ilastikrag Tutorial: Anisotropic Data¶
In [1]:
%matplotlib inline
import matplotlib
import pylab
pylab.rcParams['figure.figsize'] = (20.0, 20.0)
from ilastikrag.util import colorize_labels
from collections import OrderedDict
from itertools import groupby
from IPython.display import display
In [2]:
import numpy as np
import pandas as pd
import vigra
import ilastikrag
Load Data¶
In [3]:
# Note: Rag inputs must be VigraArray (with axistags)
grayscale_2d = vigra.impex.readImage('grayscale.png', dtype='NATIVE').withAxes('yx')
membranes_2d = vigra.impex.readImage('membranes.png', dtype='NATIVE').withAxes('yx')
superpixels_2d = vigra.impex.readImage('superpixels.png', dtype=np.uint32).withAxes('yx')
In [4]:
vigra.multiImshow(OrderedDict( [('grayscale', (grayscale_2d, 'img')),
('membranes', (membranes_2d, 'img')),
('superpixels', (colorize_labels(superpixels_2d), 'img'))]), (1,3))
(Fabricate 3D data)¶
In [5]:
# Just for this tutorial, we're cheating:
# We'll just stack copies of the 2D data
grayscale = np.tile(grayscale_2d, (20,1,1))
grayscale.axistags = vigra.defaultAxistags('zyx')
membranes = np.tile(membranes_2d, (20,1,1))
membranes.axistags = vigra.defaultAxistags('zyx')
superpixels = np.zeros(shape=grayscale.shape, dtype=np.uint32)
superpixels = vigra.taggedView(superpixels, 'zyx')
# Every slice must have unique ids
for z in range(superpixels.shape[0]):
superpixels[z] = superpixels_2d + superpixels_2d.max()+1
Create Rag: flat_superpixels=True¶
In [6]:
rag = ilastikrag.Rag(superpixels, flat_superpixels=True)
Compute features along isotropic axes (edge_group='yx')¶
In [7]:
iso_edge_features = rag.compute_features( grayscale,
['standard_edge_mean', 'standard_sp_quantiles_50'],
edge_group='yx' )
iso_edge_features[:5]
Out[7]:
Compute 'flatedge' features along the anisotropic axis (edge_group='z')¶
In [8]:
aniso_edge_features = rag.compute_features( grayscale,
['standard_flatedge_mean', 'standard_flatedge_quantiles_50'],
edge_group='z' )
aniso_edge_features[:5]
Out[8]:
Optional: Join features into one table¶
In [9]:
all_edge_features = pd.concat((iso_edge_features, aniso_edge_features))
# If tables don't have the same column names, you'll see NaNs.
# Depending on your intentions, you may want to rename the
# columns before calling pd.concat(), or call DataFrame.fillna() on the result.
display(all_edge_features[:5])
display(all_edge_features[-5:])
Show supported feature names¶
In [10]:
for _, grouped_names in groupby(rag.supported_features(), lambda s: s.split('_')[:-1]):
for feature_name in grouped_names:
print feature_name
print ""
In [ ]: