Displacement mapping

My housemate just did something very similar to this while modelling a coffee bean (for an explanatory model of a coffee cherry), using 3DS Max. It took me a few minutes to knock this up in 3DS Max myself:

It would be interesting to see how this software could make this process more accessible, I do have a few questions.

How does it unwrap the object and map the displacement map to it?
How does it cope with wrapping the map around an object and making it tile/loop? Rather than the relatively flat object in your example or mine,  will it produce a seem where the map starts and ends when it loops around an object? Or does that depend on the input map itself, I noticed that your example displacement map is tileable.

Is it smart enough to know how to automatically tile the map, or do you have to tweak that yourself?

Have you looked at displacement mapping techniques used by games and films? As it has been used there for numerous years.

Sorry for the delay... Our code doesn't do mapping, only the conversion to a triangle mesh. The image can be tiled or mirrored and the texture coordinates can be scaled, translated and rotated. Seams in the mapping result in a stair step, unless the image happens to match over the seam.

At the moment, the algorithm has been implemented as standalone software, but a plug-in is one of the development possibilities.

The original use of the code was for texturing plastic injection molds with a laser engraving tool. The precision requirements were considerable for very fine textures, like artificial leather, and we ran into serious problems with using the existing tool base, including 3ds Max, which produced a prohibitive amount of triangles. (This was with the 2010 version.)

Caswal, you seem to have applied an optimization step after the displacement. What was the original polygon count? The picture in my original message is a direct output of the triangulation method and involves no post-processing steps for optimization.

The code is also able to directly produce an STL of the difference between the displaced and the original surface, which is quite useful for traditional subtractive manufacturing. This would be a very time-consuming process as a post-processing step. We are planning to try it out for adding details with a 3d printer on the surface of an existing object.

Games and films nowadays almost exclusively use per-pixel displacement mapping, which is not directly applicable to manufacturing. A slicer that directly interprets the displacement map could be possible, but is probably not worth the effort.

Not since DX11 became available as a standard for games, it is now a combination of tessellation and vertex displacement. DX11 can dynamically subdivide an object so each triangle to will take up a certain area of pixels on the screen. Objects further away have lower amounts of tessellation compared to object much closer to the screen.

Why were the amounts of triangles prohibitive? What sort of counts are we talking about?

While generating meshes with a resolution in the less than 10 micron range, we easily ran into tens of millions of triangles. 3ds Max also tended to skip some of the details, leaving horrible "scars" in the mesh, especially if the base mesh was badly shaped.

The existing tools are all based on subdivision of the mesh. This leads to problems when trying to create sharp features. Pure subdivision-based algorithms have a problem with edges that cut across a corner in the shape, creating corners that look like they have "bites" taken out. To address this, one needs to make the mesh fine enough that the "bites" are within the required tolerance.

Maya includes a code (called Feature-Based Displacement) that partially addresses this by adjusting the locations of the created vertices after subdivision, but it is only used for very sharp features. Maya also has problems with uneven adaptation and lacks direct control of the spatial tolerance.

Subdivision-based codes seem to be very sensitive to the quality of the underlying mesh, which is a problem when the starting point is something like an STL triangulation. STLs need to be processed anyway in order to achieve a good unfolding, but still there is usually an abundance of very narrow triangles.

Our algorithm is not based on subdivision at all. Instead it uses direct insertion of new vertices at points of interest, based on a greedy algorithm that chooses the vertex locations by eliminating the largest deviations first. The result is that the inserted vertices very nicely follow the features of the displacement map and effectively eliminates all edges that cut across the displacement features. If the displacement map contains flat regions, effectively no vertices at all are inserted into those regions.

Even with highly optimized meshes some of our path generations for laser engraving took hours. I don't think that the resolutions with extrusion-based 3d printing is yet at a level where the resolution requirements would be high enough to cause significant problems.

The attachment includes an extreme example of a 3d model that was first processed into a depth map in ZBrush and then displaced onto a rough mesh patch. The original triangulation of the model is quite well preserved. A tolerance setting level below the actual spatial resolution of the image results in an accurate reproduction of the aliasing on the vertical features.

I guess there are quite a lot of tessellation-based engines still in use for rendering, but the future is in pixel shader-based GPU implementations, such as parallax mapping or relief mapping. These techniques completely remove the need for any mesh subdivision for rendering even the most complex displacement maps.

wow this is some good work !