[tom]

Thanks for the scene, Max. I found several problems and corrected them all.
Here are 8 min humble test renders with my P4.
Tomorrow, I will send you the corrected scene and make further detailed explanations.


* The above is an APNG image and you can view it animated with Firefox or Opera.



Also here's a good reference image:

Code: Select all

http://upload.wikimedia.org/wikipedia/commons/1/17/Space-Navigator.jpg

Basically, the problem with noise in your render is all about unrealistic camera and emission parameters. Literally, it's burnt.
There were numerous other things I've changed with the materials and you'll see them better in the scene I will send.

[tom]

Link has been PMed.

[Bubbaloo]

My guess is that this is more of a core issue

Nice render, Tom/Maximus!

[Maximus3D]

Good job Tom, and thank you so much for taking the time to take a closer look at that scene to see all the bad stuff i done with it. i'll take a look at it tomorrow, had no time today as i just got back home from Copenhagen.

Your result is very clean and soft looking, it sounds like it rendered pretty fast on your ancient hardware. Impressive is probably the only fitting word i can find now.

Now i clearly see the difference between a real professional in action and a noob

/ Max

[simmsimaging]

Tom: any chance you could explain a bit about what those changes were? Some kind of a hit list for optimizations would be very helpful as this problem crops up pretty often I think.

b

[KurtS]

thats an excellent rendering. Better than "other engine" as far as I can see...

(and thanks Tom for pointing to the Apng format. I've downloaded the apng editor plugin for firefox, and it works great!)

[-Adrian]

As simmsimaging, i'd like to know what makes the difference in this scene.

p.s. I really hope apng takes off, a gif replacement is overdue

[wimver]

Tom,

does that mean that this kind of noise can always be eliminated? I had similar problems a while ago, and had a hard time denoising in photoshop (neatimage can't handle this kind of noise either, without ruining the environment...)

Wim

[tom]

I'm glad you like it and thanks for the comments.
OK, here we go step by step:

1) First of all, when I open the scene I saw this torus emitter in the core having 8192 triangles. A real show stopper. Simply do-not-do-it.



Not only because it has a lot of faces, it's because torus as a shape creates more indirect illumination between the gap. Doing so will end up with heavy noise, moire patterns and long rendertime. Instead, find more efficient and clever ways to simulate the desired illumination.

I quickly got rid of this and replaced it with Studio's primitive geosphere from the library. You may think it still has lots of faces and yes, normally I wouldn't do that and model something simpler but I had to proceed fast. Then I scaled down this sphere in Z axis and located this newly created ufo-like shape under the controller. I made sure it's penetrating it and decided to left most of this object inside the controller because I only need bottom hemisphere for the emission. Finally wrapped it with MXI for adding more fun. Now it's faster and clean!



2) Emission for that small LED was 5000 watts @ 13.8 efficacy. Truly unrealistic and a real candidate for breaking the camera limits and as well as physical properties of other materials under a possible camera-compensated exposure. Always stay in reasonable amounts or expect the unexpected.

3) Camera Shutter was 4179.592 @ fStop 1.4 under above illumination. It's not guilty for noise but you may guess how this tricky exposure messes up the setups of other materials. Because, if the rest of the materials were OK, then they had to render good. This means, when building materials you should work with certain realistic values under real-life illumination.

4) There was an unseen part in the middle of the above torus emitter increasing indirect calculations. Keep in mind that you should better remove or hide unseen (truly useless and which won't affect the render in any way) areas before to render your scenes. Especially, if they are directly related to emitters. For example, you have 2 boxes and 1 has an emitter inside of it. This would slow your render down. It means most of you have a lot of chance to optimize your scenes before leaving Maxwell to suffer alone.

5) Again this is not about noise but HDR map was not 2:1 (latitude map) but a 1:1 (lightprobe map). You can easily make these kind of panoramic transformations in HDRShop. Otherwise, expect incorrectly placed reflections from the environment. Well, it's up to you, this was just a suggestion and maybe you already knew it.

6) And finally, there's nothing else I dramatically changed more. Just tweaked and optimized the materials for more realism and some speed improvement. Like if it's shiny black plastic, I go for 1 BSDF with low Nd instead of wasting an additional black or dark color lambertian in the composition. The anisotropic metal on the other hand was a real mess (sorry Max..hehe ) because I saw it has every possible mapping in every parameter like rotation, aniso, bump, rough bla bla. Now, it's much simple and of course render faster and more realistic but it's completely a subject for another thread anyway.

So, what was all about the noise you posted? It was mainly because exceeding emission under targeted exposure. This means it would be clean after a very long time... "one day" ... and well, it would converge to white, so there's no joy in waiting it to render, especially when you also* have a gigantic emitter made of 8192 faces in an unseen part of the scene. That's a real trap.

I hope these cues will help for next time and thank you for reading.

[tom]

does that mean that this kind of noise can always be eliminated?

I can easily say 95% of the time, noise has solution on user side. Remaining 5% is about known inefficient calculations like dispersion, coatings, caustics under certain conditions etc.

(and thanks Tom for pointing to the Apng format. I've downloaded the apng editor plugin for firefox, and it works great!)

p.s. I really hope apng takes off, a gif replacement is overdue

In fact the idea is not new, there was MNG like APNG but it's great today seeing APNG is on the right track. GIF is a real dino and since more than 10 years I'm trying to understand how can it survive....

[simmsimaging]

Great explanations - thanks very much Tom!

b

[wimver]

Thanks a lot Tom!
(my wife says at night I am dreaming loud about your next video tutorial.... )

wim

[flower]

Thanks Tom for a really useful insight on scene optimisation.

Bim

[tom]

Thank you all and thanks to Max for sharing this scene as an example.

[rivoli]

always a pleasure to get schooled like this, wish you'd do it more often