Landmarks

In order to demonstrate the functionality of the landmark package we will start by loading a couple of landmarked images using the menpo.io package.



In [ ]:

    
import menpo.io as pio

images = list(pio.import_images('/vol/atlas/databases/lfpw/trainset/*.png', max_images=2))
img = images[0]

In this case, images have been automatically loaded with some landmark information attached. We can confirm that by accessing the property .landmarks on the images.



In [ ]:

    
print 'Is img landmarked?'
print ' - {}'.format(img.landmarks.has_landmarks) 

print 'How many landmark groups does it have?'
print ' - {}'.format(img.landmarks.n_groups) 
    
print 'What labels do these groups have?'
for g in img.landmarks.group_labels:
    print ' - {}'.format(g)

The semantic meaning of landmarks is given by the different the property labels of each landmark group:

Landmarks with no particular semantic meaning are given the label "all" by default.



In [ ]:

    
print 'PTS semantic labels is?'
for l in img.landmarks['PTS'].labels:
    print ' - {}'.format(l)

As a shortcut, we can use None as a key value if it is unambiguous.



In [ ]:

    
l_None = img.landmarks[None]
l_PTS = img.landmarks['PTS']
print 'Does None perform the same as the explicit key if there is no ambiguity? : {}'.format(l_None is l_PTS)

New semantic meaning can be added to a group of existing landmarks by labelling them with a predefined labelling function. This will effectively create a new landmark group with the appropiate semantic labels.



In [ ]:

    
from menpo.landmark.labels import labeller, ibug_68_points

# Label some landmarks
images = labeller(images, 'PTS', ibug_68_points)

print 'How many landmark groups does it have?'
print ' - {}'.format(img.landmarks.n_groups) 
    
print 'What labels do these groups have?'
for g in img.landmarks.group_labels:
    print ' - {}'.format(g)

print 'ibug_68_points semantic labels are?'
for l in img.landmarks['ibug_68_points'].labels:
    print ' - {}'.format(l)

We can view particular landmark groups and their particular labels.



In [ ]:

    
%matplotlib inline
# View a particular group
img.landmarks['PTS'].view_new()



In [ ]:

    
# View a particular group
a = img.landmarks['ibug_68_points']['reye'].view_new()

Note that now we can't use the None trick



In [ ]:

    
try:
    img.landmarks[None]
except ValueError as e:
    print e



In [ ]:

    
from menpo.landmark.labels import ibug_68_trimesh

# Label some landmarks
images = labeller(images, 'PTS', ibug_68_trimesh)

import numpy as np
from menpo.shape import PointCloud

# Set a new set of landmarks via a PointCloud
new_points = PointCloud(np.random.random([100, 2]) * np.max(img.shape))
img.landmarks['new_points'] = new_points
img.landmarks.view()



In [ ]:

    
# Look at the group labels
print img.landmarks.group_labels

# Look at labels inside a group
print img.landmarks['ibug_68_trimesh'].labels

# View a particular label - a PointCloud
img.landmarks['ibug_68_trimesh']['tri'].view()

# Copy some landmarks
img.landmarks = images[1].landmarks
print img.landmarks.group_labels
img.landmarks['ibug_68_trimesh'] = images[0].landmarks['PTS']
print img.landmarks['ibug_68_trimesh']._target is img