Maxwell Render | Official public forum

It took 54 hr 13 min 7 sec to reach 25 samples. The white liquid (plastic material) and its reflection in the glass with the red liquid (dielectric) are still noisy.

The simulation uses the "cut surface" method for showing liquid in a glass.



This render was started before beta 1.2.2 or 2a were released. So I don't know if this was a known or fixed issue.

In these cases, especially if you have an already noisy surface that will be reflected on another surface, make your render at double or triple the size, and scale it down. The noise will become almost imperceivable and rendertime will most likely be cut in half, or more.

Beautiful image Frances, Maxwell really is starting to replace reality!

a bit noisy ( the milk above all) perhaps... but so beautifull

The glass with the red probably alcoholic liquid hehe it looks better than the milk one does, also the red one is less noisy than the milk one is. The one with the milk in it looks like it would be sugar in the glass :/

Anyways, i still like your glass image Frances. Especially the glass with the red alcoholic liquid in it

/ Max

[quote="PiledotNET"]Frances, I'm having the same problem with dieletrics.

In my case I used a Plastic Material with a subtle SSS. The milk is noiseless, but the dieletric is still very noise. Even with heave postproduction (specular bloom and image resize) you can still see the noise.


the glass with milk in my opinion look quite more noiser that the one with wine..

Well, my point of this thread was the noise in the plastic material. After 54+ hours and 25 samples, IMO it just shouldn't be there at all.

I notice noise on the groundplane (diffuse mat) behind the two glasses. The noise seems to be coming from light passing through the dielectric material onto a plastic or diffuse material. It doesn't seem to show up with light passing through a dielectric material onto another dielectric mat (the glass to the wine).

Dielectric material settings:
Glass: IOR 1.51 Abbe 155 Absorbance .1
Wine: IOR 1.33 Abbe 155 Absorbance 0

Mihai Iliuta wrote:In these cases, especially if you have an already noisy surface that will be reflected on another surface, make your render at double or triple the size, and scale it down. The noise will become almost imperceivable and rendertime will most likely be cut in half, or more.

Yes, I'm well aware that reducing the size of the image lessens the appearance of the noise. Here is the image reduced to 50%:




The noise is still noticeable, but much reduced. So for 54 hours this is what you get - a roughly 300x250 pixel semi-noiseless image.

Here is the tiny version. You can't see any noise, but you can't see much of any detail either. So 54 hours for a 150x125 pixel noise-free image. I think the noise would have been resolved much more quickly had I just rendered the original at 150x125, but that's beside the point.



But doesn't rendering a larger image take longer to resolve to a certain level of quality? In order to get a 1000x1000 pixel final size, you would render at 2000 or 3000 pixels sq. Any estimates on how long a 3000x3000 pixel render would take to resolve to a noise level that is the same as the original I've posted here?

The issue here is with the noisiness of the materials and the cause of it - not the workaround.

In this case it will be better to render large, and render it for say 10 hours.

http://www.maxwellrender.com/forum/viewtopic.php?t=3990

He says he rendered the smaller images also for 20h (not the one you see displayed) and it looked a lot worse than the larger reduced image.

Well, I guess this goes into the "Strange But True" chapter of the Great Big Book Of Maxwell.

Mihai Iliuta wrote:In this case it will be better to render large, and render it for say 10 hours.

http://www.maxwellrender.com/forum/viewtopic.php?t=3990

He says he rendered the smaller images also for 20h (not the one you see displayed) and it looked a lot worse than the larger reduced image.

Here is the result after 10 hours. Original size 2400x1800 pixels, resized to 800x600 and cropped:



Here is the "milk" - 800x600 (cropped) after just 1 hour.



I still think this needs to be addressed by NL as a dielectric material problem.

Frances, I think it is the same problem like here:
http://www.maxwellrender.com/forum/view ... 14&start=0

The light travel two times through the glass, one time inside to the plastic and than back to the camera. It seems to be, that is the worst case for Maxwell. Here NL must do some improvements in the future.

... and the shower example show the same case.

http://www.maxwellrender.com/forum/viewtopic.php?t=3990