min / max rate

[Thomas An._cg]

What is the significance of the min/max rate concept ? (any links explaining the theory behind this ?) ... why do their values have to be appart ? and how far appart is far enough ?

[flipside]

min / max rate

Here's some usefull info I once wrote for vray for max:
http://www.vray.info/topics/t0101.asp

IR map does nothing more than placing sample points on your surfaces to calculate the lighting in these points. Of course it is crucial where these points are placed, and how well lighting is calculated in these points. On large, flat surfaces you will need less points than in highly detailed areas of your model. On the large surface for example, lighting will not change that much over a given distance. So vray takes a few samples, calculates lighting for them, and then for all points in between the samples it interpolates these lighting values. Same for detailed areas, but more samples will be needed to accurately represent the correct lighting.

Min/max is a way of undersampling, first pass is a rough calculation and further passes will refine the calculations. (see above link, the min/max has to do with resolution etc...) Zero means the ir map is calculated on the resolution of the actual image, -1 is half the resolution, -2 is half the resolution of the -1 pass and so on. A good rule for decent quality GI at 640*480 images is a min/max of -3/0 or for extreme high quality -2/1.

Where the points are placed is decided by the clr/nrm/dist tresholds. These are tresholds to compare surrounding samples with each other, to see if additional samples are needed or not. Clr is the most important one, it compares the color of two samples. So if the color change between two samples is very low, this means that these two samples have very similar lighting, so no additional samples in between these two are necessary. If the clr change is large, it is best to take a new sample in between the two existing ones to see how the lighting is there. The lower the clr treshold, the more detailed your GI will be (=longer rendertime). A good starting point is usually 0.35. Lower than 0.3 will increase rendertime a lot. The relationship is not at all linear, so going from 0.3 to 0.2 can give you way higher change in rendertimes than for example going from 0.4 to 0.3. This is a very sensitive setting, so take care.

Nrm: check the vrayformax manual for more info. This settings usually has not such a big impact as the clr or dist. http://www.spot3d.com/vray/help/VRayHelp150beta/

Dist: Higher values will take more samples in regions where objects are close to each other, resulting in more detail and longer rendertimes.

Hsph subdivs is the quality of the calculations in each of the chosen points. Higher values means more accurate. Note that for example 50 actually means 50*50, so take care not to change too much at once.

When all points are placed and calculated, then the interpolation between these points kicks in. The 'samples' setting right blow the hsph subdivs setting is very important here. This option should actually be called 'interp. samples' as it does in vrayformax. Small values will result in not much interpolation, large values result in high interpolation. This means that high values will tend to blur out the GI solution, while low values will not. So high values will blur out small GI detail, but it will smooth out the whole solution, so at first it may look cleaner. Low interp values will keep the detail but you will see the blotches more.

Note that when using lightcache for secondary bounce, this can greatly speed up IR map calculations. The LC is done first, and the IR map uses the LC information to speed up its own calculations. Especially in situations where there is much secondary bounced light, you will find that IR+LC is the preferred combination. (for example interior renders).

Ok I can go on for hours, maybe it is best to read the vray manual and do some test to see what each setting does. The basics of IR map are extremely important, as you can change 2 small values resulting in a render taking forever without any actual quality difference.

[Thomas An._cg]

min / max rate

Wouter,

Firstly, I would like to thank for your tutorial efforts ! They have been helping me a lot during this early stage and I have print outs of them that I carry around to read. I like your style (especially how you put context in the numerical values indicating how much is too much)

Here's some usefull info I once wrote for vray for max:
http://www.vray.info/topics/t0101.asp

Oh, thanks that was a good insight. It helped clarify the mechanism behind those values. However, something I am still not very clear about this resolution independent aspect ...

Say we have a microscopic image at 64x32 and we set [min,max]=[-3,0], then prepass01 = 8x4 (64/(2^3) x 32/(2^3)), prepass02 = 16x8, prepass03 = 32x16 and prepass04=64x32

Now if the image is 1024x512 we should set the [min,max]=[-7,-4], so prepass01 = 8x4 (1024/(2^4) x 512/(2^4)), prepass02 = 16x8, prepass03 = 32x16 and prepass04=64x32.

In both cases the final prepass will be 64x32 which means the irradiance map has the same amount of information. Proportionally speaking, isn't that a bad thing ? (In the second case we have a large image with a small irradiance map. So the ratio of (imageResolution) / (Irradiance map size) is different).

Min/max is a way of undersampling, first pass is a rough calculation and further passes will refine the calculations.

Ok, here is another question. Why do the calculations need to be refined in a cascade fashion like this ? Can we perform a single pass and make it equivalent to three pre-passes ?

A good starting point is usually 0.35. Lower than 0.3 will increase rendertime a lot. The relationship is not at all linear, so going from 0.3 to 0.2 can give you way higher change in rendertimes than for example going from 0.4 to 0.3. This is a very sensitive setting, so take care.

This puts the ranges of these values in perspective. Thank you.
(I didn't know how much was too much)

-Thomas

[craiggorton]

min / max rate

Learning a lot from this thread- thanks.

But what's the reason for having resolution-dependent settings like that (min/max)?

In normal practice, I will use 320x240 to get the lighting correct, then when satisifed I want to adjust the resolution and hit render. I don't want to tweak settings and hope I got them right. Maybe there's some reason for this, but I would think it would be easier to do this math internally and let me go along on my happy, ignorant way.

Craig

[flipside]

min / max rate

Wouter,
In both cases the final prepass will be 64x32 which means the irradiance map has the same amount of information. Proportionally speaking, isn't that a bad thing ? (In the second case we have a large image with a small irradiance map. So the ratio of (imageResolution) / (Irradiance map size) is different).

That ratio is indeed different. Usually you would test the GI quality on a 640 or 800px wide image. If the GI quality is satisfying there with for example -3/0 on 800px, and you apply that rule for a 3200px wide image and thus use -5/-2, the GI quality will be the same (of course small errors in the 800px wide will now look a bit bigger!). But that is the same with everything else, small modeling errors will also be more visual in the 3200px wide image.

In vray for max there is an option to turn off the passes and use a single pass. But this is slower, since with the passes vray can better identify regions that need low sampling vs regions that need higher sampling.

Craig, that is indeed the drawback of a biased renderer. But once you get the hang of the settings, you can do as you want. For example for product shots (object on groundplane with skylight and some area lights), I simply know that -3/0 with hsph=50, clr=0.3=nrm=dist will be high quality on a 800px wide image. So if I wanna render 1600px, I just lower min and max by one and I still know quality will be OK. Of course for anims some tweaking can greatly improve rendertimes.

Important note, if you use direct light, area lights (without the store with IR map option) etc... In other words, 'direct' lighting, you can get away with lower ir map settings, because most of the lighting detail (=shadows) will be computed by these lights. The rays coming from a skylight are already first bounce GI light, thus calculated with IR map. The rays of a normal light are direct. So if a skylight hits an object, the light that bounces off is already second bounce, while with a direct light this will only be first bounce light. This means that by using direct lights, the total amount of first and secondary bounces will greatly decrease. So if you use lower settings for these, you won't notice it that much since there is not much bounced light. This means you can get lower GI calculation times. But, if you use area lights, the rendering itsself will take longer because there are area shadows to be computed

[Thomas An._cg]

min / max rate

In vray for max there is an option to turn off the passes and use a single pass. But this is slower, since with the passes vray can better identify regions that need low sampling vs regions that need higher sampling.

I see. We could make a single prepass by setting min=max...
... but you are saying a single prepass would be slower than all 4 prepasses put together (for the same quality) ? Is that it ?
Is 4 prepasses the golden rule for speed an quality ? (how about only 2 or 3 ?)

So if a skylight hits an object, the light that bounces off is already second bounce, while with a direct light this will only be first bounce light.

I have been thinking about this lately, but I am not fully clear about it.

Say we store a direct light into the irradiance map. Now this light becomes a second bounce light (becomes part of indirect lighting) ... now here is my comfusion: The primary engine handles primary bounces and the secondary engine handles secondary bounces. So, if the irradiance map is the primary engine and we store the light in the irradiance map, then it will still be a first bounce source. No ? (I mean it will still be handled by the primary / direct engine. No ?) I think I misunderstand the text here: http://www.spot3d.com/vray/help/VRay...gi.htm#bounces

[flipside]

min / max rate

There is no rule for number of passes, too much will be useless because the initial one will be for such a small resolution image that it won't matter. 4passes is usually good, I always use 3 or 4 passes.

Indeed, these four passes will be quicker than one single pass, because of undersampling.

If you use store with IR map, the light doesn't emit secondary bounce it emits first bounce (like skylight light). If it is not stored, the light emits direct light, which is not GI light. Of course when it hits an object, it bounces of and that bounce is first bounce. So if you use store with IR map, it emits first bounce (computed by IR map) and then bounces off and becomes secondary.

Here an example:
Only direct light:



Direct and first bounce (IR map)



Direct and first bounce and store with ir map option:



So in the last example, all the main shadows are calculated by the IR map, because the rect light emits first bounce GI light. I then thought that when this light bounces off a surface it would become secondary light, but I think this is wrong. Because in my assumption, in the last image there wouldn't be any color bleed (since I don't have sondary turned on). So maybe the rect light indeed emits first bounce GI light, but when this light bounces off, it still is treated as first bounce. Maybe Vlado can explain this better, by doing this test I get confused myself.

But anyway, the store with ir map option treats the light as first bounce, so it will be computed with the ir map. So the shadows quality will depend on the ir map settings and will not be raytraced anymore.



edit: I think something is wrong with store with ir map:


Even the shadow under the white sphere look green tinted, that seems weird to me.

[Thomas An._cg]

min / max rate

In terms of Direct light versus GI light...

Is this how it works ?

[vlado]

min / max rate

Even the shadow under the white sphere look green tinted, that seems weird to me.

It doesn't look too weird to me, how does it look without the store option? Anyways, the "store with irradiance map" option simply adds the direct lighting on top of the already calculated GI in the irradiance map.

Best regards,
VLado

[vlado]

min / max rate

In terms of Direct light versus GI light...

Is this how it works ?

Replace the light with "camera" and you'll get it. Bounces are counted from the camera towards the scene, because this is how V-Ray traces them.

Best regards,
Vlado

[Thomas An._cg]

min / max rate

Replace the light with "camera" and you'll get it. Bounces are counted from the camera towards the scene, because this is how V-Ray traces them.

Thanks Vlado,

I did some more homework ... so this has got to be it. No ?


So, my understanding so far:

• Each red dot is a 3d point of the irradiance point cloud.
• Each gray hemisphere is a random group of bounced rays. The number of bounced rays in the hemisphere is controlled by the Hmsr Subdivs parameter. At least one of these rays must hit a light source, otherwise the color of that pixel would be black.
• Also, each red dot is considered a sample and the "interp Samples" parameter controls how many of these samples to blend/blur together so as to show a smoother result.

Am I getting closer ?