ValueError: iswt2 only supports 2D arrays. see iswtn for a general n-dimensionsal ISWT

[adhaka3]

While running, wavelet='db1' Cooef = pywt.swt2(Image[:,:], wavelet, level=2) Image = pywt.iswt2(Cooef, wavelet) it gives the error: ValueError: iswt2 only supports 2D arrays. see iswtn for a general n-dimensionsal ISWT

Shape of Cooef= (2, 2) Shape of Cooef[0][0]= (155, 240, 240) #(approx. coeff) Shape of Cooef[0][1]= (3, 155, 240, 240) #(detailed coeff) Shape of Cooef[1][0]= (155, 240, 240) #(approx. coeff) Shape of Cooef[1][1]= (3, 155, 240, 240) #(detailed coeff)

I am pretty new to the concept of Wavelets; this might be a stupid question to ask but I can't get my head around this!!!!

[grlee77]

This looks like Image[:, :].shape is 3d (155, 240, 240)? Is your intention to do a batch of 155 2d transforms? (By default swt2 will transform the last two axes). The output of the forward transform looks as expected. This issue is just that iswt2 doesn't support batch transforms (it doesn't have an axes argument).

You can use swtn and iswtn to do batch 2d transforms. In each case, specify axes=(-2, -1) so that only the last two axes are transformed (assuming a batch 2D transform was your intention). The coefficient format will be a list of dictionaries as described in the docstring.

[grlee77]

I do think having axes available for swt2, but not iswt2 is a little odd from an API standpoint. Internally, swt2 just calls swtn which has axes support, so it works there. iswt2 is an older implementation that did not support axes (it does not call iswtn internally). I think the simpler iswt2 implementation was retained because it is a little easier to read and follow the logic than the more general iswtn case and so may be useful to future maintainers/developers as a reference.

[adhaka3]

I am trying to do image fusion by doing a batch transform through SWT on 3d images of shape (155,240,240). Hence, I wanted to get the coefficients, make some modifications and then get the inverse. Now, on implementing the solution you gave, wavelet = 'db1' cooef1 = pywt.swtn(I1, wavelet, level=2, axes=(-2, -1)) cooef2 = pywt.swtn(I2, wavelet, level=2, axes=(-2, -1)) It gives me 2 array of shape (2, ).

On running, np.shape(cooef1[0]), I get () as the output.

I am not able to merge the set of coefficients anymore bcoz of this. P.S- Thanks for this library and the help!!

[rgommers]

I do think having axes available for swt2, but not iswt2 is a little odd from an API standpoint.

Indeed, that seems like a wart. Let's leave this issue open for resolving that, probably by adding an axes keyword to iswt2.

[grlee77]

Indeed, that seems like a wart. Let's leave this issue open for resolving that, probably by adding an axes keyword to iswt2.

That was also addressed as part of #527, which I just merged, so we can now close this as well!