ORFEO 8.0.X / Windows 10 / 170+ GB ram
I often have very large rasters (ie 100+ GB, 10+ bands, Geotiff 16 Uint, 10 to 20m pixels, for example) that I would like to use ORFEO for dimensionality reduction, specifically ICA.
I note that ORFEO performs well on smaller tests, but on large files, the output often contains significant voids.
I’ve tried a few variations of settings in smaller tests that work, but fail on these very large rasters.
Are there any suggestions for settings/formats use this / dont use that, for very large rasters, or might this be outside the range of a workable input?
Thanks for your time, very appreciative of the toolbox.
Dear @oscillation,
Thank you for using OTB.
Orfeo ToolBox has been originally developed for Very High Resolution imagery (VHR), and as such is able to scale up to the size of those images. Most of the algorithms implemented in OTB supports piece-wise processing, allowing to process very large image under memory constraints. Most of them are also seamlessly multi-threaded.
I note that ORFEO performs well on smaller tests, but on large files, the output often contains significant voids.
I never noticed this, it might be a bug. Could you share more context? (What application did you use, with what parameters, what products did you use as input).
Sincerely.
Julien 
Writing large (up to Tb in my own experience) images is usually not a problem with OTB, but it can depend on the application (the RAM is well adressed to manage the pixel computation, but the internal storage of, e.g. a large matrix can be an issue).
As you do not mention it, I recommand all producers of large geotif images to work with the following extended file name keywords : “output.tif?&gdal:co:COMPRESS=LZW&gdal:co:BIGTIFF=YES&gdal:co:TILED=YES”
1 Like