Gisd issue?

Started by j meyer, July 18, 2015, 02:28:07 PM

Previous topic - Next topic

j meyer

Hi,
ran into something lately,which,from my point of view at least,may well be
a little fault.
Look here,in particular where the arms of the cross meet the surface of the block.
[attachimg=1]
The higher the sun the less visible the effect.

With the sun on the opposite side something strange shows up,too.
[attachimg=2]
Varies with sun elevation also.

When the render has stopped and the 'after-render-calculations' start anything
looks as it should,but on finishing the render it pops up.
[attachimg=3]

This effect shows up with different models and setups.Tested with separate
pieces of geometry penetrating each other,with separate pieces on top of
each other and one piece geometry,but alas...

Kadri


I think the Occlusion weight is high for imported objects(maybe in general?).
Can you try 0.1-0.25 and see if the spilling gets lower too ?

Dune

You mean the lighter areas? That is due to the 'bounce to the ounce'. Especially where two light areas meet it's quite obvious.

j meyer

Kadri - Even with Occlusion weight at 0.1 or lower it's there.
Bounce to the ounce is of more influence.
Tested before my first post btw.

Ulco - yep,that's what I meant.On some occasions it looks like there would
be an additional bump there.And that is definitely not like it should be.
At least from my point of view.Like so it can be a 'showstopper'.

Dune

You could soften it more, but that takes more render time. Or decrease its effect. I agree it looks a bit too much in your sample.

Kadri


I am not much into the technicalities in rendering.
I might be wrong but it looks like the way GISD works does have problems like this in general.

Maybe Matt used those default numbers more according to landscapes then objects.
Because just by using smaller numbers in all-any of the GISD inputs the appearance changes more as i think it should look for objects.

Using a different rendering method (i have no clue :) ) or having a separate option to use different GISD inputs for object and landscapes might be a help maybe.



Oshyan

GISD has limitations in what it can best handle, certainly. Adjustments to settings can improve the effect though. In this case I'm wondering how high the Bounce to the Ounce setting was. I do not think Occlusion Weight would override it, so to speak, so even if you had high settings for both, to fix this you'd want to reduce Bounce to the Ounce. I do see the potential value of being able to specify GISD weights differently for imported objects, but I'm not sure how difficult it would be from a development perspective. We'll keep it in mind...

- Oshyan

j meyer

Default settings were used for the examples.
Lowering Bounce to the ounce is not always an option.
Anyway,thanks.

Oshyan

Interesting that those are default settings! Might be a bug then in that case. I'll check with Matt.

- Oshyan

Matt

For flat/smooth faces meeting each other at right angles, or similar, the default GISD settings (occlusion weight and bounce to the ounce) are slightly too high to be realistic. However, in many other cases the default settings are too low to be very realistic. The defaults are slightly too low for most vegetation, and much too low for spheres making contact with the ground (this is a case that GISD really struggles with, no matter how strong the weights are).

The defaults are a compromise between the high weights needed for some scenes and the low weights needed for others.

It might look more natural if you increase the radius. Increasing GI cache detail also helps, as it produces a more realistic occlusion and bounce at smaller scales that will blend with the GISD results more naturally.

Matt
Just because milk is white doesn't mean that clouds are made of milk.

j meyer

Noticed the 'sphere making contact with the ground' phenomenon before,
with tires also.Most of the time these spots are not the or within the
focus of the viewers attention,though.But here it was.On first sight I
actually thought the geometry of the model got accidentally messed up
somehow.Fortunately it was just a test model.
I'll give radius and GI cache detail a try tonight.

Thanks Matt,J.

TheBadger

Hi guys,
I just ran into the same thing. I can tell you that it happened to me do to normals. When I averaged normals in Maya one to many times on one part of my temple model, I got the same effect as shown in the OP example. Hard to imagine that there are more than one cause for this, anyway, it is repeatable for me at least. So normals has something to do with it from my experience tonight.

IT happened all over the model, but to greater and lessor degrees depending on the part of the object, and how many times I averaged. Actually, I did not mean to average more than once but did it anyway.
It has been eaten.

j meyer

Tried with increasing the radius,but that exacerbates the effect.
Gi cache detail has no visible influence on the effect at all.
Here are some of the results for the ones interested.
[attach=1]



Michael did you try with your model before your treatment,too?
Because the effect should be visible there as well.It shows up
on imported models,procedural models (TG cube etc) and terrain.
You just need the 'right' lighting conditions and angles between
planes.

Dune

The last quarter looks very natural.

j meyer

Apart from the bottom end of the vertical column.