Community
Arnold for Maya Forum
Rendering with Arnold in Maya using the MtoA plug-in.
cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

Noise workflow - some doubts

9 REPLIES 9
Reply
Message 1 of 10
CiroCardoso3v
761 Views, 9 Replies

Noise workflow - some doubts

Hi everyone,

I am fairly new to Arnold. At our studio we seeing if Arnold could replace Corona.

So, we are exploring ways to easily create a bridge between Corona to Arnold. Starting with the noise, I already understand how to use the AOVs to see where the noise is coming from. However, what is the best way to do this? Rendering small regions until the noise is clean?

Is taking me like 2 or 3 hours just go through all the samples I need to tweak. There must be a quicker way to do this.

Also, is there any "magic" to calculate how long an image will take o render? This probably is coming from Corona, where you set a timer and no matter what the render will stop after X hours. I was trying to see if there is any correlation between the image size and the rendering time.

Any tips are welcomed

Lead Enviroment Artist @Axis Studios

Arnold Discord Server


Ciro Cardoso

EESignature

Tags (2)
Labels (2)
9 REPLIES 9
Message 2 of 10
lee_griggs
in reply to: CiroCardoso3v

These tutorials may help you determine where the noise is coming from. Can you post some sample renders (cropped if necessary) and/or scene screengrabs with info of what is in the scene, lights, shader etc to give a better idea?

Lee Griggs
Arnold rendering specialist
AUTODESK
Message 3 of 10
madsd
in reply to: CiroCardoso3v

Nice project Ciro!
Once you nail the AOV probing and understand the render panels samples ( and lights samples ), you should move to the noice.exe denoiser. It is basically what the Corona denoise slider in the frame buffer does. It can remove noise after a render is done, and decrease render time significantly, while still having some noise in the final frame.
You can knock out stuff at insane speed when you combine optimization and the post denoiser.

If I want ultra fast speed, I flip the process around, I make sure I render with tons of noise, then use the high quality post denoiser, and push that to the limit. Once I found limit of post denoiser, I start to increase quality of problematic noisy rays in the actually render. This way, I check the post denoised image, and not the actual render.
This approach will ensure you arnt overkilling with render settings and thus get a slower output. No point in cranking the render samples if the post denoiser can clean it up - kinda approach.

I'll follow your progress and assist when needed.

Message 4 of 10

Thank you both for your reply.

@Lee Griggs, I have been using those to clean the render. Using the workflow where you set all the samples to 0 and start with the diffuse and clean each AOV. The scene where I am testing Arnold is this one

3935-arnold6k-500px0000.jpg

All the materials are Substance files. Apologies for the watermark, but I sent this to render farm to see how long would take. The render was 6500px and took 39 hours on a Xeon Gold 6126 2.60GHZ with 128GB Ram.

To get a clean render like this I had to use the following settings:

3936-samples.jpg

These are the samples that cleans the render. I also have a Skydome with 7 samples and 20 spotlights with 6 samples.

@Mads Drøschler a big thanks for those tips. It makes perfect sense. However, just to double check. The sample settings I use for a 2k render will be the same for a 6k? Because I can then quickly do a preview in 2K and send the final render as 6k.

Lead Enviroment Artist @Axis Studios

Arnold Discord Server


Ciro Cardoso

EESignature

Message 5 of 10
thiago.ize
in reply to: CiroCardoso3v

Give adaptive sampling a try. Set the diffuse/specular/transmission/etc. samples to 1 or maybe at most 2-3 for some of them if you want to tune it, and instead let the AA adaptive sampling do the heavy lifting. It might not always be as optimized as careful tuning, but it's much easier to do and there's less risk of messing up the samples and actually getting slower renders.

Then as others have mentioned, use the denoiser to get rid of the remaining noise (increase the AA_adaptive_threshold to make it noisier).

Message 6 of 10
madsd
in reply to: CiroCardoso3v

Skylight should come from an arnold light node set to skydome, the lamp has samples, adjust the samples on the light before the render panel. 11 diffuse is a ton, indicating a bad lamp setup.

Dont forget to work with the adaptive sampler and probe the AA_Inverse AOV.

Try farm a couple with a good stack of noise on and work your way into the noice.exe denoiser.
You will soon find 39 hours is not needed.

Message 7 of 10
lee_griggs
in reply to: CiroCardoso3v

Are you using light portals outside the window? That way you shouldn't need so many skydome light samples (I usually use 3 or 4 samples at most for all lights). You might be able to reduce the Diffuse samples using a light portal too as I am sure 11 is very expensive to render.

Increasing ray depth by 1 can effectively double your render times. Do you need 4 for both Diffuse and Specular? I imagine 1 or 2 is fine for specular ray depth as there are not many shiny surfaces in the scene.

Again, I don't see any glass surfaces, so you could probably reduce the transmission ray depth.

Transmission 7 would only be necessary if you had a lot of rough glass in the scene so that could be reduced too.

Lee Griggs
Arnold rendering specialist
AUTODESK
Message 8 of 10

Thanks everyone for all your replies. I never been in a community so friendly and helpful.

Let me go by stages

I am using a skydome with a HDRI connect to it. I am using 7 samples, because at the time I tested it that seem the best value.

The idea to use the Adaptive Sampler, is to help me identify where we need more samples?So for example, if is a glass surface I need to check my Transmission samples?

I tested the standard settings, which took 2 hours to render a 6.5K image and the denoiser, did an amazing fantastic job. I can hardly believe the final output. However, I have 2 questions:

(1) does the denoiser affects the AOVs? If I need my AOVs clean to reconstruct the beauty pass is there a option to clean then?

(2) any idea why my render looks like this

3950-render.jpg

And my denoised file like this

3951-denoise.jpg

My Render file was saved with a Gamma of 1.0.

I am using Light Portals outside the windows. I wonder if the IES files (you don't see them because I am using a decay filter) could be contributing for some of the noise? 11 samples is taking me 39 hours to render. As a side note, Corona still took 27 hours to render the same image.

What do you recommend in terms of Ray depth? I found that 4 for Diffuse helped with the light, but I think in that case, I am doing things wrong. I will have a look on that.

The reason I increased so much the samples is because while checking the AOV, those were the values that would clean the indirect AOVs.

Thank you so much for all your help.

Lead Enviroment Artist @Axis Studios

Arnold Discord Server


Ciro Cardoso

EESignature

Message 9 of 10
madsd
in reply to: CiroCardoso3v

The MAX post grading is not injected into the AOV's so if you crank the exposure up or down, or play with mid low and highs in the environment panel, those settings wont carry over to the AOV's that may explain the difference you see.

3 diffuse bounce is enough in most cases, you wont get that much out of the 4th - from 1-2 you see the biggest difference.

And yes, the denoiser is awesome, I could not belive back when I tried it the first time, just how good it is at removing noise while presering texture detail, also in animation.

For adaptive you need to probe the "AA_inv_density" AOV, it will feedback intensive areas, marked as black.

I added some examples here in no paricular order, with a bunch of different settings to optimize my particular case here.
Note the plant is dominant, the actual problem area is under the table, the wall between the legs on the left chair. The plant just flashes since SSS.

Do note I use the GPU in an early version but the same applies still.

3952-1.png

3956-3.png

3953-2.png

3954-07.jpg

3955-09.jpg

I tend to use the free VFB+ framebuffer, you can grade with it, and load a linear AOV into it, and sync up with your wanted grading. since it processes the image post, you will get identical results.

Message 10 of 10
michiel.van.gasse
in reply to: madsd

Ah samples control for GPU. ^^

Can't find what you're looking for? Ask the community or share your knowledge.

Post to forums  

Autodesk Design & Make Report