#1465 MultiLinearGeometry: evaluation of Jacobian at tip of pyramid
Metadata
| Property | Value |
|---|---|
| Reported by | Ansgar Burchardt (burchardt@igpm.rwth-aachen.de) |
| Reported at | May 14, 2014 17:51 |
| Type | Bug Report |
| Version | Git (pre2.4) [autotools] |
| Operating System | Unspecified / All |
| Last edited by | Oliver Sander (oliver.sander@tu-dresden.de) |
| Last edited at | May 16, 2014 19:38 |
| Closed by | Oliver Sander (oliver.sander@tu-dresden.de) |
| Closed at | May 16, 2014 19:38 |
| Closed in version | Unknown |
| Resolution | Fixed |
| Comment | Patches applied in |
| c61bb04523aa7f85ebd2ff3f8f6d58111768f5e5 | |
| aae38c56f11e91176a13f71cf1a53d483cf0f84e | |
| 3f69645cb8a5cf600a15e22da8084badb9252f5a | |
| 7ade11ecbfbbe327b5ef647f10e2f5c8007b38ab | |
| 00add6759466a04c49e6472ae87f44e1b4dd8799 |
Thanks! |
Description
MultiLinearGeometry cannot evaluate the jacobian of pyramids at the tip: at some place the code evaluates 0/0, resulting in a NaN.
While the jacobian is in general not even defined in the tip, it is defined for affine-linear transformations of the base (which results in the transformation of the pyramid to be affine-linear as well). By replacing the 0/0 with 0/1, one can get the right result for affine-linear transformation, see the attached patch.
However when the base transformation is non-linear, this code will still return a directional derivative. This might make it easier to find buggy code doing this.
Note that use of CachedMultiLinearGeometry will hide the bug: for affine transformations, CachcedMLG will not evaluate the jacobian in the tip of pyramids.
I would also like to add a few more tests, but didn't get to do so today.
Any comments?
Ansgar