PDA

View Full Version : Noisy radiosity?! HELP!!!



cyclopse
12-07-2016, 08:19 PM
So, I'm using radiosity to light a scene... but holy crud is LW radiosity noisy! Any way to get rid of this noise? Here's a test:

https://youtu.be/i0WyHHn8uWg

Any help is greatly appreciated this is driving me NUCKING FUTS!!!

kolby
12-07-2016, 09:24 PM
It doesn't look like noise but flickering from interpolated GI.

cyclopse
12-07-2016, 09:30 PM
It doesn't look like noise but flickering from interpolated GI.

So should I crank the RPE, or uncheck interpolated, or what would help?

kolby
12-07-2016, 09:47 PM
Safest way is to uncheck interpolated. It will raise render time significantly because of noise you must kill with higher AA settings.

Danner
12-08-2016, 02:17 AM
You can use interpolated, but you need to enable animated radiosity cache.

kolby
12-08-2016, 02:37 AM
I'm not sure but I think interpolated radiosity does not work well with deformed object.

S0nny
12-08-2016, 07:17 AM
Do it in non-interpolated if you can: your render will takes longer, but it's much faster to setup and troubleshoot.
With interpolation you'll waste a good amount of time just testing how much RPEs, how tight the GI solution needs to be, and you'll get a very slow computation for animated radiosity cache.

gerardstrada
12-08-2016, 08:47 AM
You might want to try DP Denoiser to minimize render times with non-interpolated GI. Here's a setup for outdoor scenes:

http://forums.newtek.com/showthread.php?71751-Extra-Buffer-nodes&p=1397271&viewfull=1#post1397271



Gerardo

RebelHill
12-08-2016, 09:40 AM
You can either do it brute force, no interpolation, which will add a LOT of rendertime, or work through finding the right interpolated settings to get what you're after. Animated cache will NOT help you here.

Start with this...


https://www.youtube.com/watch?v=0YFZ2av-BLg

gerardstrada
12-08-2016, 10:04 AM
Do it in non-interpolated if you can: your render will takes longer, but it's much faster to setup and troubleshoot.
With interpolation you'll waste a good amount of time just testing how much RPEs, how tight the GI solution needs to be, and you'll get a very slow computation for animated radiosity cache.

Agree :)



Gerardo

Chrusion
12-08-2016, 11:38 AM
I used to use Interpolated MC and animated cache (bake scene primary rays at a frame step of 10), but after NT changed caching to individual files for each step, instead of the original single file of LW 9, I decided to give brute force (non-interp) MC a try. Yes, overall scene render time increased, but it actually feels like its a draw, because I don't have to bake the scene and experience progressively longer frame bake times after about 1,000 frames.

Now I only use MC with 5 primary rays and 2 bounces for all of my outdoor renderings. Indoor scenes probably need 10 - 20 rays. With AA set to 8 samples (lights and surfaces set to 1 in render globals) and 32 passes at threshold of 0.001 and 0.5 subsampling, the pixel-level noise gets nicely cleaned up, leaving a hint of "film" grain. 1080 HD frames take on average 10 - 20 mins to render, but now I don't have to worry, EVER, about splotchies, flickers, and lack of GI definition (contact shadows), nor spend a couple hours baking an animated GI cache and have that extra set of data files transfer over the render farm network.

RebelHill
12-08-2016, 12:13 PM
I don't have to bake the scene and experience progressively longer frame bake times after about 1,000 frames.

If you get that, you're doing the baking wrong.

Chrusion
12-08-2016, 12:29 PM
How can you do baking wrong? You set RPE and SBR as desired to get good GI definition, set frame step for amount of motion in scene (5-10 for high motion, 10-20 for low) and bake. How is that wrong? Every 10 frames in most all of my camera tracks and object motions gave decent results. I always thought the slowing of baking was due to greater and greater areas of the scene becoming visible adding more and more sample points to the cache as prior samples went out of view, but remained in the cache.

RebelHill
12-08-2016, 01:40 PM
I always thought the slowing of baking was due to greater and greater areas of the scene becoming visible adding more and more sample points to the cache as prior samples went out of view, but remained in the cache.

That is why it happens, but it shouldnt. If you get this ocurring you're using the animated/not animated mode in the wrong context. Its not the case that "somethings moving, so animated on".

Niko3D
12-09-2016, 03:02 AM
just bake the scene in automatic...and after locked. Very easy...very usefull...

gerardstrada
12-13-2016, 03:09 PM
Now I only use MC with 5 primary rays and 2 bounces for all of my outdoor renderings. Indoor scenes probably need 10 - 20 rays. With AA set to 8 samples (lights and surfaces set to 1 in render globals) and 32 passes at threshold of 0.001 and 0.5 subsampling, the pixel-level noise gets nicely cleaned up, leaving a hint of "film" grain. 1080 HD frames take on average 10 - 20 mins to render, but now I don't have to worry, EVER, about splotchies, flickers, and lack of GI definition (contact shadows), nor spend a couple hours baking an animated GI cache and have that extra set of data files transfer over the render farm network.

Yes, using non-interpolated and just few rays with a good denoiser in post tend to do the trick in many cases.

Interpolated has its place as well, but it's very scene dependent. One of the main downsides is the lack of shading definition. Mainly in small details and complex objects. Some parts get shades some others not (smaller details) and an AO pass don't really help because that's like adding radiance occlusion twice. We need to get a radiance occlusion pass, subtract from AO pass and add this result instead.

When geometry displaces and deforms, a trick that sometimes work (instead of baking the sequence) is to force quick interpolation calculation for each pass. This is done by using Dithering Motion Blur and an envelope in lighting or focal-plane settings (i.e. a cycle per frame in a light intensity or motion blur). LW averages all interpolated solutions from each AA pass and this tends to avoid flickering. As the other solutions, it depends on each case and quality level expectations.



Gerardo