Python, GDAL and building raster attribute tables

Question

I have an integer raster for which I would like to build a raster attribute table using Python and GDAL. I can create a GDAL raster attribute table in Python as follows:

>>> rat = gdal.RasterAttributeTable()

This works fine, as we can see:

>>> rat
<osgeo.gdal.RasterAttributeTable; proxy of <Swig Object of type 'GDALRasterAttributeTableShadow *' at 0x0000000002A53D50> >

The table thus created has no rows or columns:

>>> rat.GetRowCount()
0
>>> rat.GetColumnCount()
0

I create a column called "Value" to store the unique values in the raster:

>>> rat.CreateColumn("Value", gdalconst.GFT_Integer, gdalconst.GFU_MinMax)
0

This is fine, and the column count is updated:

>>> rat.GetColumnCount()
1

Now I have to add values (records) to the column for it to be of any use. I can get a list of unique values from the raster's band like so:

>>> data = band.ReadAsArray(0, 0, dataset.RasterXSize, dataset.RasterYSize)
>>> vals = list(numpy.unique(data))
>>> vals
[3, 7, 8, 10, 11, 12, 13, 14, 17, 18, 20, 22, 23, 25, 27, 28, 41, 45, 52, 56]

What I would like to do is create a for loop to loop through vals and populate the column in the attribute table. I thought I could do something like this:

>>> for i in range(len(vals)):
 rat.SetValueAsInt(i, 0, vals[i])

...where i is the row (record), 0 is the field index and vals[i] is the integer value I want to insert. But it causes an error:

Traceback (most recent call last):
 File "<pyshell#32>", line 2, in <module>
 rat.SetValueAsInt(i, 0, vals[i])
 File "C:\Python27\lib\site-packages\osgeo\gdal.py", line 1139, in SetValueAsInt
 return _gdal.RasterAttributeTable_SetValueAsInt(self, *args)
TypeError: in method 'RasterAttributeTable_SetValueAsInt', argument 4 of type 'int'

The error is caused because I use vals[i] in the call to SetValueAsInt() rather than using an integer directly. For example, rat.SetValueAsInt(0, 0, 0) works fine, but is useless for populating the column if I just want to loop over the list of unique values.

Is this a known issue? Google has so far not been very useful. What can I do to get around this problem?

Answer 1

The SetValueAsInt method is expecting a python int type, not a numpy uint16 type.

>>> print type(vals[0])
<type 'numpy.uint16'>

The following works:

rat.SetValueAsInt(i, 0, int(vals[i]))

Answer 2

If you use vals = numpy.unique(data).tolist() instead it will automatically convert each value to a python int type.