Displacement has random breakups

[lacilaci]

When mesh is displaced, it shows at random points some break ups, even when it's not at UV seam.

can something be done about it?

here's the testscene with object and disp. map

Downloads.zip

(1.58 MiB) Downloaded 174 times

[lacilaci]

Playing around some more with luxcore displacement and it's really hard to make it work

here is the problem

So a simple test on a tire. A 6K displacement map and 33M triangles using ~50% of Vram and yet I still see some bad jagged edges on the tire.

few problems I have with disp. in luxcore:

I don't know how displacement offset works. When I used 16 bit disp. map and wanted to use gray color as a base surface level (white and black goind outwards and inward the mesh accordingly) I had to use a value of 0.002 for offset to make it work. I understand that some color management and whatnot comes to play here but the parameter is just not very intuitive to set

Mas edge screen size parameter is not making any sense to me, this should be an absolute value in pixels imho, just like other renderers do.

subdivision should be adaptive (using previously set max edge length in pixels). Right now you alwasy start with evenly subdividing the complete mesh, which makes it impossible for fine detail to be useful in any way cause those polycounts would go through the roof.

simplify node/shape is useless. It is extremely slow and I don't even know what would I get in the end.

[Dade]

When mesh is displaced, it shows at random points some break ups, even when it's not at UV seam.
Annotation 2020-04-03 111019.jpg

can something be done about it?

here's the testscene with object and disp. map
Downloads.zip

There is something very strange going on in this scene, I have the feeling it may be related to UV. If you move around, the white highlight will do strange things. I'm looking into it.

[Dade]

So a simple test on a tire. A 6K displacement map and 33M triangles using ~50% of Vram and yet I still see some bad jagged edges on the tire.

This is a consequence of both using an image map (i.e.pixels are jagged by definition) and a regular subdivision grid but, as far as I know, it is normal. Does it work in a different way in Blender/Cycles/Corona/whatever ?

[lacilaci]

So a simple test on a tire. A 6K displacement map and 33M triangles using ~50% of Vram and yet I still see some bad jagged edges on the tire.

This is a consequence of both using an image map (i.e.pixels are jagged by definition) and a regular subdivision grid but, as far as I know, it is normal. Does it work in a different way in Blender/Cycles/Corona/whatever ?

Well.. If I could specify a max edgelength of 1 pixel and not worry about the whole mesh being subdivided like crazy, then I could also decide preciselly how big texture is needed for given detail... Right now the control over displacement subdivision is very limited and unintuitive

[Sharlybg]

Really also wonder how it look in Cycles. Can make a cycles test ?

[Dade]

Well.. If I could specify a max edgelength of 1 pixel and not worry about the whole mesh being subdivided like crazy,

But if use an image map as displacement map, it should make no difference, the jags should be still there

I mean, I assume you need a procedural or vectorial description as displacement map or you will see jags even with edge length of 1 pixel (if you zoom enough).

[lacilaci]

Well.. If I could specify a max edgelength of 1 pixel and not worry about the whole mesh being subdivided like crazy,

But if use an image map as displacement map, it should make no difference, the jags should be still there

I mean, I assume you need a procedural or vectorial description as displacement map or you will see jags even with edge length of 1 pixel (if you zoom enough).

No, im talking pixels as screen space. Currently there is screen size parameter but its not in pixels. And it also still divides whole mesh so not adaptive

[lacilaci]

Maybe this kinda tesselation would help?

https://developer.nvidia.com/gpugems/gp ... n-surfaces

[Dade]

No, im talking pixels as screen space. Currently there is screen size parameter but its not in pixels. And it also still divides whole mesh so not adaptive

And I'm talking of Blender/Cycles/Corona/whatever, they will produce the same result. If you use 2x2 pixels displacement map, you will get some really BIG jag, no matter of what you use. And for any displacement map pixel resolution, you can zoom enough a still see the jags (because an image has a finite resolution).

As I wrote, you need a procedural or vectorial description as displacement map or you will see jags, first or later.