Heavy depth of field noise levels - I don't get it

[snakebox]

Hi everyone,

As topic mentions, I am playing around with rendering heavy depth of field in Corona and I am getting unacceptable noise levels no matter which settings I seem to be trying.

Basic setup is an object in the foreground, interior room in the background, camera using iso 100/shutterspeed 100, f-stop 2, 2.8 or 4.

I have tried a number of different approaches mentioned here on the forums but none of them really seems to solve my issue.

It usually happens around objects that high highlights that naturally gets noisy when blurred due to the DOF, so say curtains that looks perfect without DOF, becomes impossible to keep noise free with DOF.

Settings I have tried tweaking:
Light sampler using values from 2.0 to 10.0
GI/AA balance, default 16 to 2.0
Internal res (VFB) from 1 to 4 in values
And I have tried to just let things run for 500+ passes

All these settings all change things very slightly but no combination so far seemed to actually smooth things out. So I am wondering if I am not aggressive enough on the settings or how you guys approach this? (Other than the obvious doing DOF in post).

Thanks guys! I will post some image examples of my problems ASAP.

[Adanmq]

Last time i face this problem, the solution was use Bucket mode, but it takes lots of time to render.
There is another post on the forum covering the same issue.

[juang3d]

Try to configure your bucket render to 1 initial sample, or 2 maximum, and 5 or 6 passes.

Also try to put the light sampler back to 2, the GI/AA balance, leave it at 16, the initial res to 1.

Check what happens with these settings.

Cheers!

[Stan_But]

Try to configure your bucket render to 1 initial sample, or 2 maximum, and 5 or 6 passes.

Also try to put the light sampler back to 2, the GI/AA balance, leave it at 16, the initial res to 1.

Check what happens with these settings.

Cheers!

The strong DOF with GI/AA balance 16? It was never. Only GI/AA 2-4
The light sampler leave in the default value.
Early, in A5, it was a little speed up with using bucket mode. But now the buckets and the PT are equal in the noise cleaning speed

[juang3d]

I disagree with that, check the attached pictures, the bucket picture is cleaner in the same time, also notice that in bucket mode it has less Rays/s, but it's faster cleaning noise.

BTW this pic is with GI/AA at 16.

Cheers.

[racoonart]

Bucket rendering is working fine, it's just a little basic at the moment.

Bucket rendering is working this way:
"Initial samples" are the number of passes (think of passes like in progressive mode) that are made for each pixel no-matter-what. After that each pixel in the image will be compared to it's neighbors and the "adaptive thres." condition determines how much contrast between those pixels is allowed. If the comparison fits the condition the pixel is considered to be sufficiently sampled. If, however, the condition is not met, then the next round of passes is started. The next pass is now rendered with 4 times as much samples as the pass before (if your initial samples parameter was set to e.g. 3, the next pass will be 4*3 and the next one 4 times as much as this one etc.). This will be done as many times as the "passes" setting is allowing the renderer to do.
Be aware that very low initial samples settings (1 or 2) may actually make things worse. If - by chance - the initial pass is producing a low contrast solution in some pixel areas it may not be evaluated ever after - thus potentially producing problematic spots. I'd recommend to go for 3 initial samples in most cases (depends on your GI and Light samples multipliers of course).

[romullus]

Can't wait till progressive engine gets adaptivity too. I dislike buckets to much to choose it over progressive just for adaptivity.

[juang3d]

For me it's working flawlessly, I used 1-5 for my scenes and it worked great, but I understand what you say about using 2 or 3 as base, I'll do from now on.
But what I can say is that it feels much faster than progressive and specially it feels like a sniper for noise, I don't know if it's just a feeling, but it feels like it's dedicating power where it's most needed.

I would like to be able to see where is it working but it's not a big problem.

Cheers.

[snakebox]

WOah! thank you guys!!

I will definitely try these settings today! I shall report back, but thank you! (I kind of thought the cool thing about corona was that the progressive render could do it all :/ )  one day maybe.

[juang3d]

It's just that corona is sitll evolving, is pretty awesome we can use an Alpha version of a render engine in production :D

Cheers.

[snakebox]

Yeah exactly, here at the office we have changed our production render from Vray to Corona (at least for now) because we mainly do stills, and can't be happier.

[juang3d]

We do mainly animation, and even when Corona lacks some features for animation, we can't be happier either, it's proving to be the best render engine we've used so far :D

Cheers.

[snakebox]

Update:

So I have tried bucket render and wow! What a difference. Definitely better at handling the noise in heavy depth of field scenes, and the speed is totally fine too.  I am wondering if it's generally just safer / easier to use over progressive for images / animation without depth of field? It seems to do the same as progressive, but better?

[juang3d]

Cool!

i'm happy you found this useful, a lot of people refuse to try the bucket mode, but for final rendering, if it works well, i think it doesn't matter, and IMO you have more control over quality hehe

Cheers!

[snakebox]

I don't fully understand why the bucket render has a limit of 10 adaptive passes? is the idea that this should be enough based on the threshold?  because so far when rendering fullsize 4-5K using a threshold at 0.1 it does a worse job than progressive actually, when NOT using Depth of field...