Possible not optimal zipping for nii.gz?

[SebastianoF]

Hi,
In loading and re-saving an image with nibabel I noticed an increase in memory.
Below some IO from an ipython session:

In [19]: im = nib.load('initial.nii.gz')

In [20]: ! du -h initial.nii.gz
388K	initial.nii.gz

In [21]: im.get_data_dtype()
Out[21]: dtype('<u2')

In [22]: im.set_data_dtype('<u2')

In [23]: nib.save(im, 'initial_re_saved_same_dtype.nii.gz')

In [24]: ! du -h initial_re_saved_same_dtype.nii.gz
872K	initial_re_saved_same_dtype.nii.gz

Here I open the image with ITK-snap, and re-save it under the name initial_re_saved_same_dtype_re_Saved_itksnap.nii.gz:

In [25]: ! du -h initial_re_saved_same_dtype_re_Saved_itksnap.nii.gz
388K	initial_re_saved_same_dtype_re_Saved_itksnap.nii.gz

Same issue happens even if not changing the data-type, but only in loading and re-saving.

In [26]: im = nib.load('initial.nii.gz')

In [27]: nib.save(im, 'initial_just_re_saved.nii.gz')

In [28]: ! du -h initial_just_re_saved.nii.gz
872K	initial_just_re_saved.nii.gz

Is it an issue or am I doing something wrong? Any possible workaround?
Thanks!

[effigies]

Possibly related: #382

You can adjust the default compression level. I can look up where, if you can't quickly find it. (I think it's in openers.py.)

[matthew-brett]

Nibabel uses a relatively low level of compression by default. This usually doesn't make much difference, but maybe that's not true for you image. Try doing:

nib.openers.Opener.default_compresslevel = 9

before you load and save, does that help?

[SebastianoF]

In [1]: import nibabel as nib

In [2]: nib.openers.Opener.default_compresslevel = 9

In [3]: im = nib.load('initial.nii.gz')

In [4]: ! du -h initial.nii.gz
388K	initial.nii.gz

In [5]: nib.save(im, 'initial_just_re_saved.nii.gz')

In [6]: ! du -h initial_just_re_saved.nii.gz
352K	initial_just_re_saved.nii.gz

Bingo!
Thanks for the quick answer!

[satra]

@matthew-brett - would it make sense to up the default to something like 5? may provide a reasonable balance between time to compress and compression ratio. looks like decompression isn't bad in terms of time.

https://catchchallenger.first-world.info/wiki/Quick_Benchmark:_Gzip_vs_Bzip2_vs_LZMA_vs_XZ_vs_LZ4_vs_LZO