[m-Que]

Going back to the OP - wouldn't something like this do the job?

All the credits go to abed-sabeh;
The material is available HERE @resources.Maxwell

[feynman]

Yes, that's the one I started using a while ago, first with the adaptive setting as set in that MXM; but then I changed it with a larger texture because of much larger flat carpeting areas.

[Polyxo]

It is theoretically possible to output the pretess as a mesh, so ultimately its still a mesh.

Just to clarify each increment of subdiv doubles the number of polys.

Thank you for this clarification! I could not believe this, especially as a lot of other renderers subdivide differently.

But it would be a good explanation why Maxwell completely fails to transfer Displacement from Zbrush for instance.
Here if one has a mesh with say 6 SubdLevels, saves the geo with 2subd's and writes a very good displacement-map
from the highest SubD.Level....
The result neither does reflect the shape nor is it any smooth. It's just completely off.
The strange thing is that on the fly doesn't work either. Totally incompatible it seems, which is a pity.

[JDHill]

Just to clarify each increment of subdiv doubles the number of polys.

I am not sure about that; the only way for it to be the case would be for each subdivision level to bisect each tri, which would quickly lead to a very poor mesh. I would guess it to be done more along these lines:

Code: Select all

            /\                     /\                     /\
           /  \                   /  \                   /  \
          /    \                 /    \                 /____\ 
         /      \               /      \               /\    /\
        /        \             /--------\             /  \  /  \
       /          \           / \      / \           /____\/____\
      /            \         /   \    /   \         /\    /\    /\
     /              \       /     \  /     \       /  \  /  \  /  \ 
    /________________\     /_______\/_______\     /____\/____\/____\

[Polyxo]

I am not sure about that; the only way for it to be the case would be for each subdivision level to bisect each tri, which would quickly lead to a very poor mesh.

So would you say that every quad-input gets triangulated right away (before the first subdivision gets applied? (That should lead to a poor mesh too...)

[JDHill]

There is no such thing as a quad in an MXS; if you have quads (or ngons) in your host app, they must be triangulated by your plugin before it can write them into an MXS.

[feynman]

Bingo! Thanks, JDHill. Maxwell Render = triangle mesh.

Btw, I would not have believed the benefit of good old ASCII art in todays computing environment :D

[Polyxo]

There is no such thing as a quad in an MXS; if you have quads (or ngons) in your host app, they must be triangulated by your plugin before it can write them into an MXS.

Well, I know that all mesh-based renderers at some point start working on triangles. But they obviously have the choice when doing so.
That's why I asked.

It makes a just tremendous difference on shape-integrity whether one triangulates a mesh right away (and keeps creating smaller triangles in further subdivisions) or whether one performs quad-based subd's first and triangulates the last step. This at least is true if the source-model is high-res quads, got transferred to a low poly cage and maps and is meant to be replicated by displacement.

Vray at least seems to treat quads as quads first (Catmull Clark) and Ngons and Tris with Loop-subdivision which
seems similar to the one you sketched out. Source

[feynman]

Final findings

[eric nixon]

Do you think its perhaps a possibility that there isnt any pretess-disp in those last two renders? Why not say something like my pretess disp is very small, then we at least get to chuckle, now your gonna say its a bug right?

[feynman]

You can see for yourself here https://www.dropbox.com/sh/xx811gkz4uugzqt/RrFsh8slZE When I select Pretessellated, the displacement always looks flat. Whether it's a bug or a carpet beetle, I don't know...

[Polyxo]

Ok, Eric will probably deliver an elaborate examination.
For starters: One can definedly make your material to work with pretesselated displacement and also make the machine consume ram.
A lot of ram. One can also get it to render quickly.

I could not get the .mxs to work as it always looked for an external .mxm although I had embedded it.

So I only had a look at the mxm. You should have posted this earlier, really...
There's people with far better expertise but one did not even have to look at the delicate stuff to find setup-errors.

The tiling was far too loose: 1.5 on 2*2m. The image detail you get per face at such a scale is quite neglectable,
especially as your source-map is 8bit only. Just scale the the map to 2m in Photoshop and have a look what you get.

On the other hand you want to raise a map from an ordinary flat carpet of maybe 5-6mm hight up to 2.5 cm.
A displacement offset of 0.1 also hardly makes sense.
The base-layer had a ND of 1 (=air), while the ND of the Spec-Layer (probably not even required Layer) with 3 is too high.
Using a Normalmap instead of bump would certainly help and maybe making the map smaller but then tiling more nicely.

The attached Screenshot is from Fire.
That's of course far from any good looking but it at least captures the knot-structure of the original-map somewhat.
To get an outcome of a fat fluffy rug with large knots (which I believe you would like to create) one would have to use
another set of maps which captures large detail in great resolution.

I still have no explanation why your Ram does not increase when rendering, but as Eric I would expect that the displacement
does not get evaluated at all.

[feynman]

A field of beet roots - autumn is neigh :D

I used Pack and Go, but maybe that is not the right way to publish a complete setup for others... Thanks for your hints, but still, using your suggestions, with pretessellated it looks absolutely flat. I tried it on another PC even. I think, I better give up and hope to be able to use 2.7 soon and then do it with grass as suggested. At least, maybe others looking for quick carpeting with displacement will surely find tons of useful information from the other users in this now thickly carpeted thread.

Still can't get over that render with the knotty rug and these tiny hairs protruding ever so realistically...

[Polyxo]

"feynman"]A field of beet roots - autumn is neigh

Garbage maps in, garbage render out...
Under what carpet did you find that image?

Thanks for your hints, but still, using your suggestions, with pretessellated it looks absolutely flat. I tried it on another PC even.

What you rather should try is isolating the .mxm as embedded into another scene.
The carpet .mxm sucks but if I import it into a fresh file and apply it to a plane it definedly displaces the mesh.
Screenshot of a 2m*2m plane attached.
If that also doesn't help and you really don't see this it might make sense that one of the NL_guys had a look via Screensharing.
You have some content-linking going on in your mxs, maybe here is the reason for the issue.

I think, I better give up and hope to be able to use 2.7 soon and then do it with grass as suggested.

Well, that's not a drag and drop beauty-solution too...
Good displacement has a lot to do with good maps with a rich tonal range. You might as well just scan a piece
of carpet, buy a cheap but cool map-creation-tool like Shadermap and start from scratch.

Still can't get over that render with the knotty rug and these tiny hairs protruding ever so realistically]

Rustebergs image?
That was a photo dude! He just wanted to give Eric a hard time.

the unchanged .mxm from your file.

[feynman]

Cheers; what is content-linking? Don't find that in the manual...

Please try the earlier suggested "abed-sabeh carpet MXM" - with on the fly it renders three-dimensional, change it to pretessellated and will render two-dimensional, no matter if in a "fresh" or "old" file. Map quality is a different issue; even a "bad" map should displace in both modes, no?