Marble Hill

Started by fleetwood, April 16, 2014, 09:41:48 PM

Previous topic - Next topic

fleetwood

Kadri : I'm using only a power fractal for plant distribution. But this test shows the banding is not related to any populations.

I  disabled all plants and rendered only the ground in this test. As I watched the render Banding definitely appears just during the last seconds of the render during the application of the GI surface pass. On the left is the render when it finishes the last bucket and the timer stops counting. On the right is 20 seconds later when the last GI is applied.

zaxxon : Thanks, I'll experiment with the darkness of the shadows. I think you're right. It's okay if they are dark but ideally some detail should be there if you try to see deeper. 



yossam

I had this same issue with one of my renders. It is almost like a grid to my eyes (my eyes are old). Mine was in some fake stone shaders with image maps applied. Thought I was seeing things at first.  :o

Dune

It's in the terrain for sure, and the plants follow those lines, kind of, and it's exagerated by GISD. Maybe due to a fractal warp shader that causes a kind of erosionlike bands? Or one insignificant seeming PF set with stretched Y?

Kadri

#18

I was just posting something like Ulco said.
GISD sometimes exaggerates parts that are in shadows -cracks,edges- more then i would like to.
I would go closer to the ground and to that part to see what is going on and change the sun position too maybe.

But it could still be also related to the render engine picking more then one condition and exaggerating all of it.
Curious what is going on.
Have you tried it with a different sun position?

fleetwood

Thanks for the suggestions - this is starting to feel like a "bad seed" kind of problem.
I'm doing many, many tests now, changing one variable at a time. Have taken out all plants, all fake stones, and taken out all surface group shaders except the base color. This leaves only the terrain block . It uses only the default nodes, but with another seed and a smaller displacement value and little different simple shape .

Even with only the basic terrain enabled, the banding still appears at GI phase. I've run about 30 different variables so far.... like sun position, ambient occlusion instead of global illumination, GI levels, gamma, fall off, occ weight, ray detail, camera lens fov, prepass padding, etc.,etc.

Will report my results when I come to some useful conclusions.

Kadri


Interesting. Have you tried it with GISD off too?

fleetwood

Oh yes, GISD off entirely is the easiest work around. With GISD off there is simply no banding. I could render with GISD off and then use that to comp in the shadow areas. Or better, I can use GISD and do the render using Narrow Cubic which does not produce the bands.

Some preliminary findings -
1. A very small compute terrain patch size (0.01) can minimize the appearance of the bands. Tends to make them resemble cracks.
2. A very small Radius in GISD such as 3 or 4 pixels reduces the bands but this is in my opinion equivalent to just slowly turning off GISD. I assume a radius of 0 would mean effectively no GISD.

3. Using all setting the same except the pixel filter method :

The Catmull-Rom sharp pixel filter has a banding problem                                         
The Mitchell Netrivali      pixel filter has a banding problem but bands will be in new spots
The Narrow Cubic          pixel filter has little or no banding
The Cubic B-spline soft   pixel filter has minimal bands

These images have the fake stones tuned back on and all used the same very small gradient patch size so the banding is already quite minimized.


choronr

Great presentation. My taste says Narrow Cubic looks the best.

Kadri

#23
Interesting that the pixel filters changes the appearance in that way.
Maybe time for Planetside to say something.
Maybe there is something we don't see.
Not sure of course but curious if this belongs kinda a little in the "Bug" area ?

Have you tried "Microvertex jittering" , "Detail jittering" and "Do ray traced shadows" on and off Fleetwood ?
What kind of terrain is it? Standard powerfractal on a planet or a hightfield?

fleetwood

Kadri :  Yes I tried both Microvertex jitter and Detail jittering off with no effect.
I tried turning ray traced shadows off, but it removes almost all shadow so there is not enough shadow area to be a test. It becomes a radically different scene.

The terrain is very much the default scene terrain (just those same four nodes) except the fractal terrain power fractal substitutes a different seed and the fractal warp uses a scale of 500 instead of 1000.
The simple shader is the same size and shape as default but I turned on displacement for it and gave the displacement a smooth step edge profile and a big width. That's it.

Kadri


I see.
Not much left to try i can think of right now.
Do you still try different things?

fleetwood

Thanks for your suggestions.
I'm still curious to know if the problem is the in terrain fractal itself or if maybe I have corrupted my TGD file. I may try a couple more tests, maybe rebuild the same terrain in a brand new file, not just copy the old file over. But for now planning to tweak a few things and re-render with Narrow Cubic.

Kadri


I am curious too.
Actually if you could share only the terrain i would like to test it too.
If possible of course. If you don't want no problem of course ;)

yossam

Sorry to interrupt.............is the pic attached basically the same thing? It is the default scene with 4 surface layers............that's all.

Dune

That also looks like a problem posted a while ago by (forgot the name... Yesmine, maybe) where banding occurred in grass pops. Did you use the GISD in that last file Richard? And what filter? I looks like the same issue.