Help wanted: implementation of 3D medial axis skeletonization

Josh Warner silvertrumpet999 at
Fri Nov 20 00:23:49 EST 2015

It looks like the lobster and bonsai can be downloaded directly as raw 
volumes (8 bit only, but will serve these purposes perfectly well) here:

This simple wrapper for np.fromfile will load them

import numpy as np

def loadraw(rawfile, shape=None, dtype=np.uint8):
    Load RAW volume to a NumPy array.

    rawfile : string
        Path to *.raw volume.
    shape : tuple
        Shape of the volume. If not provided, output will be a rank-1 stream
        which can be reshaped as desired.
    dtype : NumPy dtype
        Dtype of the raw image volume.
    vol = np.fromfile(rawfile, dtype=dtype)

    if shape is not None:
        vol = vol.reshape(shape)

    return vol

For the lobster, use shape=(56, 324, 301) and recall the voxel spacing has 
a ratio of 1.4:1:1
For the bonsai, use shape=(256, 256, 256) and the volume is isotropic 
(1:1:1 spacing)

On Monday, November 9, 2015 at 8:42:26 PM UTC-5, stefanv wrote:

Hi Kevin 
> On 2015-11-07 09:46:17, 'Kevin Keraudren' via scikit-image <
> scikit-image at> wrote: 
> > I don’t want to volunteer for this project, but I just wanted to 
> > mention that the 3D skeletonization from ITK is easily accessible to 
> > Python through SimpleITK, see example below for the lobster 
> > dataset. SimpleITK could be used for comparison or validation of the 
> > proposed scikit-image algorithm. 
> Thanks for the pointer.  In this case, one of the purposes of the 
> exercise is to stay away from a heavy dependency such as ITK. 
> > PS: is there another way to load those *.pvm datasets in Python 
> > without converting them to raw and hardcoding the image dimension and 
> > pixel type? An plugin? 
> I have no idea about .pvm files, but perhaps we should start a set of 
> plugin gists on the wiki somewhere? 
> Stéfan 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the scikit-image mailing list