diff --git a/bob/bio/face/config/extractor/luan.py b/bob/bio/face/config/extractor/luan.py
new file mode 100644
index 0000000000000000000000000000000000000000..474a8a661df465311b7d291b1ef9efc77239ae32
--- /dev/null
+++ b/bob/bio/face/config/extractor/luan.py
@@ -0,0 +1,6 @@
+#!/usr/bin/env python
+
+import bob.bio.base
+import bob.bio.face
+
+extractor = bob.bio.face.extractor.LuanExtractor()
diff --git a/bob/bio/face/config/preprocessor/luan.py b/bob/bio/face/config/preprocessor/luan.py
new file mode 100644
index 0000000000000000000000000000000000000000..8e375ac1b579127932300c6a8484395a0a63ba63
--- /dev/null
+++ b/bob/bio/face/config/preprocessor/luan.py
@@ -0,0 +1,18 @@
+import bob.bio.face
+
+# This is the size of the image that this model expects
+CROPPED_IMAGE_HEIGHT = 96
+CROPPED_IMAGE_WIDTH = 96
+
+RIGHT_EYE_POS = (CROPPED_IMAGE_HEIGHT // 5, CROPPED_IMAGE_WIDTH // 4 - 1)
+LEFT_EYE_POS = (CROPPED_IMAGE_HEIGHT // 5, CROPPED_IMAGE_WIDTH // 4 * 3)
+
+face_cropper = bob.bio.face.preprocessor.FaceCrop(
+ cropped_image_size = (CROPPED_IMAGE_HEIGHT, CROPPED_IMAGE_WIDTH),
+ cropped_positions = {'leye' : LEFT_EYE_POS, 'reye' : RIGHT_EYE_POS},
+ color_channel='rgb'
+)
+
+preprocessor = bob.bio.face.preprocessor.FaceDetect(face_cropper=face_cropper, color_channel='rgb')
+
+
diff --git a/bob/bio/face/extractor/Luan.py b/bob/bio/face/extractor/Luan.py
new file mode 100644
index 0000000000000000000000000000000000000000..7fe3559bed0abeee168aad275e9197a961319a3c
--- /dev/null
+++ b/bob/bio/face/extractor/Luan.py
@@ -0,0 +1,112 @@
+#!/usr/bin/env python
+# encoding: utf-8
+
+"""Features for face recognition"""
+
+import numpy
+import bob.io.base
+from bob.bio.base.extractor import Extractor
+
+import bob.io.base
+import tensorflow as tf
+import bob.ip.base
+
+from bob.learn.drgan.networks import DRGAN
+
+class LuanExtractor(Extractor):
+ """
+
+ **Parameters:**
+
+ """
+
+ def __init__(self):
+
+ Extractor.__init__(self, skip_extractor_training=True)
+
+ # not relevant (discriminator)
+ self.identity_dim = 208
+ self.conditional_dim = 13
+
+ # relevant and to be checked with the saved model
+ self.latent_dim = 320
+ self.image_size = [96, 96, 3]
+
+ import tensorflow as tf
+ self.session = tf.Session()
+
+ self.drgan = DRGAN(image_size=96, z_dim=100, gf_dim=32, gfc_dim=320, c_dim=3, checkpoint_dir='')
+ data_shape = [1] + self.image_size
+
+ self.X = tf.placeholder(tf.float32, shape=data_shape)
+ self.encode = self.drgan.generator_encoder(self.X, is_reuse=False, is_training=False)
+
+ self.saver = tf.train.Saver()
+
+
+ def __call__(self, image):
+ """__call__(image) -> feature
+
+ Extract features
+
+ **Parameters:**
+
+ image : 3D :py:class:`numpy.ndarray` (floats)
+ The image to extract the features from.
+
+ **Returns:**
+
+ feature : 2D :py:class:`numpy.ndarray` (floats)
+ The extracted features
+ """
+ def bob2skimage(bob_image):
+ """
+ Convert bob color image to the skcit image
+ """
+
+ if len(bob_image.shape) == 3:
+ skimage = numpy.zeros(shape=(bob_image.shape[1], bob_image.shape[2], 3), dtype=numpy.uint8)
+ skimage[:,:,0] = bob_image[0, :,:]
+ skimage[:,:,1] = bob_image[1, :,:]
+ skimage[:,:,2] = bob_image[2, :,:]
+ else:
+ skimage = numpy.zeros(shape=(bob_image.shape[0], bob_image.shape[1], 1))
+ skimage[:,:,0] = bob_image[:,:]
+ return skimage
+
+ def rescaleToUint8(image):
+ result = numpy.zeros_like(image)
+ for channel in range(image.shape[2]):
+ min_image = numpy.min(image[:, :, channel])
+ max_image = numpy.max(image[:, :, channel])
+ if (max_image - min_image) != 0:
+ result[:, :, channel] = 255.0*((image[:, :, channel] - min_image) / (max_image - min_image))
+ else:
+ result[:, :, channel] = 0
+ result = result.astype('uint8')
+ return result
+
+ # encode the provided image
+ image = rescaleToUint8(image)
+ #from matplotlib import pyplot
+ #pyplot.imshow(numpy.rollaxis(image, 0, 3))
+ #pyplot.show()
+ #
+ image = bob2skimage(image)
+ image = numpy.array(image/127.5 - 1).astype(numpy.float32)
+
+ shape = [1] + list(image.shape)
+ img = numpy.reshape(image, tuple(shape))
+ #
+ #pyplot.imshow(image)
+ #pyplot.show()
+
+
+ encoded_id = self.session.run(self.encode, feed_dict={self.X : img})
+ return encoded_id
+
+ # re-define the train function to get it non-documented
+ def train(*args, **kwargs): raise NotImplementedError("This function is not implemented and should not be called.")
+
+ def load(self, extractor_file):
+ self.saver.restore(self.session, extractor_file)
diff --git a/bob/bio/face/extractor/__init__.py b/bob/bio/face/extractor/__init__.py
index b1b7826d09d9f3e5d8b24d428cbebebf3aa5562f..298d936f13aa628b8129dbeabd0a631341830292 100644
--- a/bob/bio/face/extractor/__init__.py
+++ b/bob/bio/face/extractor/__init__.py
@@ -2,6 +2,7 @@ from .DCTBlocks import DCTBlocks
from .GridGraph import GridGraph
from .LGBPHS import LGBPHS
from .Eigenface import Eigenface
+from .Luan import LuanExtractor
# gets sphinx autodoc done right - don't remove it
def __appropriate__(*args):
@@ -22,5 +23,6 @@ __appropriate__(
GridGraph,
LGBPHS,
Eigenface,
+ LuanExtractor,
)
__all__ = [_ for _ in dir() if not _.startswith('_')]
diff --git a/setup.py b/setup.py
index e55963a87f982d6d2051496c274190bcb0060452..5b12961cafb9b6e5b042e01887cc08acc0604fc8 100644
--- a/setup.py
+++ b/setup.py
@@ -163,6 +163,8 @@ setup(
# histogram equalization w/o face-crop
'self-quotient = bob.bio.face.config.preprocessor.self_quotient_image:preprocessor_no_crop',
# self quotient image w/o face-crop
+
+ 'drgan = bob.bio.face.config.preprocessor.luan:preprocessor',
],
'bob.bio.extractor': [
@@ -170,6 +172,7 @@ setup(
'grid-graph = bob.bio.face.config.extractor.grid_graph:extractor', # Grid graph
'lgbphs = bob.bio.face.config.extractor.lgbphs:extractor', # LGBPHS
'eigenface = bob.bio.face.config.extractor.eigenface:extractor', # Eigenface
+ 'luan = bob.bio.face.config.extractor.luan:extractor',
],
'bob.bio.algorithm': [