[gdal-dev] gdalwarp different results (vrt vs direct)

[Even Rouault]

Stefan,

> We have a lot of aerial images. Each image has a size of 8000x8000 pixels
> and the resolution is 12.5cm/pixel (-> 1 sqkm). They form a mosaic w/o any
> gaps and overlaps. For transforming them into a new reference frame I'm
> creating an vrt first and then using gdalwarp for the transformation (in a
> for-loop). With this approach we can force the resulting images to have
> bounding boxes with "nice" coordinates:
> 
> gdalwarp -s_srs "+proj=somerc +lat_0=46.952405555555555N
> +lon_0=7.439583333333333E +ellps=bessel +x_0=600000 +y_0=200000
> +towgs84=674.374,15.056,405.346 +units=m +units=m +k_0=1
> +nadgrids=./chenyx06a.gsb" -t_srs "+proj=somerc +lat_0=46.952405555555555N
> +lon_0=7.439583333333333E +ellps=bessel +x_0=2600000 +y_0=1200000
> +towgs84=674.374,15.056,405.346 +units=m +k_0=1 +nadgrids=@null" -tr 0.125
> 0.125 -te 2611000 1233000 2612000 1234000 -co 'TILED=YES'  ortho2014.vrt
> output.tif

Side remark: if you specify both +towgs84 and +nadgrids in the same SRS 
definition, the +towgs84 will be ignored by proj.4 (the grid is supposed to 
contain all transformation shifts from/to WGS84), so you can just remove it.

> 
> The resulting image does have some strange artefacts, similar to a breaking
> edge almost vertically through the whole image [1].

This occurs in the middle of a source tile, or at the boundary between 2 tiles 
?

> 
> When using gdalwarp with the same -s_srs and -t_srs parameters for a single
> image there is no breaking edge [2].

Just to be sure: you mean you are using the above -s_srs and -t_srs on one of 
the tile that makes the ortho2014.vrt ?
And what if you make a VRT with just that source image ?

On that tile, did you try "gdal_translate ortho2014.vrt out.tif -projwin ULX 
ULY LRX LRY" where ULX,YLY,LRX,LRY are the exact bounds of the tile, and check 
that out.tif perfectly overlaps on top of the source tile ? There could be 
perhaps some rounding error in the way the placement of the source tile is 
made in the VRT.

> 
> Any ideas what's going on?

Not really. Having a way to reproduce would be interesting.

Hum, perhaps try adding -et 0 to your gdalwarp command line, but I fail to see 
why this would make a difference between a VRT or TIFF source.

> 
> I'm using gdal 1.11.

Any chance you can try with 2.0 or the dev version ?

Even

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com