As I have indicated previously, a change in method beyond a more
robust linear solver is required. I have taken the time to fool with
Armadillo. At the very least I can pass along a trick to improve the
numerical accuracy.. some.
It would help me if you would share the data because I could actua
On 09/09/2011 08:11 AM, Even Rouault wrote:
You might try latest GDAL trunk, in particular
http://trac.osgeo.org/gdal/changeset/22876 . It offers the option to
use libarmadillo to speed-up matrix inversion in thinplatespline.cpp
(speed-up when TPS are in the thousands), but perhaps as a side
Le vendredi 09 septembre 2011 03:23:29, Big Bear a écrit :
> The linear solver in the TPS routine is naive for any number of
> reasons.1,2,3,... At the same time, one is going to suffer
> considerably when the number of control points is in the thousands.
> Slow to evaluate so many coefficients fo
The linear solver in the TPS routine is naive for any number of
reasons.1,2,3,... At the same time, one is going to suffer
considerably when the number of control points is in the thousands.
Slow to evaluate so many coefficients for each point.
It is not such a task to improve the routine. Best
Not sure whether this can be considered a bug, so I give it for what it
is worth.
I'm doing thin plate spline transformation from one set of projected
coordinates to another. Both sets have values between -60 and 60
(meters). A typical set of gcps looks like:
-gcp 62402 -74383 18191
