Weirdness with DMC as primary GI

[thomthom]

I've been modelling a scene; while doing so I've left the settings pretty much standard, IR+DMC.


However, I wanted to do some testing on how to ensure I get sharp details, I started of first by upping the quality of the default settings, but I felt that with IR as primary I was beginning to have to push the settings so high that the render time just slowed down too much.


I also tried to enable the Detail Enhancement, which did give me some better contrast between the light and the dark areas. But I still wasn't happy about IR for this model.


So I switched to DMC and LC. But then I got a very unexpected result. As you see, the LC pass has the colours as I would expect it, but when the DMC pass render everything turns grey. From what I understand, DMC is the most accurate and give the most correct result. I expected it to be some slight colour variation, but nothing like this. Initially I thought it was due to the reflection layer of the material, but when I removed it there was no change. Can anyone shed some light to why there's such a dramatic change from IR to DMC as primary GI? ???



All the tests: http://picasaweb.google.no/thomassen.t/AssassinsGuild#

[sepo100]

Re: Weirdness with DMC as primary GI

I can see the renders now. Before not really. Comented on stone and roof cover. Lok forward to see this going further.

[Jackson_cg]

Re: Weirdness with DMC as primary GI

Sorry, I can't help with your problem Thomas, but there is one thing I may have to contribute to the issue: I'm going to commit what might be regarded as VRay blasphemy here, but I never use DMC.

I have tested and tested and tested and have yet to come across a render yet, (interior or exterior) where the massive increase in render time was worth the increase in quality offered by DMC. One day I may well look back at this post and think "you poor naive fool", but after 2 years of using VRay I now always use IM for Primary and LC for secondary. I'm sure a VRay expert would take one look at my renders and say "no sh*t!", but I have done so many identical test renders using IM & LC, IM & DMC and just DMC (left that one overnight) and flicking between them for comparison the quality difference is so minute it's just not worth using DMC IMO. Of course you could argue that the fact that I render on a 3 yr old Core 2 Duo 1.83GHz laptop has influenced my choice, but I know that when I buy my next laptop I'll still be addicted to rendering as fast as possible.

You say "I still wasn't happy about IR for this model."- can you explain what it is that you are unhappy about? Lack of detail? Lack of contrast?
If there is one quality that I feel is most lacking in VRay (no matter which settings I use) it is the effect of ambient occlusion- I always feel internal corners are too bright and indistinct. Damien once suggested decreasing the LC multiplier which I have always done for exteriors (sometimes as low as 0.1) and often for interiors (usually no less than 0.8 or everything gets too dark), but while this definitely helps pick out the details and internal corners in exterior renders, with interior renders it affects the lighting of the whole scene too much to be really useful.

Just an idea- could you run a test render of that file with the following settings:

IM Primary, Multiplier 1
LC Secondary, Multiplier 0.1

IM

-3 Min 0 Max
50 HSph. Subdivs Samples 40

LC
However many subdivs are appropriate for the size of your render.

I'm not sure you'll like the result, but I'm curious to hear your feedback about this setup.

Jackson

p.s. I know we've got a nice BIG forum now, but any chance you could crop your images? They still don't fit the forum so the text goes off the screen and scrolling down to the bottom to reach the sideways scroll bar to scroll across, then scrolling back up to read the end of your first sentences is horrible!

p.p.s. Forgot to say I hope you had a great time in London and probably best not to bump the SCF thread now, but I can't believe you managed to get so many missions done! Glad you enjoyed Sir John Soane's House- my tutors dragged me there, me thinking "What does a young modernist architecture student like me need to see a dusty old pile like this for?" and they had to drag me back out again! Since then I have visited it every time I'm in London for more than a day. The way that natural light is used not in an obvious "more is more" way, but as a means of highlighting features and the character of spaces and how each storey/floor interacts with those above and below via voids and gratings (before laminated glass floors were even conceived) was mind-blowing for me as a student..... and still is.

[thomthom]

Re: Weirdness with DMC as primary GI

I'll resize the images later when I get time. (Hopefully they'll implement my CSS example I gave them that will limit text's line length to 60-70 character but still allow images to overflow and extend without causing them annoying scrollbars.)



I resided to try out DMC instead of IR because I had to up the IR settings so high they where beginning to take quite a bit of time to render. The DMC could retain the details in less time. I also saw this confirmed in some tutorial or manual.

If you look at this image; http://picasaweb.google.no/thomassen...82038155591842 you can see the splotchyness around the band which indicate floor separation. (Not sure what the architechtural name for them is.)

Going from 50-75 HSpSubdivs didn't do a whole lot; http://picasaweb.google.no/thomassen...82051345909778

Here with just the samples visible: http://picasaweb.google.no/thomassen...82058114461106

I tried lowering the Normal Threshold, also without too much luck; http://picasaweb.google.no/thomassen...82074437613538

When I set the Max Rate to 1 I was starting to get clean details. But at a much higher render time: http://picasaweb.google.no/thomassen...82082085843874

I also reduced the Sample Rate, as higher values blur more details: http://picasaweb.google.no/thomassen...82087370263234



In regard to Ambient Occlusion; I wish we'd have that feature as well. I read in the Vray docs that turning on IR's Detail Enhancement is somewhat the same effect;
Before: http://picasaweb.google.no/thomassen...82104163842338
After: http://picasaweb.google.no/thomassen...82109911108498
But it's very noisy.


Now, the first test with DMC, details are about as sharp, though render is darker. But notice that it took 2.19s as oppose to around 10m for the IR; http://picasaweb.google.no/thomassen...82119616745746


Off course, DMC comes with a lot of noise. So the settings needs to be higher;
http://picasaweb.google.no/thomassen...82165496523090
http://picasaweb.google.no/thomassen...82199539967090

http://picasaweb.google.no/thomassen...82213573487698 (this one is now 15m, slower than the IR, but very sharp details.)
Also, DMC is better for lots of glossy ray from what I understand. And it can be quicker using IR.



Whether or not I'll end up using IR or DMC, I'd like to know why I get such a weird colour result with DMC. I've avoided DMC before as I found it to render differently, but now I want to learn what is going on and how to use it correctly.



I'll try your settings when I get home.
I see you got a very low multiplier for LC, I only reduced it to 0.8 in my tests. Why so low?



re: p.s.s. The missions was lots of fun. I'll be sure to do it again next time I go for a trip anywhere. I still which I'd managed to do more. And I'd like to spend more time at the Soane museum.


Thanks for your feedback!

[Micha_cg]

Re: Weirdness with DMC as primary GI

I use LC+IM all days too, becaue DMC is to slow for my need (also per 5xQ6600@3GHz). DE is a nice option to get fine detailes like thin gaps, but but most I try to use a finer IM max rate (default -1, for more details 0) because the better speed and DE cause noise some times. My experience is, that IM is the fastest way to get a clean image done and I'm happy that the IM calculation is so stable. Exteriors should be rendered super fast and your floor problem looks like a geometry problem. I would try to use IM subdiv 50 + samples 20 and small LC sample size like 0.005 or 0.003 (the IM based on the LC pass, so fine details should be there too).

"floor separation" - could it be that two surfaces match the same place, wall and floor? If yes, than try to avoid it, because it cause artefacts and slower rendering. If you disable the secondary ray bias (set 0), than all doubled geometry should renderd black and so you can find it.

[thomthom]

Re: Weirdness with DMC as primary GI

I used IR+DMC for my IR based tests, as that's the default setting where I started from.


There are no floors, I was just referring to the bands that runs around the façade of the building located about where the floor would be. The bands where made using Profile Builder. PB uses the Follow Me tool to generate the geometry; so I expect that there's some faces that overlap where the backside of the bands faces the wall. I'll have a look at this as well when I get home.



But no clue to the big colour difference?

[Jackson_cg]

Re: Weirdness with DMC as primary GI

LOL, I must admit that at that resolution I'm having difficulty seeing much difference in quality between many of those renders! The splotchyness is obvious in the first one- occasionally I go up to 100 HSph. Subdivs and 60 Samples and that usually seems to get rid of splotchyness, but I'm not as "particular" (i.e. meticulous) as you.

Re: the low multiplier for LC secondary- it simply creates much darker interiors, darker internal corners and picks out external details better in exterior renders. You can't do it if you need to have quite bright views into rooms from the outside, but in reality a camera set up for exterior architecture photography would usually produce very dark interiors anyway so I often keep it quite low.

As with most of my VRay theories, there's no real science (or understanding haha) behind it, just that if secondary bounces are set to contribute less to the scene, areas which are mostly lit by bounced light will be darker and if that helps it look more the way I want then that's what I do.

OK, I'm off to play with Detail Enhancement- I've ignored it until now!

[Micha_cg]

Re: Weirdness with DMC as primary GI

My experience of lowering the LC multiplier is, that it cause that white objects are looking greyish, because the light transport of a grey object is simulated. A multiplier 0.1 means 10% of the light is indirect reflected - it's the same like it can be expected by a dark grey color. So, I try to keep the multiplier at 0.6 for exteriors and 0.8 for interiors. But at the end the look of the renderded image counts.

[dalomar_cg]

Re: Weirdness with DMC as primary GI

Okay, where the heck do I start with this one. First off, I don't know why DMC would look that much different...Its almost like its double gamma correcting or something. Second, AO is coming...thats 100%. Thirdly, imho DMC+DMC are really the only way to get the super sharp details that you're trying to get from this render. The reason being that IR and LC are, by their design, going to smooth over these details. Therefore, in order to try and "get them back" you're going to have to fight them and bend them to work against their nature. Not only does this typically require a lot of tests, but in the end also takes a whole lot of time. I'm not saying DMC is fast, but just that its the right tool for the job (this may be one of the few situations where that is actually the case).

About lowering the multiplier for LC. Whether you knew it or not, no matter how you're using light cache, it ALWAYS calculates ALL bounces. Primaries and Secondaries, all the time, and with theoritically infinite number of bounces. So if you're using LC for just secondaries (which is really the only recommended usage), then you're actual having the primary bounces being taken into account twice. By decreasing the secondary bounce multiplier, this becomes a way to partially manage this.

I guess I'll explain a little bit about DE as well. DE is actually just the DMC process that is tied into Irradiance mapping. When there is a IR sample and within the distance specified there is some sort of geometric detail, V-Ray "leaves" that calculation for later and doesn't take that sample into account. During the actual render pass, V-Ray remembers that there are these "gaps" in the IR calculation, and uses DMC to solve them instead. So, effectively, you can use the radius value as a mixer between IR and DMC. Of course, if you're getting noisy results from a DMC calculation, more subdivisions are needed.

I'm sure I missed something as there's a lot in this thread already, so point me to it if that's the case.

JAMIE01 - your question has been separated out into its own thread called "Components working with V-Ray"
Thomas - your post was removed because in a split thread there's no need to talk about spliting the thread

[thomthom]

Re: Weirdness with DMC as primary GI

First off, I don't know why DMC would look that much different...Its almost like its double gamma correcting or something.

What would help in order to determine what's going on? I've used some Arroway textures, so the whole model it quite big I assume; not that convenient to upload. Especially if you include the HDRI. But if the model with just the low-res diffuse textures are ok, I can upload it some place.
It'd be nice to know if it's my settings or if I've come across a bog of some sort.

Whether you knew it or not, no matter how you're using light cache, it ALWAYS calculates ALL bounces.

I heard this. But I've never heard of anyone setting the LC multiplier as low as what Jackson did.

[Micha_cg]

Re: Weirdness with DMC as primary GI

The gaps in the IM pass are only a "bridge for the mind" - a little test with DE radius 500 show me no faster IM pass with calculation gaps. So far I understand it, the IM pass is calculated like without DE, but DE add details at the final pass at areas with geometric details in the search radius.

[dalomar_cg]

Re: Weirdness with DMC as primary GI

This is not what I have understood from Vlado. He has said that IR maps with DE are calculated differently that w/o and has described the result as having "holes". He has said that maps calculated with DE are not valid for use without DE because of these "holes". I don't believe that DE is meant to speed up the IR pass, since V-Ray will still have to calculate those subdivisions to see how big and where the "holes" are.

[thomthom]

Re: Weirdness with DMC as primary GI

I haven't gotten the impression that DE is any speed boosts either. I read that it added some brute force calculation to areas within the threshold set. I assumed that the brute force calculation would be a DMC pass; the noise from the DE looks like DMC as well.

[dalomar_cg]

Re: Weirdness with DMC as primary GI

The intent for DE is not to make anything faster, but to combine that advantages of IR with DMC.

I read that it added some brute force calculation to areas within the threshold set. I assumed that the brute force calculation would be a DMC pass; the noise from the DE looks like DMC as well.

Could you have possibly read that from my post above??? just wondering

I guess I'll explain a little bit about DE as well. DE is actually just the DMC process that is tied into Irradiance mapping.

[thomthom]

Re: Weirdness with DMC as primary GI

Partly from your post, partly from the V-Ray manual;

The detail enhancement option is a way to calculate those smaller details with a high-precision brute-force sampling method. This is similar to how an ambient occlusion pass works, but is more precise as it takes into account bounced light.