In [1]:
%matplotlib inline
%pylab inline


Populating the interactive namespace from numpy and matplotlib

In [2]:
repeat = 3

Load test data


In [3]:
import menpo.io as mio
from menpo.landmark import labeller, ibug_face_49
from menpofast.utils import convert_from_menpo

path = '/data/'
group = 'ibug_face_49'
db = 'lfpw'

test_images = []
for i in mio.import_images(path + 'PhD/DataBases/faces/' + db + '/testset/', 
                           verbose=True, max_images=None):
    
    # convert the image from menpo Image to menpofast Image (channels at front)
    i = convert_from_menpo(i)
    
    labeller(i, 'PTS', eval(group))
    i.crop_to_landmarks_proportion_inplace(0.5, group='PTS')
    i = i.rescale_landmarks_to_diagonal_range(150, group=group)
    
    if i.n_channels == 3:
        i = i.as_greyscale(mode='average')
    test_images.append(i)


- Loading 224 assets: [====================] 100%

In [6]:
from menpo.visualize import visualize_images

visualize_images(test_images)


Active Appearance Models

Fitting Parameters


In [ ]:
from alabortijcv2015.aam.algorithm import SIC

algorithm_cls = SIC

sampling_step = 1

sampling_mask = np.require(np.zeros((17, 17)), dtype=np.bool)
sampling_mask[::sampling_step, ::sampling_step] = True

n_shape = [3, 12]
n_appearance = [50, 100]

noise_std = [None, 0, 0.02, 0.04, 0.06]
max_iters = 20
prior = False

Global PWA


In [8]:
from alabortijcv2015.utils import pickle_load

aam = pickle_load(path + 'PhD/Models/ijcv2015/exp1_global_pwa_int')

In [9]:
from alabortijcv2015.aam import StandardAAMFitter

fitter = StandardAAMFitter(aam, algorithm_cls=algorithm_cls, n_shape=n_shape,
                           n_appearance=n_appearance, sampling_step=sampling_step)

In [10]:
from alabortcvpr2015.utils import pickle_dump
from alabortcvpr2015.result import SerializableResult

for n in noise_std:
    
    fitter_results = []
    
    for seed in xrange(repeat):

        np.random.seed(seed=seed)

        for j, i in enumerate(test_images):

            gt_s = i.landmarks[group].lms

            if n is not None:
                s = fitter.perturb_shape(gt_s, noise_std=n)
            else:
                s = gt_s

            fr = fitter.fit(i, s, gt_shape=gt_s, max_iters=max_iters, prior=prior)

            fitter_results.append(fr)
            fr.downscale = 0.5

            #print 'Image: ', j
            #print fr
    print n

    results = [SerializableResult('none', fr.shapes(), fr.n_iters, 'FastSIC', fr.gt_shape) 
               for fr in fitter_results]

    pickle_dump(results, path + 'PhD/Results/ijcv2015/exp1_global_pwa_int_fastsic' + str(n))


None
0.0
0.02
0.04
0.06

In [11]:
from menpofit.visualize import visualize_fitting_results
    
visualize_fitting_results(fitter_results)


Global TPS


In [12]:
from alabortijcv2015.utils import pickle_load

aam = pickle_load(path + 'PhD/Models/ijcv2015/exp1_global_tps_int')

In [13]:
from alabortijcv2015.aam import StandardAAMFitter
from alabortijcv2015.aam.algorithm import FastSIC_GN

fitter = StandardAAMFitter(aam, algorithm_cls=algorithm_cls, n_shape=n_shape,
                           n_appearance=n_appearance, sampling_step=sampling_step)

In [14]:
from alabortcvpr2015.utils import pickle_dump
from alabortcvpr2015.result import SerializableResult
    
for n in noise_std:
    
    fitter_results = []
    
    for seed in xrange(repeat):

        np.random.seed(seed=seed)

        for j, i in enumerate(test_images):

            gt_s = i.landmarks[group].lms

            if n is not None:
                s = fitter.perturb_shape(gt_s, noise_std=n)
            else:
                s = gt_s

            fr = fitter.fit(i, s, gt_shape=gt_s, max_iters=max_iters, prior=prior)

            fitter_results.append(fr)
            fr.downscale = 0.5

            #print 'Image: ', j
            #print fr
    print n

    results = [SerializableResult('none', fr.shapes(), fr.n_iters, 'FastSIC', fr.gt_shape) 
               for fr in fitter_results]

    pickle_dump(results, path + 'PhD/Results/ijcv2015/exp1_global_tps_int_fastsic' + str(n))


None
0.0
0.02
0.04
0.06

In [15]:
from menpofit.visualize import visualize_fitting_results
    
visualize_fitting_results(fitter_results)


Patch


In [16]:
from alabortijcv2015.utils import pickle_load

aam = pickle_load(path + 'PhD/Models/ijcv2015/exp1_patch_int')

In [17]:
from alabortijcv2015.aam import StandardAAMFitter
from alabortijcv2015.aam.algorithm import FastSIC_GN

fitter = StandardAAMFitter(aam, algorithm_cls=algorithm_cls, n_shape=n_shape,
                           n_appearance=n_appearance, sampling_step=sampling_step)

In [18]:
from alabortcvpr2015.utils import pickle_dump
from alabortcvpr2015.result import SerializableResult
    
for n in noise_std:
    
    fitter_results = []
    
    for seed in xrange(repeat):

        np.random.seed(seed=seed)

        for j, i in enumerate(test_images):

            gt_s = i.landmarks[group].lms

            if n is not None:
                s = fitter.perturb_shape(gt_s, noise_std=n)
            else:
                s = gt_s

            fr = fitter.fit(i, s, gt_shape=gt_s, max_iters=max_iters, prior=prior)

            fitter_results.append(fr)
            fr.downscale = 0.5

            #print 'Image: ', j
            #print fr
    print n

    results = [SerializableResult('none', fr.shapes(), fr.n_iters, 'FastSIC', fr.gt_shape) 
               for fr in fitter_results]

    pickle_dump(results, path + 'PhD/Results/ijcv2015/exp1_patch_int_fastsic' + str(n))


None
0.0
0.02
0.04
0.06

In [19]:
from menpofit.visualize import visualize_fitting_results
    
visualize_fitting_results(fitter_results)


Linear PWA


In [20]:
from alabortijcv2015.utils import pickle_load

aam = pickle_load(path + 'PhD/Models/ijcv2015/exp1_linear_global_pwa_int')

In [21]:
from alabortijcv2015.aam import LinearAAMFitter
from alabortijcv2015.aam.algorithm import FastSIC_GN

fitter = LinearAAMFitter(aam, algorithm_cls=algorithm_cls, n_shape=n_shape,
                         n_appearance=n_appearance, sampling_step=sampling_step)

In [22]:
from alabortcvpr2015.utils import pickle_dump
from alabortcvpr2015.result import SerializableResult
    
for n in noise_std:
    
    fitter_results = []
    
    for seed in xrange(repeat):

        np.random.seed(seed=seed)

        for j, i in enumerate(test_images):

            gt_s = i.landmarks[group].lms

            if n is not None:
                s = fitter.perturb_shape(gt_s, noise_std=n)
            else:
                s = gt_s

            fr = fitter.fit(i, s, gt_shape=gt_s, max_iters=max_iters, prior=prior)

            fitter_results.append(fr)
            fr.downscale = 0.5

            #print 'Image: ', j
            #print fr
    print n

    results = [SerializableResult('none', fr.shapes(), fr.n_iters, 'FastSIC', fr.gt_shape) 
               for fr in fitter_results]

    pickle_dump(results, path + 'PhD/Results/ijcv2015/exp1_linear_global_pwa_int_fastsic' + str(n))


None
0.0
0.02
0.04
0.06

In [23]:
from menpofit.visualize import visualize_fitting_results
    
visualize_fitting_results(fitter_results)


Linear Global TPS


In [24]:
from alabortijcv2015.utils import pickle_load

aam = pickle_load(path + 'PhD/Models/ijcv2015/exp1_linear_global_tps_int')

In [25]:
from alabortijcv2015.aam import LinearAAMFitter
from alabortijcv2015.aam.algorithm import FastSIC_GN

fitter = LinearAAMFitter(aam, algorithm_cls=algorithm_cls, n_shape=n_shape,
                         n_appearance=n_appearance, sampling_step=sampling_step)

In [26]:
from alabortcvpr2015.utils import pickle_dump
from alabortcvpr2015.result import SerializableResult
    
for n in noise_std:
    
    fitter_results = []
    
    for seed in xrange(repeat):

        np.random.seed(seed=seed)

        for j, i in enumerate(test_images):

            gt_s = i.landmarks[group].lms

            if n is not None:
                s = fitter.perturb_shape(gt_s, noise_std=n)
            else:
                s = gt_s

            fr = fitter.fit(i, s, gt_shape=gt_s, max_iters=max_iters, prior=prior)

            fitter_results.append(fr)
            fr.downscale = 0.5

            #print 'Image: ', j
            #print fr
    print n

    results = [SerializableResult('none', fr.shapes(), fr.n_iters, 'FastSIC', fr.gt_shape) 
               for fr in fitter_results]
    
    pickle_dump(results, path + 'PhD/Results/ijcv2015/exp1_linear_global_tps_int_fastsic' + str(n))


None
0.0
0.02
0.04
0.06

In [27]:
from menpofit.visualize import visualize_fitting_results
    
visualize_fitting_results(fitter_results)


Linear Patch


In [28]:
from alabortijcv2015.utils import pickle_load

aam = pickle_load(path + 'PhD/Models/ijcv2015/exp1_linear_patch_int')

In [29]:
from alabortijcv2015.aam import LinearAAMFitter
from alabortijcv2015.aam.algorithm import FastSIC_GN

fitter = LinearAAMFitter(aam, algorithm_cls=algorithm_cls, n_shape=n_shape,
                         n_appearance=n_appearance, sampling_step=sampling_step)

In [30]:
from alabortcvpr2015.utils import pickle_dump
from alabortcvpr2015.result import SerializableResult
    
for n in noise_std:
    
    fitter_results = []
    
    for seed in xrange(repeat):

        np.random.seed(seed=seed)

        for j, i in enumerate(test_images):

            gt_s = i.landmarks[group].lms

            if n is not None:
                s = fitter.perturb_shape(gt_s, noise_std=n)
            else:
                s = gt_s

            fr = fitter.fit(i, s, gt_shape=gt_s, max_iters=max_iters, prior=prior)

            fitter_results.append(fr)
            fr.downscale = 0.5

            #print 'Image: ', j
            #print fr
    print n

    results = [SerializableResult('none', fr.shapes(), fr.n_iters, 'FastSIC', fr.gt_shape) 
               for fr in fitter_results]

    pickle_dump(results, path + 'PhD/Results/ijcv2015/exp1_linear_patch_int_fastsic' + str(n))


None
0.0
0.02
0.04
0.06

In [31]:
from menpofit.visualize import visualize_fitting_results
    
visualize_fitting_results(fitter_results)


Parts


In [32]:
from alabortijcv2015.utils import pickle_load

aam = pickle_load(path + 'PhD/Models/ijcv2015/exp1_parts_int')

In [33]:
from alabortijcv2015.aam import PartsAAMFitter
from alabortijcv2015.aam.algorithm import FastSIC_GN

fitter = PartsAAMFitter(aam, algorithm_cls=algorithm_cls, n_shape=n_shape,
                        n_appearance=n_appearance, sampling_mask= sampling_mask)

In [34]:
from alabortcvpr2015.utils import pickle_dump
from alabortcvpr2015.result import SerializableResult
    
for n in noise_std:
    
    fitter_results = []
    
    for seed in xrange(repeat):

        np.random.seed(seed=seed)

        for j, i in enumerate(test_images):

            gt_s = i.landmarks[group].lms

            if n is not None:
                s = fitter.perturb_shape(gt_s, noise_std=n)
            else:
                s = gt_s

            fr = fitter.fit(i, s, gt_shape=gt_s, max_iters=max_iters, prior=prior)

            fitter_results.append(fr)
            fr.downscale = 0.5

            #print 'Image: ', j
            #print fr
    print n

    results = [SerializableResult('none', fr.shapes(), fr.n_iters, 'FastSIC', fr.gt_shape) 
               for fr in fitter_results]

    pickle_dump(results, path + 'PhD/Results/ijcv2015/exp1_parts_int_fastsic' + str(n))


None
0.0
0.02
0.04
0.06

In [35]:
from menpofit.visualize import visualize_fitting_results
    
visualize_fitting_results(fitter_results)