Dear list, I am trying to use gdal.ReprojectImage to warp an image within Python. Usually when the image is large i like to do this in smaller chunks, based on the blocksize of the destination image.
Today, when doing this for a target dataset which was striped, row based (blocksize = [1, ysize]), i noticed significant artifacts when using a resampling method other then nearest neighbor. To do the other resampling methods you would need surrounding pixels as well, but since you provide only the input dataset to GDAL, this is not something you can control, GDAL is doing all the reading from the source data in the background. So whats the best way to ensure correct results when using ReprojectImage? I can look at the destination blocksize and make it a bit larger in case of a single pixel dimension, process 5 or 10 lines at once for example. But how can i be sure that something like that would solve the source of the artifacts. I cant check all valid blocksize combinations, and like to avoid surprises. I'm really hoping to avoid any windowed approach, eg warping to a larger area and slicing of the borders. Here is a simplified example replicating the artifacts i'm seeing: http://nbviewer.jupyter.org/gist/RutgerK/ef6d1aae57625e9a2679 Needless to say that the gdalwarp cmd utility (of course) works for fine for striped data. Regards, Rutger -- View this message in context: http://osgeo-org.1560.x6.nabble.com/Using-gdal-ReprojectImage-on-chunk-block-basis-tp5258018.html Sent from the GDAL - Dev mailing list archive at Nabble.com. _______________________________________________ gdal-dev mailing list gdal-dev@lists.osgeo.org http://lists.osgeo.org/mailman/listinfo/gdal-dev
