Author Topic: Mental ray user with a couple of questions  (Read 11006 times)

2014-02-21, 22:11:03

blank...

  • Active Users
  • **
  • Posts: 209
    • View Profile
Hello everyone

Yesterday i tried out Corona for the first time. Well, i just run benchmark scene, but seeing how fast it rendered i had to test it further.
I dug out an old scene i once used to test Vray against Mental, to see how Corona works. Since i know nothing about it's options (i see same lingo as in Vray, so i gues Vray users are in luck there), so i left everything at default, just switched secondary solver to HD cache, as Rawalanche suggested in one post about interiors.

Here are the results. Mental rendered in 33:04, Corona did 22 passes in 33:13.

Mental ray:
http://i.imgur.com/kKPTNXb.jpg


Corona:
http://i.imgur.com/V8sfN6e.jpg


Beside different color cast (sun in both renderer is set from sky), and considerably grainier Coronas render, there are a couple of other things like missing fine detail shadows, and some strange shadows.

Here are they highlighted with Mental ray cutouts:
http://i.imgur.com/52xvow5.jpg


Materials are relatively simple, i don't think i messed up there (i hope i didn't :) ), so what could be the problem? What settings should be tweaked to get better results, and get them, if not faster, at least in the same time? Considering how fast that benchmarks scene popped out, something must be "wrong" in my scene.

Here are both scenes, if someone would be so kind to take a stab at it:

Corona:
https://app.box.com/s/h68fz09k2s3cy7n2e0wb

Mental:
https://app.box.com/s/3rqai98f4nbvq8zqepun

Thanks in advance!

2014-02-21, 22:16:25
Reply #1

Ludvik Koutny

  • VIP
  • Active Users
  • ***
  • Posts: 2557
  • Just another user
    • View Profile
    • My Portfolio
The shadows mental ray produces using AO are the incorrect ones. Corona does a lot more correct GI than mental ray, but if you are used to AO look, you often expect contact shadows even on places where they do not belong. You can test it simply by doing reference rendering of your MR scene in iRay ;)

Also, another thing that may cause lack of contact shadows are unrealistically bright materials. Your white should never be over RGB 220 ;)

I think the reason your rendering looks weird is that you are used to a lot of fakes from Mental Ray, and you try to apply them to Corona as well :)

2014-02-21, 22:34:41
Reply #2

blank...

  • Active Users
  • **
  • Posts: 209
    • View Profile
I think the reason your rendering looks weird is that you are used to a lot of fakes from Mental Ray, and you try to apply them to Corona as well :)

Quite possible, what about the speed? What can be done to speed things up and get cleaner render in less time? Or is it unreasonable to expect both better, and faster, from an unbiased renderer?

2014-02-21, 22:37:56
Reply #3

racoonart

  • Active Users
  • **
  • Posts: 1446
    • View Profile
    • racoon-artworks
I had a quick look on your scene:
1) put a portal in both of the windows, you were using 1 which was not quite well aligned, the other window was portal-free
2) walls could be a bit less bright, they are at rgb 215 atm, I'd go for 180 maybe.
3) more subdivisions for the couch and the jagged shadows will be gone ;) It's a problem related to smoothing and not enough subdivs
4) as rawalanche said, shadows are quite fine for the fabric, but this material was also quite high, maybe you'll be more pleased with the results if you lower the brightness.
Any sufficiently advanced bug is indistinguishable from a feature.

2014-02-21, 22:43:42
Reply #4

blank...

  • Active Users
  • **
  • Posts: 209
    • View Profile
1) put a portal in both of the windows, you were using 1 which was not quite well aligned, the other window was portal-free

On purpose, because MR scene also has only one portal.

Quote
2) walls could be a bit less bright, they are at rgb 215 atm, I'd go for 180 maybe.

Agreed.

Quote
4) as rawalanche said, shadows are quite fine for the fabric, but this material was also quite high

Yes it is, honestly, i remember when i first created that scene there was a reason why i left all of materials with diffuse level at 1 (which i usually never do), but now, i have no idea why :) But for testing purposes i also left all of Corona diffuse levels at 1.

2014-02-21, 22:50:12
Reply #5

racoonart

  • Active Users
  • **
  • Posts: 1446
    • View Profile
    • racoon-artworks
For the speed issue... I disabled bitmap paging in your scene and the rays/s shot up from 600.000 to nearly 3.000.000  :D
I guess this may help a bit

[Edit]
maybe it's not a big deal with MR using only 1 portal (I don't know mr very well) but it's not quite useful to do so in corona - it's possible but it won't help so much getting rid of noise.
Any sufficiently advanced bug is indistinguishable from a feature.

2014-02-21, 22:59:28
Reply #6

Ludvik Koutny

  • VIP
  • Active Users
  • ***
  • Posts: 2557
  • Just another user
    • View Profile
    • My Portfolio
For the speed issue... I disabled bitmap paging in your scene and the rays/s shot up from 600.000 to nearly 3.000.000  :D
I guess this may help a bit

[Edit]
maybe it's not a big deal with MR using only 1 portal (I don't know mr very well) but it's not quite useful to do so in corona - it's possible but it won't help so much getting rid of noise.

Bitmap paging setting is stored with scene? O_O

And to blank...  exactly as DeadClown says... always make sure you have bitmap paging in asset tracked disabled. I am also currently working on your scene...  and i will post the result once i am finished :)

2014-02-21, 23:02:49
Reply #7

blank...

  • Active Users
  • **
  • Posts: 209
    • View Profile
For the speed issue... I disabled bitmap paging in your scene...

Damn, how did i miss that, it shouldn't be enabled in the first place :/

2014-02-21, 23:05:04
Reply #8

blank...

  • Active Users
  • **
  • Posts: 209
    • View Profile
Bitmap paging setting is stored with scene? O_O

That's a new one for me as well :D (i never have it enabled).

Quote
and i will post the result once i am finished :)

Thank you very much! Looking forward to experiment with it, i'm currently applying changes DeadClown suggested.

EDIT:
Damn! 15 minutes in and i stopped it, it was there, same quality as MR. I knew bitmap paging slowes things down, but i didn't expect it to have that much of an impact.
http://i.imgur.com/gBDSbiA.jpg



Now, to see what MR does to this scene with paging off.


EDIT2:
3) more subdivisions for the couch and the jagged shadows will be gone ;) It's a problem related to smoothing and not enough subdivs

Even after i added more sudivs, shadows are still jagged. Is this a bug, or a known "problem" that is not actually a problem but just the way Corona works?

http://i.imgur.com/EWL1WzA.jpg
« Last Edit: 2014-02-22, 00:14:59 by blank... »

2014-02-22, 02:09:40
Reply #9

Juraj

  • Active Users
  • **
  • Posts: 4768
    • View Profile
    • studio website
The odd bump behaviour is my only problem with Alpha5 which I use currently. The jaggednes might also be the terminating error with unbiased method as mentioned above.

Regarding lack of contact shadows, sometimes they indeed are too meager, although it might be quite correct behaviour, pure PT+PT does help slightly even if 4096 PT subdivs are used in HDCache  (I always use 4096 for finals) but of course that is not the best thing to do in interiors and Corona is largely built around this as main feature. [my white wall is between regular RGB150! to 180 if it's supposed to be quite new/white :- ) ]
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2014-02-22, 08:55:43
Reply #10

Ludvik Koutny

  • VIP
  • Active Users
  • ***
  • Posts: 2557
  • Just another user
    • View Profile
    • My Portfolio
If the jagged bump problem does not go away with increased subdivision, then simply try to lower bump amount a bit. Its hard to calculate normal smoothing with strong high-detail bump map. Also, make sure none of your bump maps has completely disabled filtering (set to 0.01)... i usually use 0.1 for bump maps... Lastly, i have done some work on your scene, but right now i am at office doing weekend shift, so result wont be sooner than this evening :)

2014-02-22, 10:11:38
Reply #11

blank...

  • Active Users
  • **
  • Posts: 209
    • View Profile
Also, make sure none of your bump maps has completely disabled filtering (set to 0.01)... i usually use 0.1 for bump maps

In this case bump is a mix of celular and bitmap, both are blured to 1.

Quote
right now i am at office doing weekend shift

I'm very sorry to hear that :)

Quote
so result wont be sooner than this evening :)

No problem, there's no rush. In the menatime, considering they are named similarly, are Coronas HD cache and Path tracing at all similar to Vrays Light cache and Path tracing? Would it help to watch some Vray tutorials on Light cache and path tracing to better understand Coronas settings/theory and what's going on behind them?

2014-02-22, 10:28:36
Reply #12

Ludvik Koutny

  • VIP
  • Active Users
  • ***
  • Posts: 2557
  • Just another user
    • View Profile
    • My Portfolio
No problem, there's no rush. In the menatime, considering they are named similarly, are Coronas HD cache and Path tracing at all similar to Vrays Light cache and Path tracing? Would it help to watch some Vray tutorials on Light cache and path tracing to better understand Coronas settings/theory and what's going on behind them?

Not really, they work similar internally, but knobs are quite different. Considering PT and HD, here are all the basics you need to know:

Increasing PT samples increases amount of secondary rays per eye ray. secondary rays handle quality of GI, reflection, refraction (including glossiness of these parameters), translucency, and so on, while primary rays handle quality of AA, DoF and Motion Blur. So decreasing PT samples is suggested in scenes with fine geometry, that needs a lot of AA, or scenes with strong DoF and/or motion blur. While increasing PT samples is suggested in scenes with difficult indirect lighting. Default works fine in most cases, and i would never go under 2 and above 64 in PT samples parameter.

HDcache is simple, you can ignore most of the knobs, except two. PT samples in HDcache define qualit of HDcache samples, and precomp density their density. Default values work fine. Except for final renders, you can raise PT samples in HDcache to 512, or even 1024 if you have lot of complex indirect bouncing. And on the other side, for quick previews, you can drop precomp density from 1 to 0.2, and then put it back to 1 for finals :)

If you have windows, or some large glass elements like glass railings or glass walls, then set refraction mode of your glass to hybrid, so it will generate transparent shadows instead of caustics and render a lot faster. If you have glass planes in windows that are just one polygon, without thickness, then you can use twosided glass mode, which does the same, but generates no refraction. Similar to thin walled mode in Arch&Design material.

Lastly... watch diffuse brightness of your materials. Your should never have too bright (over RGB 220) materials. In mental ray, you may be used for this, as mental ray will give you contact shadows even on superbright materials because you usually set final gather to very low diffuse bounces, so light falls off a lot quicker than in reality. Corona has 25 bounces by default, so light does not fall of that quickly and you get more light bouncing = less contact shadows. Also, in Corona, superbright materials can slow down rendering a bit, unlike mental ray :)

2014-02-22, 10:49:49
Reply #13

blank...

  • Active Users
  • **
  • Posts: 209
    • View Profile
Increasing PT samples increases amount of ... <CUT> rendering a bit, unlike mental ray :)


This is going down in my notebook :) Thank you!

2014-02-24, 20:21:40
Reply #14

Ludvik Koutny

  • VIP
  • Active Users
  • ***
  • Posts: 2557
  • Just another user
    • View Profile
    • My Portfolio
Ok, so i took a look at the scene.

First of all, i did change a few things:

I merged the scene into a new file and immediately assigned gray material to all objects. I always do that to prevent any garbage getting carried with the scene. Therefore i didn't match all your materials 1:1. I made gold a bit more golden too.

I also made another cloth sim. The out of the scale and floating one was not really helping the realism. The cloth now has translucent material, so it may appear that contact shadows are missing, but its just light passing through. You can always put there some opaque material to get stronger indirect shadows as light wont pass through and scatter.

I also deformed seat pillows on the sofa and fixed their smoothing as it was another thing that was kinda killing realism.

Lastly, i moved camera a bit as i didn't feel like that crazy camera shift was necessary.

So, on my PC, which should be more than twice slower than yours, i managed to get similar noise level as you at similar time. This is the output i made, both postprocessed and raw versions:





as well as  the scene itself. But to be able to render the scene, you will need last daily build. If you do not have access to dailies yet, just go to the daily build section, get daily build and request an activation. I will make sure you will get an activation code ;)

2014-02-25, 09:40:34
Reply #15

blank...

  • Active Users
  • **
  • Posts: 209
    • View Profile
Thank you very much! Tomorrow i'll have some time to dig in to it, and then i'll probably have more questions :)