosed model with beveled edges etc.

In this scene, the pool is just a simple plane, extruded down. There is no bottom, just normals facing up. Hard edges, and no textures.

Maybe having no bottom in the model speeds up Indigo's MLT since it doesn't have to shoot rays at the bottom of it. Or having no bevelled edges and textures does likewise....

I don't know why, but the caustics showed up a lot faster compared to before, and the surface is quite a high freuency bump map!

Didn't bother with basic Path tracing or MLT, since the rays have a hard time lighting up the underwater object

Bidirectional Path tracing at 500+ samples. Doesn't look like the caustics will show anytime soon, but some parts of the underwater objects did light up. Maybe after a few thousand samples, all the those little dots become caustics

Then Bidirectional MLT. At 2000+ samples, still needs more rendering time to smooth out, or I may need to use smaller Max Mutation Range settings. I used default Indigo settings here of 0.01 since I found a smaller settingS like 0.005 to 0.003 caused a lot of multi-coloured artifacts.
The underwater cube should have some caustics right?

I will have to investigate some more, I'm trying the same scene now but instead of using bump map on the water surface, I have the same texture doing displacement. So far it is rendering more or less the same speed, just looks better.
And then texture and bump map the pool surface, which has slowed things way down in the past.

reference etc
I see a bit of dispersion, although rendering it takes forever!

[ENSLAVER]

Those caustics look sweet CoolColJ, I want to jump in lol - What bump map did you use? Is it shareable?

[CTZn]

The underwater cube should have some caustics right?

Wrong, you're shooting it's backlit face, it's in shadows. In fact I'm pretty sure you can guess a grazing light on it, might as well be subjective though

[CoolColJ]

Those caustics look sweet CoolColJ, I want to jump in lol - What bump map did you use? Is it shareable?

I took the bump map off the Thea render demo


I kept the render going and even at 3100 samples, it still hasn't really smoothed out much, or the caustics taken on a sharper appearance like these pics - will try different settings, smaller Mutation Range of 0.25
Or maybe the bump map isn't sharp enough

Yeah the dispersion is pretty obvious, but I'll wait till I get a faster computer before I try rendering them

[CoolColJ]

after experimenting on many different types of texture bump maps will poor results, I finally had the bright idea of trying Indigo's built in Perlin noise shader for bump mapping! The noise itself even looks like a smooth water type shape

Scaling it up 400%, and using an output multiplier of 1% seems to be doing a good job on so far, will see when the render progresses how it turns out...
It certainly is more organic in theory, with better precision/resolution

Then to try it out with displacement, but should smoothing be turned on? And will Indigo crap up the mesh when doing so?

[Soup]

Smoothing on for sure. Just make sure the mesh has a high poly count (or turn subD up) and you'll be okay. You'll only get problems with displacements if you can see the edges of the mesh, but as your pool is enclosed it'll be fine
ISL is probably the best bet though. Post results!

[CoolColJ]

some quick tests - using Indigo Shading Perlin noise to do the bump mapping on a simple low poly water mesh

Basicly what I did was blend 2 different Perlin noise. I had one at 400% scaling and 75% value, and the other at 900% and 25% value. And basicly added them together so total value = 100%
Then set the total output to 1%

It works fine if I keep the total multiplier output low/bump map amount at 1% or under.
But the underwater Cube does not seem to be lighting up.... and it definitely should have caustics on it...

At 700 samples

150 samples

if the total output is at 2%, you start to get these dark areas forming. You can see some caustics on the back wall and water in the sphere reflection

250 samples

Black artifacts are not an issue with a higher angle though. Caustics look to high in frequency. Will try reducing the noise octaves, but keeping the ratios the same.

200+ samples

I'll try displacement next, which should get around the black artifacts stuff

[CoolColJ]

This is what it looks with the noise scaling at 400 and 675% for the 2 layers of Perlin noise respectively.
Displacing the water mesh, but with combined output of the noise set at 5%. The higher level creates more defined caustics, but seems to render slower

Looks ok above water

But a bit too simple underwater, maybe another layer of noise at a lower frequency will help. And maybe instead of adding the noise layers together, using mulitplication

real