The flickerfree animation challenge

[Fox]

So I decreased the lookup radius of the caustic cache and made it 10 times smaller and the result I get with a value of 0.0005 m is good.
It takes forever, almost 10times. And this is just a 1/3 of the resultion. So at the end 1 frame will take 10min*3*3=180min. Uhh.

Looks like the Oidn denoises better small radius even for the BiDirVM.
And the smaller normal map scale of 0.1 on the wall, denoises also better.

Code: Select all

            "bidirvm.lightpath.count": 131072,
            "bidirvm.startradius.scale": 0.00005,
            "bidirvm.alpha": 0.999,

64 spp
https://streamable.com/r1dujj

[lighting_freak]

Hi,

Knowing the average luminance of the image to render is the only Metropolis requirement

Why does the sampler need this? Isn't it a bit strange to have the result as an prerequisite? What about a MonteCarlo sampling strategy? Does it also need the average luminance?

BR

[mischterlampe]

Here is the caustic only rendering with 100% cpu for light tracing and only sun.

https://streamable.com/9cx6ia

[mischterlampe]

I may have spotted the chain of events that is the source of the problem.

Because it is my birthday , I would like to ask you what your plans are about this issue. I know that you have a lot of tasks to do, so I just want to know if you are working on that or not? My hope is still alive.

[epilectrolytics]

Luminance is not constant across different frames.

Oh I see

But I have another stupid idea

From what I see the problem is to correctly gauge the average luminance from a very short warmup run.
In my experience light tracing is always way faster than eye tracing, when run on all cores it converges in seconds where the GPUs needs minutes.
If extending the randomly sampled warmup run substantially (like 10x or more) would help with getting better estimates of the average luminance necessary for the metropolis run this would not slow down OCL rendering because light tracing is always finished earlier.
It would only look strange in the viewport where for the first minute or so the added light pass has bad quality because of random sampling before metropolis magic kicks in.

There could be a problem with SDS-cache though if that needs good light pass quality from the start.

[Fox]

With BlendLuxCore from may 13, there is still little flickering, but maybe longer render would cure it.
If you make floor perfect mirror, there is strong sun spot on the wall, but with roughness textured and bumped, it's very random.
Same with the shadow of decorative branch in vase, which reflects from black plastic to wall, when the black plastic is very rough, then the branch shadow flickers on the wall.

Random
Light Rays 90%
256spp (Ryzen 3900x ~2.5 min)
Caustic Cache with periodic update: 0.25, 0.003 m, 10d, 1, 99.571, 0.001 m

https://streamable.com/gf2a59

[mischterlampe]

I found a solution

It is not the render setting or coding solution, but a post production solution.
In this case I don't mind, because the rendering goes to after effects anyways and the result is how i wanted it to be.

I used Neat Video, a plugin to mainly reduce noise, but also to reduce flickering in lamps, for example.
https://streamable.com/rgfkp1
It takes about 20sec. to get it set up. Boom.

Thank you all guys for your tips, and, of course, it would be better, if the renderer can produce such a result by its own.
I don't know if it is possible to do a pre-animation rendering, were just the brightness of such reflections is calculated and averenged, e.g. for 4frames.


PS: I spent the money to the luxcore team.

[B.Y.O.B.]

I used Neat Video, a plugin to mainly reduce noise, but also to reduce flickering in lamps, for example.
https://streamable.com/rgfkp1
It takes about 20sec. to get it set up. Boom.

I wonder if the result would improve if you would de-flicker a caustics-only rendering (cpu render, 100% light paths).
You could then add it on top of a render without caustics.