I thought 1.0 would produce the same results as having detail enhancement turned off. That’s sure what the help file makes it sound.
Can someone explain this better? I can’t seem to understand how the numbers work. I like to know the scale of things before I just go changing numbers around (although, I have done some of that already).
If you mean the radius parameter, then a value of 0.0 would be as though DE is turned off.
Otherwise, if you use “World” scale, the value is a radius in world units. If you shoose the “Screen” scale, the value is pixels (approximately).
Best regards,
Vlado
Not sure what you mean by “1.0 would produce the same results…” but I’ll chime in here until someone smarter does. The numbers relate to hemispheric sampling and how far it looks in a scene. When vray is calculating an irradiance map it fires a ray from the camera to a surface. when it hits the surface it then bounces rays of this surface in a half sphere shape (or hemisphere) and sees what they hit.
This is how vrays hsphere samples work in the irmap. If you turn on detail enhancement, what happens is for the fine detail in a scene that would sometimes be lost by a blurry gi method like irradiance mapping can be detected and then a far more accurate method used to calculate lighting for those details in a way similar to qmc based gi. So what you end up getting is the speed and smoothness of irmaps in the more “boring” areas and the sharpness of qmc gi in the detail areas.
In terms of the radius settings, what this describes is how big an area the hemispheric sampling will cover. Say for example you have an object in your scene and another large object 100 max units away. With the radius setting, you can control how far away from the area the detail enhancement is sampling will take into account with it’s sampling. If we take our example of our object with another object 100 units away, our DE radius would have to be at least 100 to bother taking into account the other object. If we set the radius parameter to 50 for example, the hsphere rays from the detail enhancement will shoot out in a hemisphere but cut off within 50 units distance from the point being sampled and not make it far enough to consider the object 100 units away. It won’t have an effect on the area we are using detail enhancement on.
Effectively what it does is allows us to terminate the detail enhancement early so that we get quicker renders. Admittedly it might be a bit less accurate, but there’s a good possibility that an object really far away won’t have any effect on the lighting of an object anyway. What you get to choose is how accurate versus how fast you want the render to be.
As regards screen versus world space, you might end up with a scene that covers a very small area such as a render of an object or an object that covers a very large area such as a render of a building exterior. Having a similar scale of DE sampling makes perfect sense for a small object so screen space works nicely there since there isn’t much depth. As per the help files, it’s basing the early cut off radius on distance in pixels, so if you have a pixel using de and a screen radius of 20, it’ll consider objects within 20 pixels radius.
If you were doing an exterior however, you might have a tree in the foreground and a building in the far away background. If you keep with screen scale on a scene like this, you might have bits of the background within a 20 pixel radius of the tree - it might be so far away from the tree that it won’t affect its lighting but the DE will try and sample it anyway. You might end up with no benefit but the speed hit of trying to calculate it. For something like this you could use world space. Here the radius value makes a hemisphere around the object it’s rendering in 3d space so again if we had our tree object in the extreme foreground and the building in the background, since it’s using distance in 3d space rather than just pixels, the building will probably be way more than 20 3d units away.
In terms of a value of 1.0 turning off de, in screen space you’re nearly correct since you’re limiting it’s sampling to such a small area. In world space, a distance of 1.0 might actually be useful on a tiny space where 1 unit might make up a large area in your render.
Hope this helps!
Oh and another thing, you’ll possibly find that the bigger your radius, the darker your render gets since the de rays will possibly see more objects that can block light hitting the points being sampled - much the same as the vray dirt map used for ambient occlusion.
Sorry guys, I always omit an important detail in my post.
I’m fairly certain I have a grasp of what the radius does. Thanks for the in depth explanation joconnell. Although, in one paragraph you made it sound like if a ray hits another object within the radius you gave then it actually ignores, instead of uses the multiplier, the sample for DE.
What I actually had a question about was the multiplier.
The help file makes it sound like if I have it set to 1.0 then it should be the same as if doing nothing at all…which would make sense to me. Multiplying something by 1 shouldn’t effect the outcome right? So multiplying something by less than 1 would make detailed areas Less detailed, and a number greater than 1 would make them more detailed. This is what I am understanding with the help file and your explanations.
This understanding I have seems to be incorrect, however, so please correct me. I am getting approximately the same results, although it Is slightly more detailed, with a multiplier of 0.0 and 1.0 increases my time by about 3x. 0.0 increases my time by only a small fraction which I would expect simply because there is one more process to compute.
Thanks again for the help. Very detailed.
The subdivs multiplier controls the number of samples used for calculating the DE as a fraction of the Hemspheric subdivs for the irradiance map. The effect is to make the DE less or more noisy in exchange for calculation time.
So, a value of 1.0 will use the same hemispheric subdivs as specified for the irradiance map. Lower values will use less subdivs making DE faster, although perhaps more noisy.
Best regards,
Vlado
Subdivs multipler is qmc sampling subdivs, which will be used to reveal the details. It depends on hsph value. So subdivs mult. = 1.0 with hsph set to 50 --> means that qmc subdivs amount will be 50.
sub. mult = 0.7 --> 35 and so on
Heh Vlado 
Ah okay getcha now though I see vlado and Paul have both chimed in with smarter answers 
OK, both explanations make sense, but those aren’t results I’m getting, which is why I’m confused.
I have 50 Hemispherical subdivisions and the time on a region is 34 seconds and it is too blurry.
I turned on detail enhancement and it cleans it up adding only 2 seconds. The strange thing is, the multiplier is only 0.3, which should decrease the subdivisions.
If I change it to 1.0 the time goes up to 1:26 and gets even more detailed. Should it look at least approximately the same as not using DE at all?
No, if you set the subdivs multiplier to 1.0, you are simply using more samples for the detail enhancement.
Best regards,
Vlado
Can you explain this further. I’d kind of like to know how the numbers correlate to each other. I mean, what kind of scale does it use then.
Generally this equation is true (1*x)=x
but it sounds more like we’ve got this? (1*x)=(x*y)
there is no direct correlation, per se. You are simply replacing the irrad. map calc with a direct QMC solution depending on the radius value. It is a bit confusing I guess since the subs are calculated as a fraction of the Hemispherical subdivisions, it could as well just be set directly as in the GMC controls.
It sounds like (in case hsph=50):
we’re using 50 subdivs when this set to 1.0
and you can’t set this any higher than 1.0
but you can rise hsph value, for ex hsph=150 (so, 1.0=150, not as before 1.0=50)
this leads to higher quality and higher time
If you want lower quality in lower time:
set the subdivs multipler to smth below 1.0
for ex. 0.7
0.7=(50*0.7)
0.7=35
so now 35 subdivs used for calculation (hsph is still 50 )
Paul, I thought those calculations make sense, but that doesn’t seem to directly correlate. Also, you Can use a number higher than 1 to increase the subdivisions in those areas which of course makes you time go really high.
I think my confusion is that it maybe doesn’t use IR, but instead uses QMC? This would explain why it is slower simply having DE turned on at all and why I have increased detail even though the number is 0.0
Am I correct there?
Sorry for asking for so much explanation. I just would rather know how something works rather than just what numbers seem to work for me. It seems I can problem solve much better that way.
Thanks for being patient.
Don’t trust me huh? 
from the help index:
Detail enhancement is a method for bringing additional detail to the irradiance map in the case where there are small details in the image. Due to its limited resolution, the irradiance map typically blurs the GI in these areas or produces splotchy and flickering results. The detail enhancement option is a way to calculate those smaller details with a high-precision QMC sampling method. This is similar to how an ambient occlusion pass works, but is more precise as it takes into account bounced light.
I trust you, but was slightly confused about the GMC part. Did you mean QMC?
yes, that would be a typo
Oh… really multipler can be higher than 1.0 hahaha
/me got tired of work 
