Dept of Field noise

[almsdesign]

Hi, all!
I have an issue that is coming with me from project to project and I don’t know how to solve it. The issue is a noisy blur when you have camera depth of field (see attached, denoiser is turned on). The more depth of field I have, the more noise it has, without the depth of field image is very clean.

I tried all I thought can help and all I found here, and the result is the same…
So, here is the list of things I tried:
1. Increase samples. I had 900 samples and tried to increase them to 6000. No changes
2. Adaptive sampling, not a big difference, little better when min samples are lower.
3. Filter size, hot pixel removal also didn’t help.
4. 2x dimensions and then rescale it
5. Direct lightning and Path tracing gave the same result (PMC is taking too much time, I want to do an animation, so 12 minutes per frame is really painful)

I am rendering in 16-bit, EXR Octane. And this situation really upsets me, because I absolutely love this render, am I doing something wrong? I see a lot of octane artists with a beautiful clean depth of field and I don’t see many topics with this problem in the forum. Adding dof in post is not an option, it gives me a bad result, much worse than in octane.

I would be very grateful for ANY suggestions or tips, thank you all

[elsksa]

Hi,

Usually, it's often the following:
• Plausible shading (especially specular reflection always requiring a minimum amount of roughness, and refraction)
• Appropriate lighting (avoiding extreme "intensity values")
• Tweaking Adaptive Sampling (requires the "Noise" AOV, otherwise it's hit or miss).

Worth mentioning that Parallel Samples can be pushed to maximum (32) if VRAM allows.
Secondly, Caustics Blur would often be also maxed out when caustics (reflective or refractive) are not needed. It immensely help, depending on the scene.

Most of it has been grouped into this page (for further details and as a "handy reminder"). An other shameless plug, I've recently posted about the Universal Camera.

DoF in Octane will certainly be superior to doing it in post. It shouldn't be a problem from what I see on the provided screenshots.

[almsdesign]

Hi,

Usually, it's often the following:
• Plausible shading (especially specular reflection always requiring a minimum amount of roughness, and refraction)
• Appropriate lighting (avoiding extreme "intensity values")
• Tweaking Adaptive Sampling (requires the "Noise" AOV, otherwise it's hit or miss).

Worth mentioning that Parallel Samples can be pushed to maximum (32) if VRAM allows.
Secondly, Caustics Blur would often be also maxed out when caustics (reflective or refractive) are not needed. It immensely help, depending on the scene.

Most of it has been grouped into this page (for further details and as a "handy reminder"). An other shameless plug, I've recently posted about the Universal Camera.

DoF in Octane will certainly be superior to doing it in post. It shouldn't be a problem from what I see on the provided screenshots.

Hello elsksa, thank you so much for the answer. I will study in detail what you sent and try to apply it in the scene and post the results here.
p.s.
I'm constantly researching this forum and seeing your answers, I want to say that they are just super professional and interesting!!

[almsdesign]

Hi,

Usually, it's often the following:
• Plausible shading (especially specular reflection always requiring a minimum amount of roughness, and refraction)
• Appropriate lighting (avoiding extreme "intensity values")
• Tweaking Adaptive Sampling (requires the "Noise" AOV, otherwise it's hit or miss).

Worth mentioning that Parallel Samples can be pushed to maximum (32) if VRAM allows.
Secondly, Caustics Blur would often be also maxed out when caustics (reflective or refractive) are not needed. It immensely help, depending on the scene.

Most of it has been grouped into this page (for further details and as a "handy reminder"). An other shameless plug, I've recently posted about the Universal Camera.

DoF in Octane will certainly be superior to doing it in post. It shouldn't be a problem from what I see on the provided screenshots.

I have tried changing the settings and it seems to me that result is the same. I have tried different materials, with or without roughness (also with very little roughness). The lighting in the scene is very regular, just a simple hdri with very little power, and some area lights also with a small power (because of the size of the drone, it is like 10 cm). Adaptive sampling I also changed according to the noise map.
I also changed parallel samples to maximum and maxed out Caustic Blur. You can see my updated settings attached.

Maybe I'm wrong somewhere, but so far everything I've tried does not bring any changes. Anyway I'm very grateful to any advices

[james_conkle]

That’s looking pretty clean but yes DoF can be quite expensive. Your settings look good I would suggest using the denoiser as well and you could blend between the denoised image and the raw beauty so you can recover fine details. Otherwise I sometimes need 8-10k+ samples in extreme conditions.

The only other thing might be scene scale/ray epsilon . If my scene is super small or big I rarely but sometimes notice an improvement by adjusting the ray epsilon accordingly.

[elsksa]

Joining James. More samples will be needed when everything else has been done.
And at some point, if it "feels slow", it's a hardware matter.

[almsdesign]

That’s looking pretty clean but yes DoF can be quite expensive. Your settings look good I would suggest using the denoiser as well and you could blend between the denoised image and the raw beauty so you can recover fine details. Otherwise I sometimes need 8-10k+ samples in extreme conditions.

The only other thing might be scene scale/ray epsilon . If my scene is super small or big I rarely but sometimes notice an improvement by adjusting the ray epsilon accordingly.

Yes, I have the denoiser also turned on, it is doing a good job on not blurry parts, but with blurred dof parts it is not doing anything. I tried to adjust ray epsilon, but it seems that in this case, it does not solve the problem. May I'm too demanding and this is a normal condition for dof?

Thank you for your answer

[elsksa]

I tried to adjust ray epsilon, but it seems that in this case, it does not solve the problem. May I'm too demanding and this is a normal condition for dof?

It won't, especially when the scene-scale is plausible and technically correct.
There should rarely be a need to change the Ray Epsilon, and is often adjusted when project files have wrong scene-scales.

Crank the samples to the enough-limit by accordingly adjusting the Adaptive Sampling (where it basically keeps sampling areas that require more samples, and stop in areas that have been sampled - all of it being manually set).

[almsdesign]

Changing the amount of samples is not doing much difference. Also, I set adaptive sampling according to noise map, so I think this parameter is right.

I'm sorry if I bored you, I just have this problem often and I would like to understand if I can do something about it

[elsksa]

It looks "clean enough". It's either a matter of subtle tweaking or gentle denoising.

I set adaptive sampling according to noise map, so I think this parameter is right.

In case of uncertainty: https://youtu.be/dHTnaYjj3g4

I'm sorry if I bored you, I just have this problem often and I would like to understand if I can do something about it

You are not bothering us at all. That is what this forum is for.

Slightly off topic: this is highly recommended.