from menpofit.io import load_fitter
[docs]def load_balanced_frontal_face_fitter():
r"""
Loads a frontal face patch-based AAM fitter that is a good compromise
between model size, fitting time and fitting performance. The model returns
68 facial landmark points (the standard IBUG68 markup).
Note that the first time you invoke this function, menpofit will
download the fitter from Menpo's server. The fitter will then be stored
locally for future use.
The model is a :map:`PatchAAM` trained using the following parameters:
=================== =================================
Parameter Value
=================== =================================
`diagonal` 110
`scales` (0.5, 1.0)
`patch_shape` [(13, 13), (13, 13)]
`holistic_features` `menpo.feature.fast_dsift()`
`n_shape` [5, 20]
`n_appearance` [30, 150]
`lk_algorithm_cls` :map:`WibergInverseCompositional`
=================== =================================
It is also using the following `sampling` grid:
.. code-block:: python
import numpy as np
patch_shape = (13, 13)
sampling_step = 4
sampling_grid = np.zeros(patch_shape, dtype=np.bool)
sampling_grid[::sampling_step, ::sampling_step] = True
sampling = [sampling_grid, sampling_grid]
Additionally, it is trained on LFPW trainset, HELEN trainset, IBUG and AFW
datasets (3283 images in total), which are hosted in
http://ibug.doc.ic.ac.uk/resources/facial-point-annotations/.
Returns
-------
fitter : :map:`LucasKanadeAAMFitter`
A pre-trained :map:`LucasKanadeAAMFitter` based on a :map:`PatchAAM`
that performs facial landmark localization returning 68 points (iBUG68).
"""
return load_fitter('balanced_frontal_face_aam')