Dobromir Dyankov

Adaptive Sampling with Octane Render for Cinema 4D

An excerpt from “What’s New in Octane Render v3.x”

Adaptive Sampling is a rendering technique that disables sampling for pixels that have reached a specified noise threshold. This allows you to bump up the max. samples quite high and then rely on the adaptive sampling to figure out which pixels actually need that many samples and which don’t. It is mostly useful is in scenes that have areas that are a lot more noisy than others. If your whole image is equally noisy, this feature will not help.

Noise Threshold: This is probably the most important parameter when rendering with Adaptive Sampling. What follows below are just some of the thoughts that should be running through your head when adjusting this setting. Let’s assume you go with the default of 0.03. What does it mean?

You’d ask..
“Dear Octane Render, if the noise level of a pixel reaches the defined threshold value, please consider this pixel “clean” and stop sending additional samples in that area.”

You’d get..
“Sure, I can do that! A word of caution though – consider, with noise threshold of 0 there is no longer a noise target for me to test against. This means I will be forced to throw equal amount of samples in all areas of the image, at all times. In reality, this will render adaptivity useless thus returning the control over image quality, back to “max. samples”. On the other hand – with a noise threshold of 1, my ability to judge what’s noisy and what’s clean will be compromised. Said otherwise.. what I’d perceive as “a clean pixel” will most certainly translate to you as “wtf is this noisy mess!?” But cheer up – I will, at least, render faster! That’s pretty much it. Don’t say I didn’t warn you. I have it in writing!”

Adaptive Strength: Self-explanatory. This parameter controls how much adaptivity is applied when sampling the image. Probably 99% of the time you’d want it at the default 1. A setting like, let’s say 0.6 means adaptive sampling (with respect to the noise threshold) will be applied only 60% of the time. For the the remaining 40%, pixels will render without taking adaptivity into consideration.

Min. Samples: That’s the amount of samples that will be rendered before adaptivity kicks in. You can imagine this as a mandatory layer of processed pixels, which is there to ensure you have a good base to build on top of. Lowering this number will allow the adaptive technique to start affecting the image sooner thus reducing render times. Though this may sound tempting, taking the min. samples way down may compromise the quality of the GI solution. The default of 256 is something you should be able to live with for most scenes.

Group Pixels: Adaptive Sampling works by testing the quality of image pixels against a pre-defined noise threshold. This check-up is done on single or grouped pixels. When pixels are collected in a 2×2 or 4×4 gird, the adaptive technique will stop sending samples only if all members of the group have met the target noise levels, otherwise – we wait. It is only logical to assume that “none” should be the fastest option, and it is. However, the default here is 2×2 and until I know any better – I’d suggest you stick to it.

---

Hey, hopefully you enjoyed this post. If you did and think that the work I do is somewhat decent, chances are you may want to join the mailing list.
I keep it clean and on topic. No spam. Just saying.