Skip to content
Snippets Groups Projects
Commit 6fc98cc5 authored by Tiago de Freitas Pereira's avatar Tiago de Freitas Pereira
Browse files

IMproved documentation

parent d844c8f2
No related branches found
Tags v3.0.0
1 merge request!133Facexzoomodels
Hide whitespace changes
Inline Side-by-side
bob/bio/face/embeddings/pytorch.py
+ 82
− 28
View file @ 6fc98cc5
......@@ -20,6 +20,7 @@ from bob.bio.face.annotator import BobIpMTCNN
from bob.learn.pytorch.architectures.facexzoo import FaceXZooModelFactory
class PyTorchModel(TransformerMixin, BaseEstimator):
"""
Base Transformer using pytorch models
......@@ -272,13 +273,13 @@ class IResnet100(PyTorchModel):
)
def _load_model(self):
model = imp.load_source("module", self.config).iresnet100(self.checkpoint_path)
self.model = model
self.model.eval()
self.place_model_on_device()
class FaceXZooModel(PyTorchModel):
"""
FaceXZoo models
......@@ -289,17 +290,17 @@ class FaceXZooModel(PyTorchModel):
preprocessor=lambda x: (x - 127.5) / 128.0,
memory_demanding=False,
device=None,
arch='AttentionNet',
arch="AttentionNet",
**kwargs,
):
self.arch=arch
_model= FaceXZooModelFactory(self.arch)
self.arch = arch
_model = FaceXZooModelFactory(self.arch)
filename = _model.get_facexzoo_file()
checkpoint_name = _model.get_checkpoint_name()
config=None
config = None
path = os.path.dirname(filename)
checkpoint_path = os.path.join(path, self.arch+'.pt')
checkpoint_path = os.path.join(path, self.arch + ".pt")
super(FaceXZooModel, self).__init__(
checkpoint_path,
......@@ -312,28 +313,28 @@ class FaceXZooModel(PyTorchModel):
def _load_model(self):
_model= FaceXZooModelFactory(self.arch)
_model = FaceXZooModelFactory(self.arch)
self.model = _model.get_model()
model_dict = self.model.state_dict()
pretrained_dict = torch.load(self.checkpoint_path, map_location=torch.device('cpu'))['state_dict']
pretrained_dict_keys=pretrained_dict.keys()
model_dict_keys=model_dict.keys()
pretrained_dict = torch.load(
self.checkpoint_path, map_location=torch.device("cpu")
)["state_dict"]
pretrained_dict_keys = pretrained_dict.keys()
model_dict_keys = model_dict.keys()
new_pretrained_dict = {}
for k in model_dict:
new_pretrained_dict[k] = pretrained_dict['backbone.'+k]
new_pretrained_dict[k] = pretrained_dict["backbone." + k]
model_dict.update(new_pretrained_dict)
self.model.load_state_dict(model_dict)
self.model.eval()
self.place_model_on_device()
def iresnet_template(embedding, annotation_type, fixed_positions=None):
# DEFINE CROPPING
cropped_image_size = (112, 112)
......@@ -365,14 +366,18 @@ def iresnet_template(embedding, annotation_type, fixed_positions=None):
return VanillaBiometricsPipeline(transformer, algorithm)
def AttentionNet(annotation_type, fixed_positions=None, memory_demanding=False):
"""
Get the AttentionNet pipeline which will crop the face :math:`112 \times 112` and
use the :py:class:`AttentionNet` to extract the features
.. warning::
If you are at Idiap, please use the option `-l sge-gpu` while running the `vanilla-biometrics` pipeline.
Parameters
----------
......@@ -385,20 +390,23 @@ def AttentionNet(annotation_type, fixed_positions=None, memory_demanding=False):
memory_demanding: bool
"""
return iresnet_template(
embedding=FaceXZooModel(arch='AttentionNet', memory_demanding=memory_demanding),
embedding=FaceXZooModel(arch="AttentionNet", memory_demanding=memory_demanding),
annotation_type=annotation_type,
fixed_positions=fixed_positions,
)
def ResNeSt(annotation_type, fixed_positions=None, memory_demanding=False):
"""
Get the ResNeSt pipeline which will crop the face :math:`112 \times 112` and
use the :py:class:`ResNeSt` to extract the features
.. warning::
If you are at Idiap, please use the option `-l sge-gpu` while running the `vanilla-biometrics` pipeline.
Parameters
----------
......@@ -414,7 +422,7 @@ def ResNeSt(annotation_type, fixed_positions=None, memory_demanding=False):
"""
return iresnet_template(
embedding=FaceXZooModel(arch='ResNeSt', memory_demanding=memory_demanding),
embedding=FaceXZooModel(arch="ResNeSt", memory_demanding=memory_demanding),
annotation_type=annotation_type,
fixed_positions=fixed_positions,
)
......@@ -425,6 +433,11 @@ def MobileFaceNet(annotation_type, fixed_positions=None, memory_demanding=False)
Get the MobileFaceNet pipeline which will crop the face :math:`112 \times 112` and
use the :py:class:`MobileFaceNet` to extract the features
.. warning::
If you are at Idiap, please use the option `-l sge-gpu` while running the `vanilla-biometrics` pipeline.
Parameters
----------
......@@ -440,17 +453,25 @@ def MobileFaceNet(annotation_type, fixed_positions=None, memory_demanding=False)
"""
return iresnet_template(
embedding=FaceXZooModel(arch='MobileFaceNet', memory_demanding=memory_demanding),
embedding=FaceXZooModel(
arch="MobileFaceNet", memory_demanding=memory_demanding
),
annotation_type=annotation_type,
fixed_positions=fixed_positions,
)
def ResNet(annotation_type, fixed_positions=None, memory_demanding=False):
"""
Get the ResNet pipeline which will crop the face :math:`112 \times 112` and
use the :py:class:`ResNet` to extract the features
.. warning::
If you are at Idiap, please use the option `-l sge-gpu` while running the `vanilla-biometrics` pipeline.
Parameters
----------
......@@ -465,17 +486,24 @@ def ResNet(annotation_type, fixed_positions=None, memory_demanding=False):
"""
return iresnet_template(
embedding=FaceXZooModel(arch='ResNet', memory_demanding=memory_demanding),
embedding=FaceXZooModel(arch="ResNet", memory_demanding=memory_demanding),
annotation_type=annotation_type,
fixed_positions=fixed_positions,
)
def EfficientNet(annotation_type, fixed_positions=None, memory_demanding=False):
"""
Get the EfficientNet pipeline which will crop the face :math:`112 \times 112` and
use the :py:class:`EfficientNet` to extract the features
.. warning::
If you are at Idiap, please use the option `-l sge-gpu` while running the `vanilla-biometrics` pipeline.
Parameters
----------
......@@ -490,7 +518,7 @@ def EfficientNet(annotation_type, fixed_positions=None, memory_demanding=False):
"""
return iresnet_template(
embedding=FaceXZooModel(arch='EfficientNet', memory_demanding=memory_demanding),
embedding=FaceXZooModel(arch="EfficientNet", memory_demanding=memory_demanding),
annotation_type=annotation_type,
fixed_positions=fixed_positions,
)
......@@ -502,6 +530,12 @@ def TF_NAS(annotation_type, fixed_positions=None, memory_demanding=False):
use the :py:class:`TF-NAS` to extract the features
.. warning::
If you are at Idiap, please use the option `-l sge-gpu` while running the `vanilla-biometrics` pipeline.
Parameters
----------
......@@ -516,17 +550,23 @@ def TF_NAS(annotation_type, fixed_positions=None, memory_demanding=False):
"""
return iresnet_template(
embedding=FaceXZooModel(arch='TF-NAS', memory_demanding=memory_demanding),
embedding=FaceXZooModel(arch="TF-NAS", memory_demanding=memory_demanding),
annotation_type=annotation_type,
fixed_positions=fixed_positions,
)
def HRNet(annotation_type, fixed_positions=None, memory_demanding=False):
"""
Get the HRNet pipeline which will crop the face :math:`112 \times 112` and
use the :py:class:`HRNet` to extract the features
.. warning::
If you are at Idiap, please use the option `-l sge-gpu` while running the `vanilla-biometrics` pipeline.
Parameters
----------
......@@ -541,16 +581,22 @@ def HRNet(annotation_type, fixed_positions=None, memory_demanding=False):
"""
return iresnet_template(
embedding=FaceXZooModel(arch='HRNet', memory_demanding=memory_demanding),
embedding=FaceXZooModel(arch="HRNet", memory_demanding=memory_demanding),
annotation_type=annotation_type,
fixed_positions=fixed_positions,
)
def ReXNet(annotation_type, fixed_positions=None, memory_demanding=False):
"""
Get the ReXNet pipeline which will crop the face :math:`112 \times 112` and
use the :py:class:`ReXNet` to extract the features
.. warning::
If you are at Idiap, please use the option `-l sge-gpu` while running the `vanilla-biometrics` pipeline.
Parameters
----------
......@@ -566,17 +612,23 @@ def ReXNet(annotation_type, fixed_positions=None, memory_demanding=False):
"""
return iresnet_template(
embedding=FaceXZooModel(arch='ReXNet', memory_demanding=memory_demanding),
embedding=FaceXZooModel(arch="ReXNet", memory_demanding=memory_demanding),
annotation_type=annotation_type,
fixed_positions=fixed_positions,
)
def GhostNet(annotation_type, fixed_positions=None, memory_demanding=False):
"""
Get the GhostNet pipeline which will crop the face :math:`112 \times 112` and
use the :py:class:`GhostNet` to extract the features
.. warning::
If you are at Idiap, please use the option `-l sge-gpu` while running the `vanilla-biometrics` pipeline.
Parameters
----------
......@@ -591,10 +643,12 @@ def GhostNet(annotation_type, fixed_positions=None, memory_demanding=False):
"""
return iresnet_template(
embedding=FaceXZooModel(arch='GhostNet', memory_demanding=memory_demanding),
embedding=FaceXZooModel(arch="GhostNet", memory_demanding=memory_demanding),
annotation_type=annotation_type,
fixed_positions=fixed_positions,
)
def iresnet34(annotation_type, fixed_positions=None, memory_demanding=False):
"""
Get the Resnet34 pipeline which will crop the face :math:`112 \times 112` and
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment