LuxCoreRender Forums

Hi all

I'm making some tests with Sharlybg's Nakaging Capsule, and I've seen that with noise threshold of 1 there's still some noise on my test. But I cannot go below 1.

Can this value be set as real threshold? Like 0.001? I think we should work in this case with the same units as other render packages.

Also, I've seen that changing Adaptive strength from 0.7 to 0.95 duplicates the time but, with the same threshold, also the image gets much less noise.

This is with Adaptive Strength of 0.95 And this one with AS of 0.7 Regards

And this is with Adaptive Strength of 0:

I'm not sure to understand but you seems the confuse the role of 2 parameters: sampler "Adaptive strength" and halt condition "Noise threshold".

If you want a rendering with less noise just lower halt condition "Noise threshold". "Adaptive strength" has no role at all in the amount of noise at the end of the rendering, it can only speed up the time to get there.

Sampler "Adaptive strength" sets how "adaptive" is going to be the samples distribution over the image: 0.0 means totally uniform (no adaptive at all), 0.95 means high focus on zones with more noise (i.e. they will receive 20 times more samples than others, 1.0 / (1.0 - 0.95) = 20).

Thanks for the clarification, but giving the same threshold, I see on this scenes no advantage about adaptive. I expected much less noise on the dark area compared to the image of adaptive 0.

Hence my confusion.

Asticles wrote: ↑Thu Apr 26, 2018 2:33 pm Thanks for the clarification, but giving the same threshold, I see on this scenes no advantage about adaptive. I expected much less noise on the dark area compared to the image of adaptive 0.

The advantage of adaptive is only in the time to have all image pixels under halt noise threshold. By definition, the amount of image noise, using the same halt condition threshold, will be (about) the same.

Ok, and what about lowering noise threshold below 1?

Edit: It's also interesting that the image with adaptive 0.95 does indeed have less noise on the walls. And lasted much more time.

Thanks

Asticles wrote: ↑Thu Apr 26, 2018 2:39 pm Ok, and what about lowering noise threshold below 1?

Uh, the noise threshold is something like 0.004 (use 1/256 has base unit), are you still confusing "Adaptive strength" and "Noise threshold" ?

In BlendLuxCore I exposed the noise threshold as an integer ranging from 0 to 256, which is devided by 256 to get the noise threshold.
It should be possible to set it to 0. In this case a very small noise threshold is used: https://github.com/LuxCoreRender/BlendL ... alt.py#L36

Well & hopefully correct, I am using Noise Threshold thinking of noise quantity in percentage (1-100%). Since the binary range is 0-255, i simply multiply wanted percentage with 1/100 of that = 2.55. ie. 10% noise = 10 * 2.55 = 25.5 = ~ 26

Finding the right amount, am thus starting at 255 (100%) and then jump/switch through max-min until i find the right value within preferential style and budget (time & quality; Scene & Client dependent).
Max. 255 > fast render, low quality
Min. 3 > slow render, high quality

Maybe would be better from users POV, if this is calculated in behind. Artists and common users are used to decimal vs binary

Dade wrote: ↑Thu Apr 26, 2018 2:44 pm

Asticles wrote: ↑Thu Apr 26, 2018 2:39 pm Ok, and what about lowering noise threshold below 1?

Uh, the noise threshold is something like 0.004 (use 1/256 has base unit), are you still confusing "Adaptive strength" and "Noise threshold" ?

I see that adaptive strength only affects speed. OK

Now I'm talting about Noise threshold only. What I see is that going from threshold 1 (which shows some noise on the scene) to 0 makes the gap very large. While with values of 1 the render time is 8min, with a value of 0, after 20 minutes rendering only 18% of pixels are converged.

I was asking to have the possibility to enter values below 0 more gradually.

Regards