Interior render - Brute force vs Irradiance map

[Guest]

I am running a test, the scene is only lit via Vray sun, so there are dark areas. Below is the settings for each render. As you can see the brute force method is very noisy and the irradiance map method is very splotchy.

I consider the settings to be quite high for both methods. Where am I doing wrong? Both methods have produced less than satisfactory results. The brute force method although the best took 7 hours! If I bump up the subdivisions any higher I wont be seeing the final render any time soon.

PC Spec
Intel Core 2 Quad Q6600
4.00 GB RAM

Output size = 1600 x 1200 px

Lighting = Vray sun & Sky
Camera = Vray Physical Camera

Image A – Irradiance map method

Render Time = 1h 19m

Result = Splotchy in areas where there is little light

Adaptive DMC Sampler
Min subdivs = 1
Max Subdivs = 24
Clr thresh = 0.005

Image Sampler
Type = Adaptive DMC
Antialiasing filter = Mitchell-Netravali

Environment = Off

Colour Mapping
Type = Linear Multiply
Gamma = 2.2

Indirect Illumination
Primary Bounces = Irradiance map
Secondary Bounces = Light cache

Irradiance map
Min rate = -3
Max rate = 0
HSph subdivs = 50
Interp. Samples = 20
Clr thresh = 0.3
Nrm thresh = 0.1
Dist thresh = 0.1

Light cache
Subdivs = 1600
Sample size = 0.02
Scale = Screen

DMC Sampler
Adaptive amount = 1.0
Noise threshold = 0.005
Min samples = 8
Global subdivs multiplier = 1.0

Image B – Brute force method

Render Time = 7h 14m

Result = Lots of noise

Adaptive DMC Sampler
Min subdivs = 1
Max Subdivs = 100
Clr thresh = 0.005

Image Sampler
Type = Adaptive DMC
Antialiasing filter = Mitchell-Netravali

Environment = Off

Colour Mapping
Type = Linear Multiply
Gamma = 2.2

Indirect Illumination
Primary Bounces = Brute Force
Secondary Bounces = Light cache

Brute Force
Subdivs = 16

Light cache
Subdivs = 1600
Sample size = 0.02
Scale = Screen

DMC Sampler
Adaptive amount = 1.0
Noise threshold = 0.005
Min samples = 8
Global subdivs multiplier = 1.0

Also, there seems to be a bug, when I set the Adaptive DMC Sampler to a max subdivs higher than the default of 4, save and close the file. When I open the file the subdivs default back to 4. This happens with both of the above methods.

Has anyone experienced this before? All other settings are fine, and remain what I set them to. :/ I have also tried this on various other files and It has the same problem. It leads to think does it render at what I set it to, or does it render at 4?

[fAEkE]

I don't think your results are that strange. Since you are only lighting your scene with a sun and skylight, you need higher settings to produce a better image. The only light in the top left corner are GI, no direct light.

To produce a clean image with brute force you often need very high settings. And the more indirect light (less direct light) you have in your scene, the better settings you need. I also think that you need to study min/max values for the IR map a bit. The "sharp" blotches you get is a combination of relatively low settings for this scene and the use of 0 as max value. One other thing I'm struggling with myself is the adaptive amount. I know that you need to use fully adaptive DMC to get the most out of vray (as far as I know), but I just can't get my GI solution to be th way I want it with 1.0.

Try this please, and see what results you get:

Sun subdiv 25

Adaptive DMC Sampler
Min subdivs = 1
Max Subdivs = 12
Clr thresh = 0.005

Irradiance map
Medium preset
HSph subdivs = 70
Interp. Samples = 30

Light cache
Subdivs = 1600
Sample size = 0.02
Scale = Screen

DMC Sampler
Adaptive amount = 0.85
Noise threshold = 0.007
Min samples = 8
Global subdivs multiplier = 1.0

Notice that these settings are lower than your, but I believe the result will be a bit better. Atleast for the GI solution. And by better I mean "looks better", not "more exact".

Regarding you bug, I've got the almost the same problem with one of my scenes at home. I use Mitchell-Netravali, and every time I open the scene the values are set to 0.0... I have no idea why and I've never seen it before 2.0.

[super gnu]

id be very surprised if you got a clean result with 50 hsph subdivs for imap, this is a -very- low setting for an interior..( those are the imap default settings..) i generally find the imap needs to be anywhere between 100 and 300 subdivs to get a clean interior.

also its not recommended to use imap with an adaptive amount of 1, as it will further reduce your effective subdiv value in low detail areas, often making it impossible to get a blotch free result whatever you set the subdivs to. try reducing the adaptive amount a bit (the 0.85 default works fine with imap for example) , or increase the min samples to say 150 - 200 forcing the imap to use more subdivs as a minimum.

[Subburb]

i agree.. 50 hsph +adaptive at 1.0 will lead to bad result...

my 2 cents :

DMC:
adaptive = 0.65
min samples = at least 128 (this will help to reduce splotches in no time)
global mult = at least 4 (assuming all subdivs values, materials and lights, are default values)
noise thr : 0.003

IM:
As minrate and max rate are resolution dependant, i would suggest -5 -2.
If -4 -1 is correct at 800*600, then you can use -5 -2 at 1600*1200
hsph = 24. (24*global mult = 96 subdivs) (at least)
you can check on DE, with radius = 80, and multiplier = 0.33, wich means 32 subdivs (24hsph*4globmult*0.33 = 32)

LC, sounds correct, but be sure you've prefilter = at least 64

Adaptive sampler, i would suggest 2-6, with clr thr = 0.005. Would be enough to make correct AA. (it can be lowered to 1-5 if you double the res to 3200*2400)
The glossies will be managed by DMC sampler, and not by Adaptive sampler.

I'm pretty sure you'll get better results in less time !

[CA Portugal]

About your DMC bug ive got the same problem too with all files i use in 3dsmax2011 and vray sp2. Bug also happened in max2008.
When i had 2010 installed this bug didnt happen at all.

I think it has to do with max, but im not sure.

[Guest]

"i generally find the imap needs to be anywhere between 100 and 300 subdivs to get a clean interior. " I didn't realise you could go up that high, I always thought 50 was high enough, any higher and it will take an age to render, also start to see strange results.

Thanks for the help.

[fAEkE]

I use 100 max. 50-80 is often enough imo. I would really want to see some tests where the difference between 100 and 300 subdivs shows.

[super gnu]

it all depends on how tough the lighting situation is.. ive had many many scenes that would never end up clean withough subdivs much higher than 100... since the imap is usually quite fast relative to the rest of the render, ive never understood people being scared of turning it up or thinking there should be some kind of maximum.

one job i did (saatchi gallery in london) was all pristine white walls lit by diffused skylight. the client was (as youd expect from saatchi) uber picky about the cleanliness of the white walls, and i think i ended up using 600 hsph for the imap. looked great and didnt take an outrageous amount of time to render. obviously with stills the balance of rendertime is more biased towards the imap, and maybe BF would have been a better choice.. but this was a flythrough.

basically if you have blotches, theres only a few things you can do.. the main one being increase the hsph. subdivs.

[fAEkE]

I would recommend fill light instead of 600 hsph

[super gnu]

lol yes that is another option.. however i was trying to avoid such cheap dirty hacks the rendertime wasnt an issue, and to be honest adding a fill light with a soft shadow big enough (they were very picky about how the picture frames cast shadows on the walls... like i said.. reaaaaly picky) would have bumped my frame times though the roof... when rendering flythroughs i prefer to put all the heavy calculating into the imap, since its done in a few hours, and then with no area shads or complex stuff to do, the -3000 frame single camera path- (was a very boring anim) rendered in a reasonable time.. about 5-10 mins a frame iirc.

[Guest]

Hi,

I decided to render through the stages and post up the results. You can see the settings for each image in the PDF file here.

Comments on lighting? Models and materials are evermotion stock.

[Guest]

Here is version 5 with some post work.

[fAEkE]

I would recommend not changing the global multiplier setting.

What I wonder tho, is why the top right corner get illuminated like that.. that doesn't look natural. Can you render it with Bruteforce/LC and see if it changes?