Bug: X/Y inverted when using numpy.reshape() on a GDALRaster's GDALBand

[Vincent Letarouilly]

When importing GeoTIF data through GDALRaster with numpy package installed all the array is a mess because of a reversed X/Y parameters passed to the numpy.reshape() function.
I was able to see this bug by importing a GeoTIF via GDALRaster. The generated data did not make any sense so I decided to export the array to an image again to see if it was me doing shit or if there was a bug somewhere. When exporting the GeoTIF again to an image I I ran into something really strange. You can see in attachements the original tif and the generated one that is not correct due to this X/Y mismatch. Here is the code to reproduce:

from PIL import Image
from django.contrib.gis.gdal import GDALRaster
import os, numpy

rst = GDALRaster(os.path.join('/tmp', 'cea.tif'))
data = rst.bands[0].data()
rescaled = (255.0 / data.max() * (data - data.min())).astype(numpy.uint8)
im = Image.fromarray(rescaled)
im.save('/tmp/cea_2.tif')

You can see in the (see ​| Numpy Reshape documentation) that you should pass a tuple with the height and the width (in that order). And in the django code, the size tuple is built with width first then height at line 111:

size = (self.width - offset[0], self.height - offset[1])

The bug is located in django.contrib.gis.gdal.raster.band.py at line 148:

data_array, dtype=numpy.dtype(data_array)).reshape(size)

should be replaced by

data_array, dtype=numpy.dtype(data_array)).reshape((size[1], size[0]))

[Vincent Letarouilly]

Original GeoTIF sample

[Vincent Letarouilly]

GeoTIF after being imported

[Vincent Letarouilly]

Original GeoTIF sample

[Vincent Letarouilly]

GeoTIF after being imported

[Vincent Letarouilly]

Link to the fixed branch ​https://github.com/Opa-/django/tree/ticket_26432
And the commit ​https://github.com/Opa-/django/commit/f530405061edfe137858e3a0cfe3e7b177438804

[Tim Graham]

The pull request currently lacks a test. Please uncheck "Needs test" when you add one. Thanks!

[Vincent Letarouilly]

I've just commited tests for the patch ;)
I'm checking the length of the row & columns of the numpy + some arbitrary values.

[Tim Graham]

Please uncheck "Patch needs improvement" when you address the review comments.

[Tim Graham <timograham@…>]

In 461f74ab:

Fixed #26432 -- Fixed size tuple order when using numpy reshape on a GDALBand.

[Tim Graham <timograham@…>]

In a3265af8:

Refs #26432 -- Skipped a raster test as needed.