[gdal-dev] ogr2ogr -tps with more than 1000 control points

[Jan Hartmann]

Thanks Martin, I upgraded to GDAL 2.4.2, and now it works. The problem 
was with control points  being too near to each other. GDAL 2.4.2 even 
worked on a testcase with two control points on the same location. 
That's important, for I want to rubbersheet historical maps as exactly 
as possible, and for that I use interpolated values between the control 
point that have been set manually. In some case these interpolated 
values end up at very small distances, and now gdal doesn't break on 
them any more.

Regards,

Jan


On 13-8-2019 9:42, Martin.Franzke at telekom.de wrote:
> Hi Jan,
>
> I had a similar problem, which was solved by normalizing the TPS equations. The patch (https://github.com/OSGeo/gdal/commit/c85c58ac781ef781cd126006d91cb5eed2dd53e2) is included in gdal versions >= 2.4. This may help you as well.
>
> Martin
>
> -----Ursprüngliche Nachricht-----
> Von: gdal-dev <gdal-dev-bounces at lists.osgeo.org> Im Auftrag von Jan Hartmann
> Gesendet: Freitag, 9. August 2019 21:18
> An: Even Rouault <even.rouault at spatialys.com>
> Cc: gdal-dev at lists.osgeo.org
> Betreff: Re: [gdal-dev] ogr2ogr -tps with more than 1000 control points
>
> Yes, I already noticed that. But a run with thousand gcp's is still quite fast. I am looking for a smart way to segment these large maps.
> You cannot rubbersheet them in many  parts, because then the borders won't fit any more. The only thing I can come up with at the moment is creating common points on the borders of adjacent parts, and adding them to the control points. If you have any ideas, let me know.
>
> Jan
>
> On 8/9/2019 8:19 PM, Even Rouault wrote:
>> On vendredi 9 août 2019 20:08:31 CEST Jan Hartmann wrote:
>>> Thanks Even. GDAL was built with Armadillo, so I'm going to test the
>>> GCP's. Just wanted to be sure there was no possibility of a buffer
>>> overflow, or something like that. The program should be able to handle
>>> an unlimited number of control points, right?
>> I'd rather say there's no hard-coded limit ;-) But the processing time will
>> increase with the number of GCPs: linearly when applying the result of the GCP
>> adjustment for each point transformed, and with an initial computation cost of
>> the adjustment coefficients that has a complexiy a bit below O(n^3).
>>
>> Even
>>
> _______________________________________________
> gdal-dev mailing list
> gdal-dev at lists.osgeo.org
> https://lists.osgeo.org/mailman/listinfo/gdal-dev