AttributeError in Image cleaning and atom finding using pycroscopy

[Jack Liu]

Hello,

I'm new to pycroscopy and trying to use algorithm of Image cleaning and atom finding using pycroscopy to get rid of noise in a TEM image. But I encountered a problem shown below when I trying to execute their command. Can some body help me out? If doSVD is an old function, can somebody tell me the replacing function of it?

[Rajiv Giridharagopal]

Hi Jack,

Yes, that function is old. Is this from an existing notebook that you're working off of? For this to work it be more like:

h5_svd = px.processing.svd_utils.SVD(h5_wins)

h5_svd_group = h5_svd.compute()

h5_u = h5_svd_group['U']

h5_v = h5_svd_group['V']

h5_s = h5_svd_group['S']

where SVD is a class. If you want to dig into it, is in pycroscopy.processing.svd_utils

Best,

Raj

[Jack Liu]

Hi Raj,

Thank you for answering my question!! I did try the command lines you provided, and there are still some errors that I don't really understand what cause them or what is their meaning.

Best, 

Jack

[Rajiv Giridharagopal]

How big is the dataset h5_wins? This is a MemoryError popping up in h5py (i.e. not pycroscopy). I don't know if reducing components from 100 will help avoid this error, since it may be system-dependent...

Raj

[Jack Liu]

Thank you so much for enlighten me!!

This is the step before memory error. Do you think this is too big? I think I just input a regular 1080*1080 image, what should I do to prevent the memory error? Also, I reduced components from 100 to 50 to 3, and even the num_components is 3 I still keep getting the memory error.

Best regards,

Jack

[Rama Vasudevan]

Hi Jack,

I think the issue is that the window step is very small. You should change it to be something like 1/4 of the window size (I think the default is a step size of 1). For a small image (256x256) it should work, but for a larger image such as yours you will quickly hit a memory limit on a normal machine. Please let me know if this works.

Kind Regards,

Rama