[SciPy-User] silently different results when running scipy.signal.fftconvolve on ubuntu karmic vs. lucid

[Pauli Virtanen]

Wed, 26 Jan 2011 15:02:31 +0000, Pauli Virtanen wrote:
[clip]
> Probably the image package you used to save the images had bugs with
> dealing with arrays, as its version is different:
[clip]

This turns out to be the case: using Python-imaging 1.1.6 on Maverick,
the example program produces garbled images, whereas with 1.1.7 it
works as expected.

So, the problem is not in Numpy/Scipy and the numerical data is OK.
However, the image library apparently used to interpret the array
data given to it as 8-bit integers in all cases (resulting to
the strange 8-pixel and 16-pixel stripe structure in the figures).

-- 
Pauli Virtanen