[gdal-dev] Source pixels used when warping using cubic resampling

[Gregory, Matthew]

Hi Even,

Even Rouault wrote:
> > Hi all,
> > 
> > I am warping and upsampling an image from 4km to 30m using cubic resampling.
> >  I've been warping different sized windows from the source image like this:
> > 
> > [snip]
>
> maybe try to add
> 
> -et 0
> 
> and/or
> 
> -wo XSCALE=133.33333333333334 -wo YSCALE=133.33333333333334
> 
> where 133.3333 = 4000 / 30
> 
> XSCALE/YSCALE are "un-documented" there:
> https://github.com/OSGeo/gdal/blob/master/gdal/alg/gdalwarpkernel.cpp#L1120

Wow, thanks for this - would have never figured this out on my own.  For me, both -et and the two -wo flags were necessary to make the reprojections from the full window and the row be equivalent. 

Using the -wo flags on their own gets pretty close (~0.5% percent difference in pixel values in my test), but -et on its own was still substantially different (~7% percent differences in pixel values).

thanks, matt