I just learned about a promising replacement for CGAL booleans in OpenSCAD.
It is the relatively new mesh_boolean module in libigl by Qingnan "James"
Zhou.
Here's the paper:
http://www.cs.columbia.edu/cg/mesh-arrangements/mesh-arrangements-for-solid-geometry-siggraph-2016-zhou-et-al.pdf
Here are the claims:
- Accepts a more liberal set of meshes than CGAL and Carve.
- 6.8 times faster than CGAL (on an 8 core processor)
- (Carve is 7.72 times faster than CGAL in the same tests.)
- Uses CGAL exact arithmetic.
- Multithreaded, and can union a variable length list of shapes together in
parallel.
- More robust than Carve or CGAL. The output is guaranteed to be "without
self-intersections, without open boundaries, and with piecewise-constant
winding number", whereas both Carve and CGAL output fails to be valid in a
significant fraction of cases.
- This guarantee of output correctness applies to the exact arithmetic
version of the output. However, they provide a conversion to floating point
that succeeds 99.5% of the time.
My friend also claims that these operations are "CGAL compatible".
My understanding is that this could replace the multi-threaded CGAL
interface that we have under development, and that it would be more robust,
accepting more meshes as inputs, and succeeding more often in producing
valid STL files as outputs.
I just learned about a promising replacement for CGAL booleans in OpenSCAD.
It is the relatively new mesh_boolean module in libigl by Qingnan "James"
Zhou.
Here's the paper:
http://www.cs.columbia.edu/cg/mesh-arrangements/mesh-arrangements-for-solid-geometry-siggraph-2016-zhou-et-al.pdf
Here are the claims:
* Accepts a more liberal set of meshes than CGAL and Carve.
* 6.8 times faster than CGAL (on an 8 core processor)
* (Carve is 7.72 times faster than CGAL in the same tests.)
* Uses CGAL exact arithmetic.
* Multithreaded, and can union a variable length list of shapes together in
parallel.
* More robust than Carve or CGAL. The output is guaranteed to be "without
self-intersections, without open boundaries, and with piecewise-constant
winding number", whereas both Carve and CGAL output fails to be valid in a
significant fraction of cases.
* This guarantee of output correctness applies to the exact arithmetic
version of the output. However, they provide a conversion to floating point
that succeeds 99.5% of the time.
My friend also claims that these operations are "CGAL compatible".
My understanding is that this could replace the multi-threaded CGAL
interface that we have under development, and that it would be more robust,
accepting more meshes as inputs, and succeeding more often in producing
valid STL files as outputs.
