[Gdal-dev] gdalwarp, using GCPs with pixel positions outside the input image

[Chris Campbell]

> I think your approach is correct and the problem is likely due to
> inaccuracies and differing round-off errors in the polynomial calculation
> code.  What order polynmomial are you using?   In general I am not 
> impressed
> with the numerical quality of the GDAL polynomial solver and transformer.

I see... I am using order = 1.

> I'm not really sure why you need to do things in chunks though.  GDAL warp
> is intended to be usable for large images, and I have personally processed
> images up to 40GB without any apparent issues other than that it takes 
> quite
> a while.  It should scale up to hundreds of GB images at least.

I thought this might be pointed out :)
The scalability problem is not with gdal but with the generation process, 
which needs more than a reasonable amount of memory to generate a 
50,000x50,000 pixel image. I can generate it in smaller tiles and then join 
these together in a much less memory demanding way and then warp, but it 
takes a long time and a lot of redundant disk space until I can finally get 
a look at some rectified output.
I don't know too much about these kind of matrices, but is it possible to 
calculate the transformation coefficients using the original set of GCPs and 
then alter some values in the matrix to get a desired pixel offset?

Chris