Gaussian smooth fails with float32

[blazzter]

if I have object:

data = np.random.rand(200,200)
data = data.astype('float32')
hs_data = hs.signals.Signal2D(data)
hs_data.plot()

and apply Gaussian filter, it fails with:

Traceback (most recent call last):
File "/home/mkrajnak/.local/lib/python3.5/site-packages/hyperspyui/plugins/gaussianfilter.py", line 198, in set_preview
self.update()
File "/home/mkrajnak/.local/lib/python3.5/site-packages/hyperspyui/plugins/gaussianfilter.py", line 216, in update
record=False, out=out)
File "/home/mkrajnak/.local/lib/python3.5/site-packages/hyperspyui/plugins/gaussianfilter.py", line 112, in gaussian
im_o[:] = gaussian(im_i, sigma, *args, **kwargs)
File "/home/mkrajnak/.local/lib/python3.5/site-packages/skimage/filters/_gaussian.py", line 118, in gaussian
image = convert_to_float(image, preserve_range)
File "/home/mkrajnak/.local/lib/python3.5/site-packages/skimage/_shared/utils.py", line 243, in convert_to_float
image = img_as_float(image)
File "/home/mkrajnak/.local/lib/python3.5/site-packages/skimage/util/dtype.py", line 336, in img_as_float
return convert(image, np.float64, force_copy)
File "/home/mkrajnak/.local/lib/python3.5/site-packages/skimage/util/dtype.py", line 236, in convert
raise ValueError("Images of type float must be between -1 and 1.")
ValueError: Images of type float must be between -1 and 1.

Any clue?

[francisco-dlp]

I think that you've found a bug: in order to make the data palatable for skimage's gaussian filter, the plugin normalises the data making the (wrong) assumption that vmax - vmin > 1.

This can be easily fixed but, @vidartf. why not using ndimage's gaussian filter instead which doesn't require renormalising the images?

[vidartf]

why not using ndimage's gaussian filter instead which doesn't require renormalising the images?

If I remember correctly, this might have been a performance issue. The blur should be fast in order to be responsive when interactively changing the sigma.