[gdal-dev] VRT derived band pixel functions written in Python

Rutger kassies at gmail.com
Tue Sep 13 00:02:09 PDT 2016

Hi Even,

This is really amazing. I'm becoming more and more a fan of Numba for number
crunching, so this certainly makes my day. As soon as i can find a Win x64
dev version on a Conda channel I'll give it a try.

A use case that comes to mind, and which i run into regularly, is when i
want to do some simple aggregation before using something like gdalwarp. For
example when you have a file containing 24 hourly temperature values, and
you are only interested in the daily mean. Currently i either aggregate
before warping and write the intermediates to disk, or aggregate after
warping which is computationally inefficient. Neither is optimal.

A few questions, can you access a files metadata from within a pixel
function? This would perhaps allow for example interpolating atmospheric
data to the overpass time of a satellite image.

Do the pixel functions also work with @numba.vectorize(), in particular when
targeting 'parallel' or 'cuda'. And would that give parallel processing for
both IO and calculations?

You should give the folks at Continuum a heads up, i'm sure they appreciate
seeing Numba used like this. 


View this message in context: http://osgeo-org.1560.x6.nabble.com/gdal-dev-VRT-derived-band-pixel-functions-written-in-Python-tp5285323p5285450.html
Sent from the GDAL - Dev mailing list archive at Nabble.com.

More information about the gdal-dev mailing list