diff --git a/bob/bio/demographics/datasets.py b/bob/bio/demographics/datasets.py
index 2171d0600d15ade8f8f24fea934ba8c315b09251..317ba16288936eff653d78dc159174c1d809ff45 100644
--- a/bob/bio/demographics/datasets.py
+++ b/bob/bio/demographics/datasets.py
@@ -78,12 +78,44 @@ class MorphTorchDataset(DemoraphicTorchDataset):
self, protocol, database_path, database_extension=".h5", transform=None
):
- bob_dataset = MorphDatabase(
+ self.bob_dataset = MorphDatabase(
protocol=protocol,
dataset_original_directory=database_path,
dataset_original_extension=database_extension,
)
- super().__init__(bob_dataset, transform=transform)
+
+ # Morph dataset has an intersection in between zprobes and treferences
+ self.excluding_list = [
+ "190276",
+ "332158",
+ "111942",
+ "308129",
+ "334074",
+ "350814",
+ "131677",
+ "168724",
+ "276055",
+ "275589",
+ "286810",
+ ]
+
+ self.bucket = [s for sset in self.bob_dataset.zprobes() for s in sset]
+ self.bucket += [
+ s
+ for sset in self.bob_dataset.treferences()
+ for s in sset
+ if sset.subject_id not in self.excluding_list
+ ]
+
+ # Defining keys and labels
+ keys = [b.subject_id for b in self.bucket]
+
+ self.labels = dict(zip(keys, range(len(keys))))
+
+ self.demographic_keys = self.load_demographics()
+ self.transform = transform
+
+ # super().__init__(bob_dataset, transform=transform)
def load_demographics(self):
@@ -91,7 +123,11 @@ class MorphTorchDataset(DemoraphicTorchDataset):
metadata_keys = set(
[f"{sset.rac}-{sset.sex}" for sset in self.bob_dataset.zprobes()]
- + [f"{sset.rac}-{sset.sex}" for sset in self.bob_dataset.treferences()]
+ + [
+ f"{sset.rac}-{sset.sex}"
+ for sset in self.bob_dataset.treferences()
+ if sset.subject_id not in self.excluding_list
+ ]
)
metadata_keys = dict(zip(metadata_keys, range(len(metadata_keys))))
return metadata_keys
diff --git a/bob/bio/demographics/experiments/__init__.py b/bob/bio/demographics/experiments/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/bob/bio/demographics/experiments/evaluation/ortogonality/meds.py b/bob/bio/demographics/experiments/evaluation/ortogonality/meds.py
new file mode 100644
index 0000000000000000000000000000000000000000..d65a5f99efc4dfd4912a6bb7b961e4f3478cc1d6
--- /dev/null
+++ b/bob/bio/demographics/experiments/evaluation/ortogonality/meds.py
@@ -0,0 +1,78 @@
+import click
+from bob.bio.face.embeddings.pytorch import PyTorchModel, iresnet_template
+from bob.bio.demographics.fair_transformers import RunnableTransformer
+from functools import partial
+import os
+from bob.extension.scripts.click_helper import ResourceOption
+from bob.bio.base.pipelines.vanilla_biometrics import checkpoint_vanilla_biometrics
+from bob.bio.base.pipelines.vanilla_biometrics import execute_vanilla_biometrics
+from bob.pipelines.distributed import VALID_DASK_CLIENT_STRINGS
+from bob.bio.demographics.regularizers import AVAILABLE_BACKBONES
+
+# BACKBONES = dict()
+# BACKBONES["iresnet100"] = iresnet100
+
+
+@click.command()
+@click.argument("BACKBONE")
+@click.argument("CHECKPOINT-PATH")
+@click.argument("EXPERIMENT-PATH")
+@click.option(
+ "--dask-client",
+ "-l",
+ entry_point_group="dask.client",
+ string_exceptions=VALID_DASK_CLIENT_STRINGS,
+ default="single-threaded",
+ help="Dask client for the execution of the pipeline.",
+ cls=ResourceOption,
+)
+def meds_experiment(backbone, checkpoint_path, experiment_path, dask_client, **kwargs):
+
+ annotation_type = "eyes-center"
+ fixed_positions = None
+ memory_demanding = False
+
+ backbone = AVAILABLE_BACKBONES[backbone]["structure"]
+
+ # Loading the pipeline
+ pipeline = iresnet_template(
+ embedding=RunnableTransformer(
+ partial(backbone, pretrained=checkpoint_path),
+ memory_demanding=memory_demanding,
+ ),
+ annotation_type=annotation_type,
+ fixed_positions=fixed_positions,
+ )
+
+ face_crop_path = os.path.join(experiment_path, "../face-crop/")
+
+ # Checkpointing the pipeline
+ pipeline = checkpoint_vanilla_biometrics(pipeline, base_dir=experiment_path)
+
+ pipeline.transformer[0].features_dir = face_crop_path
+ dask_partition_size = 100
+
+ #### DATABASE
+
+ from bob.bio.face.database import MEDSDatabase
+
+ protocol = "verification_fold1"
+ database = MEDSDatabase(protocol=protocol)
+
+ groups = ["dev", "eval"]
+ execute_vanilla_biometrics(
+ pipeline,
+ database,
+ dask_client,
+ groups,
+ experiment_path,
+ write_metadata_scores=True,
+ checkpoint=False,
+ dask_partition_size=dask_partition_size,
+ dask_n_workers=10,
+ allow_scoring_with_all_biometric_references=True,
+ )
+
+
+if __name__ == "__main__":
+ meds_experiment()
diff --git a/bob/bio/demographics/experiments/evaluation/ortogonality/morph.py b/bob/bio/demographics/experiments/evaluation/ortogonality/morph.py
new file mode 100644
index 0000000000000000000000000000000000000000..3ed5fce47572afaac01162f3d27fd14427b925a4
--- /dev/null
+++ b/bob/bio/demographics/experiments/evaluation/ortogonality/morph.py
@@ -0,0 +1,81 @@
+import click
+from bob.bio.face.embeddings.pytorch import PyTorchModel, iresnet_template
+
+# from bob.learn.pytorch.architectures.iresnet import iresnet100
+from bob.bio.demographics.fair_transformers import RunnableTransformer
+from functools import partial
+import os
+from bob.extension.scripts.click_helper import ResourceOption
+from bob.bio.base.pipelines.vanilla_biometrics import checkpoint_vanilla_biometrics
+from bob.bio.base.pipelines.vanilla_biometrics import execute_vanilla_biometrics
+from bob.pipelines.distributed import VALID_DASK_CLIENT_STRINGS
+
+from bob.bio.demographics.regularizers import AVAILABLE_BACKBONES
+
+
+@click.command()
+@click.argument("BACKBONE")
+@click.argument("CHECKPOINT-PATH")
+@click.argument("EXPERIMENT-PATH")
+@click.option(
+ "--dask-client",
+ "-l",
+ entry_point_group="dask.client",
+ string_exceptions=VALID_DASK_CLIENT_STRINGS,
+ default="single-threaded",
+ help="Dask client for the execution of the pipeline.",
+ cls=ResourceOption,
+)
+def morph_experiment(backbone, checkpoint_path, experiment_path, dask_client, **kwargs):
+
+ annotation_type = "eyes-center"
+ fixed_positions = None
+ memory_demanding = False
+
+ # checkpoint_path = "/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality_hypothesis/meds/iresnet100.pth"
+ # checkpoint_path = "/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality_hypothesis/meds_identity-10.0_orthogonality-1.0/iresnet100.pth"
+
+ backbone = AVAILABLE_BACKBONES[backbone]["structure"]
+
+ # Loading the pipeline
+ pipeline = iresnet_template(
+ embedding=RunnableTransformer(
+ partial(backbone, pretrained=checkpoint_path),
+ memory_demanding=memory_demanding,
+ ),
+ annotation_type=annotation_type,
+ fixed_positions=fixed_positions,
+ )
+
+ face_crop_path = os.path.join(experiment_path, "../face-crop/")
+
+ # Checkpointing the pipeline
+ pipeline = checkpoint_vanilla_biometrics(pipeline, base_dir=experiment_path)
+
+ pipeline.transformer[0].features_dir = face_crop_path
+ dask_partition_size = 100
+
+ #### DATABASE
+
+ from bob.bio.face.database import MorphDatabase
+
+ protocol = "verification_fold1"
+ database = MorphDatabase(protocol=protocol)
+
+ groups = ["dev", "eval"]
+ execute_vanilla_biometrics(
+ pipeline,
+ database,
+ dask_client,
+ groups,
+ experiment_path,
+ write_metadata_scores=True,
+ checkpoint=False,
+ dask_partition_size=dask_partition_size,
+ dask_n_workers=10,
+ allow_scoring_with_all_biometric_references=True,
+ )
+
+
+if __name__ == "__main__":
+ morph_experiment()
diff --git a/bob/bio/demographics/experiments/evaluation/ortogonality_hypothesis/meds.py b/bob/bio/demographics/experiments/evaluation/ortogonality_hypothesis/meds.py
deleted file mode 100644
index 12ae523a690626c1aace7afabee4d7e2a3ffb676..0000000000000000000000000000000000000000
--- a/bob/bio/demographics/experiments/evaluation/ortogonality_hypothesis/meds.py
+++ /dev/null
@@ -1,32 +0,0 @@
-from bob.bio.face.embeddings.pytorch import PyTorchModel, iresnet_template
-from bob.learn.pytorch.architectures.iresnet import iresnet100
-from bob.bio.demographics.fair_transformers import RunnableTransformer
-from functools import partial
-
-annotation_type = "eyes-center"
-fixed_positions = None
-memory_demanding = False
-
-# checkpoint_path = "/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality_hypothesis/meds/iresnet100.pth"
-checkpoint_path = "/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality_hypothesis/meds_identity-10.0_orthogonality-1.0/iresnet100.pth"
-
-pipeline = iresnet_template(
- embedding=RunnableTransformer(
- partial(iresnet100, pretrained=checkpoint_path),
- memory_demanding=memory_demanding,
- ),
- annotation_type=annotation_type,
- fixed_positions=fixed_positions,
-)
-
-
-#### DATABASE
-
-from bob.bio.face.database import MEDSDatabase
-
-protocol = "verification_fold1"
-database = MEDSDatabase(protocol=protocol)
-
-# output = (
-# "/remote/idiap.svm/user.active/tpereira/gitlab/bob/bob.nightlies/vanilla-callback"
-# )
diff --git a/bob/bio/demographics/experiments/train/__init__.py b/bob/bio/demographics/experiments/train/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/bob/bio/demographics/experiments/train/mine/__init__.py b/bob/bio/demographics/experiments/train/mine/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/bob/bio/demographics/experiments/train/mine/meds.py b/bob/bio/demographics/experiments/train/mine/meds.py
new file mode 100644
index 0000000000000000000000000000000000000000..afa6441b4b4c8c538e7354c6af0e3c56472dd52e
--- /dev/null
+++ b/bob/bio/demographics/experiments/train/mine/meds.py
@@ -0,0 +1,93 @@
+from bob.bio.demographics.datasets import MedsTorchDataset, MorphTorchDataset
+
+# https://pytorch.org/docs/stable/data.html
+from torch.utils.data import DataLoader
+from bob.extension import rc
+import os
+
+import bob.io.image
+
+import torch
+from functools import partial
+
+import torchvision.transforms as transforms
+import click
+import yaml
+
+
+from bob.bio.demographics.regularizers.trainers import mine_trainer
+
+
+@click.command()
+@click.argument("OUTPUT_DIR")
+@click.option("--identity-factor", default=1.0, help="Identity factor")
+@click.option("--mine-factor", default=1.0, help="MINE factor")
+@click.option("--max-epochs", default=600, help="Max number of epochs")
+@click.option(
+ "--demographic-epochs",
+ default=100,
+ help="Number of epochs to train the demographic classifier",
+)
+@click.option(
+ "--identity-epochs",
+ default=200,
+ help="Number of epochs to train the identity classifier",
+)
+@click.option("--batch-size", default=64, help="Batch size")
+@click.option("--backbone", default="iresnet100", help="Backbone")
+def mine_meds(
+ output_dir,
+ identity_factor,
+ mine_factor,
+ max_epochs,
+ demographic_epochs,
+ identity_epochs,
+ batch_size,
+ backbone,
+):
+
+ from bob.bio.demographics.regularizers import AVAILABLE_BACKBONES
+
+ database_path = os.path.join(
+ rc.get("bob.bio.demographics.directory"), "meds", "samplewrapper"
+ )
+
+ transform = transforms.Compose(
+ [
+ lambda x: bob.io.image.to_matplotlib(x.astype("float32")),
+ # transforms.ToPILImage(mode="RGB"),
+ # transforms.RandomHorizontalFlip(p=0.5),
+ # transforms.RandomRotation(degrees=(-3, 3)),
+ # transforms.RandomAutocontrast(p=0.1),
+ transforms.ToTensor(),
+ lambda x: (x - 127.5) / 128.0,
+ ]
+ )
+
+ dataset = MedsTorchDataset(
+ protocol="verification_fold1", database_path=database_path, transform=transform,
+ )
+
+ train_dataloader = DataLoader(
+ dataset, batch_size=batch_size, shuffle=True, pin_memory=True, num_workers=2
+ )
+ # train_dataloader = DataLoader(dataset, batch_size=64, shuffle=True)
+
+ backbone_model = AVAILABLE_BACKBONES[backbone]()
+
+ mine_trainer(
+ output_dir,
+ identity_factor,
+ mine_factor,
+ max_epochs,
+ demographic_epochs,
+ identity_epochs,
+ batch_size,
+ train_dataloader,
+ backbone_model,
+ transform,
+ )
+
+
+if __name__ == "__main__":
+ mine_meds()
diff --git a/bob/bio/demographics/experiments/train/ortogonality/__init__.py b/bob/bio/demographics/experiments/train/ortogonality/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/bob/bio/demographics/experiments/train/ortogonality/meds.py b/bob/bio/demographics/experiments/train/ortogonality/meds.py
new file mode 100644
index 0000000000000000000000000000000000000000..a5d14dbe67f8b1609166aa53995d49283928a5a5
--- /dev/null
+++ b/bob/bio/demographics/experiments/train/ortogonality/meds.py
@@ -0,0 +1,93 @@
+from bob.bio.demographics.datasets import MedsTorchDataset, MorphTorchDataset
+
+# https://pytorch.org/docs/stable/data.html
+from torch.utils.data import DataLoader
+from bob.extension import rc
+import os
+
+import bob.io.image
+
+import torch
+from functools import partial
+
+import torchvision.transforms as transforms
+import click
+import yaml
+
+
+from bob.bio.demographics.regularizers.trainers import ortogonality_trainer
+
+
+@click.command()
+@click.argument("OUTPUT_DIR")
+@click.option("--identity-factor", default=1.0, help="Identity factor")
+@click.option("--orthogonality-factor", default=1.0, help="Ortogonality factor")
+@click.option("--max-epochs", default=600, help="Max number of epochs")
+@click.option(
+ "--demographic-epochs",
+ default=100,
+ help="Number of epochs to train the demographic classifier",
+)
+@click.option(
+ "--identity-epochs",
+ default=200,
+ help="Number of epochs to train the identity classifier",
+)
+@click.option("--batch-size", default=64, help="Batch size")
+@click.option("--backbone", default="iresnet100", help="Backbone")
+def ortogonality_meds(
+ output_dir,
+ identity_factor,
+ orthogonality_factor,
+ max_epochs,
+ demographic_epochs,
+ identity_epochs,
+ batch_size,
+ backbone,
+):
+
+ from bob.bio.demographics.regularizers import AVAILABLE_BACKBONES
+
+ database_path = os.path.join(
+ rc.get("bob.bio.demographics.directory"), "meds", "samplewrapper"
+ )
+
+ transform = transforms.Compose(
+ [
+ lambda x: bob.io.image.to_matplotlib(x.astype("float32")),
+ # transforms.ToPILImage(mode="RGB"),
+ # transforms.RandomHorizontalFlip(p=0.5),
+ # transforms.RandomRotation(degrees=(-3, 3)),
+ # transforms.RandomAutocontrast(p=0.1),
+ transforms.ToTensor(),
+ lambda x: (x - 127.5) / 128.0,
+ ]
+ )
+
+ dataset = MedsTorchDataset(
+ protocol="verification_fold1", database_path=database_path, transform=transform,
+ )
+
+ train_dataloader = DataLoader(
+ dataset, batch_size=batch_size, shuffle=True, pin_memory=True, num_workers=2
+ )
+ # train_dataloader = DataLoader(dataset, batch_size=64, shuffle=True)
+
+ backbone_model = AVAILABLE_BACKBONES[backbone]()
+
+ ortogonality_trainer(
+ output_dir,
+ identity_factor,
+ orthogonality_factor,
+ max_epochs,
+ demographic_epochs,
+ identity_epochs,
+ batch_size,
+ train_dataloader,
+ backbone_model,
+ transform,
+ )
+
+
+if __name__ == "__main__":
+ ortogonality_meds()
diff --git a/bob/bio/demographics/experiments/train/ortogonality/morph.py b/bob/bio/demographics/experiments/train/ortogonality/morph.py
new file mode 100644
index 0000000000000000000000000000000000000000..8e9c97e2de8e8433bed8490670bdf31683332afd
--- /dev/null
+++ b/bob/bio/demographics/experiments/train/ortogonality/morph.py
@@ -0,0 +1,93 @@
+from bob.bio.demographics.datasets import MedsTorchDataset, MorphTorchDataset
+
+# https://pytorch.org/docs/stable/data.html
+from torch.utils.data import DataLoader
+from bob.extension import rc
+import os
+
+import bob.io.image
+
+import torch
+from functools import partial
+
+import torchvision.transforms as transforms
+import click
+import yaml
+
+
+from bob.bio.demographics.regularizers.trainers import ortogonality_trainer
+
+
+@click.command()
+@click.argument("OUTPUT_DIR")
+@click.option("--identity-factor", default=1.0, help="Identity factor")
+@click.option("--orthogonality-factor", default=1.0, help="Ortogonality factor")
+@click.option("--max-epochs", default=600, help="Max number of epochs")
+@click.option(
+ "--demographic-epochs",
+ default=100,
+ help="Number of epochs to train the demographic classifier",
+)
+@click.option(
+ "--identity-epochs",
+ default=200,
+ help="Number of epochs to train the identity classifier",
+)
+@click.option("--batch-size", default=64, help="Batch size")
+@click.option("--backbone", default="iresnet100", help="Backbone")
+def ortogonality_morph(
+ output_dir,
+ identity_factor,
+ orthogonality_factor,
+ max_epochs,
+ demographic_epochs,
+ identity_epochs,
+ batch_size,
+ backbone,
+):
+
+ from bob.bio.demographics.regularizers import AVAILABLE_BACKBONES
+
+ database_path = os.path.join(
+ rc.get("bob.bio.demographics.directory"), "morph", "samplewrapper"
+ )
+
+ transform = transforms.Compose(
+ [
+ lambda x: bob.io.image.to_matplotlib(x.astype("float32")),
+ # transforms.ToPILImage(mode="RGB"),
+ # transforms.RandomHorizontalFlip(p=0.5),
+ # transforms.RandomRotation(degrees=(-3, 3)),
+ # transforms.RandomAutocontrast(p=0.1),
+ transforms.ToTensor(),
+ lambda x: (x - 127.5) / 128.0,
+ ]
+ )
+
+ dataset = MorphTorchDataset(
+ protocol="verification_fold1", database_path=database_path, transform=transform,
+ )
+
+ train_dataloader = DataLoader(
+ dataset, batch_size=batch_size, shuffle=True, pin_memory=True, num_workers=2
+ )
+ # train_dataloader = DataLoader(dataset, batch_size=64, shuffle=True)
+
+ backbone_model = AVAILABLE_BACKBONES[backbone]()
+
+ ortogonality_trainer(
+ output_dir,
+ identity_factor,
+ orthogonality_factor,
+ max_epochs,
+ demographic_epochs,
+ identity_epochs,
+ batch_size,
+ train_dataloader,
+ backbone_model,
+ transform,
+ )
+
+
+if __name__ == "__main__":
+ ortogonality_morph()
diff --git a/bob/bio/demographics/experiments/train/ortogonality_hypothesis/meds.py b/bob/bio/demographics/experiments/train/ortogonality_hypothesis/meds.py
deleted file mode 100644
index cf76727c1d694d54c750ebac0f8a3b66455b241f..0000000000000000000000000000000000000000
--- a/bob/bio/demographics/experiments/train/ortogonality_hypothesis/meds.py
+++ /dev/null
@@ -1,181 +0,0 @@
-from bob.bio.demographics.datasets import MedsTorchDataset, MorphTorchDataset
-
-# https://pytorch.org/docs/stable/data.html
-from torch.utils.data import DataLoader
-import pytest
-from bob.extension import rc
-import os
-from bob.learn.pytorch.architectures.iresnet import iresnet34, iresnet100
-from bob.learn.pytorch.head import ArcFace, Regular
-from bob.bio.demographics.regularizers.independence import (
- DemographicRegularHead,
- OrthogonalityModel,
-)
-import pytorch_lightning as pl
-import torch
-from functools import partial
-
-from pytorch_lightning.callbacks import ModelCheckpoint
-from pytorch_lightning.loggers import TensorBoardLogger
-import torchvision.transforms as transforms
-import click
-import yaml
-
-
-# demographic_epochs = 50
-# identity_epochs = 200
-
-
-@click.command()
-@click.argument("OUTPUT_DIR")
-@click.option("--identity-factor", default=1.0, help="Identity factor")
-@click.option("--orthogonality-factor", default=1.0, help="Ortogonality factor")
-@click.option("--max-epochs", default=600, help="Max number of epochs")
-@click.option(
- "--demographic-epochs",
- default=100,
- help="Number of epochs to train the demographic classifier",
-)
-@click.option(
- "--identity-epochs",
- default=200,
- help="Number of epochs to train the identity classifier",
-)
-@click.option("--batch-size", default=64, help="Batch size")
-def ortogonality_meds(
- output_dir,
- identity_factor,
- orthogonality_factor,
- max_epochs,
- demographic_epochs,
- identity_epochs,
- batch_size,
-):
-
- os.makedirs(output_dir, exist_ok=True)
- with open(f"{output_dir}/config.yml", "w") as file:
- dict_file = dict()
- dict_file["demographic_epochs"] = demographic_epochs
- dict_file["identity_epochs"] = identity_epochs
- dict_file["max_epochs"] = max_epochs
- dict_file["batch_size"] = batch_size
- yaml.dump(dict_file, file)
-
- backbone_checkpoint_path = f"{output_dir}/iresnet100.pth"
- checkpoint_dir = f"{output_dir}/last.ckpt"
-
- database_path = os.path.join(
- rc.get("bob.bio.demographics.directory"), "meds", "samplewrapper"
- )
-
- import bob.io.image
-
- transform = transforms.Compose(
- [
- lambda x: bob.io.image.to_matplotlib(x),
- transforms.ToPILImage(mode="RGB"),
- transforms.RandomHorizontalFlip(p=0.5),
- transforms.RandomRotation(degrees=(-3, 3)),
- # transforms.RandomAutocontrast(p=0.1),
- transforms.ToTensor(),
- lambda x: (x - 127.5) / 128.0,
- ]
- )
-
- dataset = MedsTorchDataset(
- protocol="verification_fold1", database_path=database_path, transform=transform,
- )
-
- # train_dataloader = DataLoader(
- # dataset, batch_size=batch_size, shuffle=True, pin_memory=True, num_workers=2
- # )
- train_dataloader = DataLoader(dataset, batch_size=64, shuffle=True)
-
- # backbone = iresnet34(
- # pretrained="/idiap/temp/tpereira/bob/data/pytorch/iresnet-91a5de61/iresnet34-5b0d0e90.pth"
- # )
- # Add this argument
- backbone = iresnet100(
- pretrained="/idiap/temp/tpereira/bob/data/pytorch/iresnet-91a5de61/iresnet100-73e07ba7.pth"
- )
-
- # list(dataloader.dataset.labels.values())
- #####################
- ## IDENTITY
- num_class = len(list(train_dataloader.dataset.labels.values()))
- identity_head = ArcFace(
- feat_dim=backbone.features.num_features, num_class=num_class
- )
-
- ######################
- ## DEMOGRAPHIC
- num_class = len(list(train_dataloader.dataset.demographic_keys.values()))
- demographic_head = DemographicRegularHead(
- feat_dim=backbone.features.num_features, num_class=num_class
- )
-
- ################
- ## Trainer
- optimizer = partial(torch.optim.SGD, lr=0.001, momentum=0.9)
-
- # demographic_epochs = 50
- # identity_epochs = 200
- # ortogonality_epochs = 400
-
- # Preparing lightining model
- model = OrthogonalityModel(
- backbone=backbone,
- identity_head=identity_head,
- demographic_head=demographic_head,
- loss_fn=torch.nn.CrossEntropyLoss(),
- optimizer_fn=optimizer,
- identity_factor=identity_factor,
- orthogonality_factor=orthogonality_factor,
- backbone_checkpoint_path=backbone_checkpoint_path,
- demographic_epochs=demographic_epochs,
- identity_epochs=identity_epochs,
- )
-
- """
- from bob.bio.face.pytorch.callbacks import VanillaBiometricsCallback
-
-
- vanilla_callback = VanillaBiometricsCallback(
- config="/remote/idiap.svm/user.active/tpereira/gitlab/bob/bob.nightlies/src/bob.bio.demographics/bob/bio/demographics/fair_transformers/transformers.py",
- output_path="./vanilla-callback",
- )
- """
- model_checkpoint = ModelCheckpoint(
- output_dir, every_n_train_steps=100, save_last=True
- )
- logger = TensorBoardLogger(os.path.join(output_dir, "tb_logs"))
-
- # Be careful with
- # https://github.com/PyTorchLightning/pytorch-lightning/issues/5325
- resume_from_checkpoint = checkpoint_dir if os.path.exists(checkpoint_dir) else None
-
- # TODO: using this code to learn too
- # so, be nice with my comments
- # callbacks=[model_checkpoint, vanilla_callback],
- callbacks = [model_checkpoint]
- trainer = pl.Trainer(
- callbacks=callbacks,
- logger=logger,
- max_epochs=max_epochs,
- gpus=-1 if torch.cuda.is_available() else None,
- resume_from_checkpoint=resume_from_checkpoint,
- # resume_from_checkpoint=resume_from_checkpoint, #https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#resume-from-checkpoint
- # debug flags
- # limit_train_batches=10, # https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#limit-train-batches
- # limit_val_batches=1,
- amp_level="00", # https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#amp-level
- log_every_n_steps=5,
- )
-
- trainer.fit(
- model=model, train_dataloaders=train_dataloader,
- )
-
-
-if __name__ == "__main__":
- ortogonality_meds()
diff --git a/bob/bio/demographics/regularizers/__init__.py b/bob/bio/demographics/regularizers/__init__.py
index e69de29bb2d1d6434b8b29ae775ad8c2e48c5391..eca908cc0bf1a6b723564de153cdd92bf41a4388 100644
--- a/bob/bio/demographics/regularizers/__init__.py
+++ b/bob/bio/demographics/regularizers/__init__.py
@@ -0,0 +1,31 @@
+AVAILABLE_BACKBONES = dict()
+from bob.learn.pytorch.architectures.iresnet import iresnet34, iresnet100, iresnet50
+from functools import partial
+
+# Organize these checkpoints
+AVAILABLE_BACKBONES["iresnet100"] = {
+ "structure": iresnet100,
+ "prior": partial(
+ iresnet100,
+ "/idiap/temp/tpereira/bob/data/pytorch/iresnet-91a5de61/iresnet100-73e07ba7.pth",
+ ),
+}
+
+
+AVAILABLE_BACKBONES["iresnet50"] = {
+ "structure": iresnet50,
+ "prior": partial(
+ iresnet50,
+ "/idiap/temp/tpereira/bob/data/pytorch/iresnet-91a5de61/iresnet50-7f187506.pth",
+ ),
+}
+
+
+AVAILABLE_BACKBONES["iresnet34"] = {
+ "structure": iresnet34,
+ "prior": partial(
+ iresnet34,
+ "/idiap/temp/tpereira/bob/data/pytorch/iresnet-91a5de61/iresnet34-5b0d0e90.pth",
+ ),
+}
+
diff --git a/bob/bio/demographics/regularizers/independence.py b/bob/bio/demographics/regularizers/independence.py
index d8a84d358a96e4d0d84cd5b392f9ccfba73828cd..6653f711e0309116f94695018513eadbae2bdc90 100644
--- a/bob/bio/demographics/regularizers/independence.py
+++ b/bob/bio/demographics/regularizers/independence.py
@@ -4,8 +4,9 @@ import pytorch_lightning as pl
import torch
import torch.nn.functional as F
import numpy as np
-
+import torch.nn as nn
import copy
+import math
class DemographicRegularHead(Module):
@@ -42,7 +43,7 @@ def switch(model, flag):
class OrthogonalityModel(BackboneHeadModel):
"""
Here we hypothesize that the sensitive attribute is orthogonal
- to the identity
+ to the identity attribute
"""
def __init__(
@@ -110,9 +111,6 @@ class OrthogonalityModel(BackboneHeadModel):
# Switching of the backbone
if not self.demographic_switch:
- self.demographic_switch = True
-
- self.backbone = switch(self.backbone, False)
self.demographic_head = switch(
self.demographic_head, True
@@ -244,3 +242,325 @@ class OrthogonalityModel(BackboneHeadModel):
def configure_optimizers(self):
# optimizer = torch.optim.Adam(self.parameters(), lr=1e-3)
return self.optimizer_fn(params=self.parameters())
+
+
+class Mine(nn.Module):
+ """
+ Implements the MINE loss from the paper:
+
+ Mutual Information Neural Estimation (https://arxiv.org/pdf/1801.04062.pdf)
+
+ """
+
+ def __init__(self, T, alpha=0.01, is_mine_f=False):
+ super().__init__()
+ # self.running_mean = 0
+ # self.loss = loss
+ # self.alpha = alpha
+ # self.method = method
+ self.is_mine_f = is_mine_f
+ self.T = T
+
+ def forward(self, x, z, z_marg=None):
+ if z_marg is None:
+ z_marg = z[torch.randperm(x.shape[0])]
+
+ t = self.T(x, z).mean()
+ t_marg = self.T(x, z_marg)
+
+ if self.is_mine_f:
+ second_term = torch.exp(t_marg - 1).mean()
+ else:
+ second_term = torch.logsumexp(t_marg, 0) - math.log(t_marg.shape[0])
+
+ return -t + second_term
+
+ def mi(self, x, z, z_marg=None):
+ if isinstance(x, np.ndarray):
+ x = torch.from_numpy(x).float()
+ if isinstance(z, np.ndarray):
+ z = torch.from_numpy(z).float()
+
+ with torch.no_grad():
+ mi = -self.forward(x, z, z_marg)
+ return mi
+
+
+class MINEModel(BackboneHeadModel):
+ """
+ Here we hypothesize that the sensitive attribute is mutually independent
+ to the identity attribute
+ """
+
+ def __init__(
+ self,
+ backbone,
+ identity_head,
+ demographic_head,
+ loss_fn,
+ optimizer_fn,
+ identity_factor=1.0,
+ mine_factor=1.0,
+ backbone_checkpoint_path=None,
+ demographic_epochs=30,
+ identity_epochs=50,
+ mine_dimension=512,
+ **kwargs,
+ ):
+ # super(pl.LightningModule, self).__init__(**kwargs)
+
+ pl.LightningModule.__init__(self, **kwargs)
+ self.backbone = backbone
+ self.identity_head = identity_head
+ self.demographic_head = demographic_head
+ self.loss_fn = loss_fn
+ self.optimizer_fn = optimizer_fn
+
+ self.identity_factor = identity_factor
+ self.mine_factor = mine_factor
+
+ self.demographic_epochs = demographic_epochs # 1. First train demographics
+
+ self.identity_epochs = (
+ demographic_epochs + identity_epochs
+ ) # 2. Train identities
+
+ # self.ortogonality_epochs = (
+ # demographic_epochs + identity_epochs + ortogonality_epochs
+ # ) # 3. Train ortogonality
+
+ self.backbone_checkpoint_path = backbone_checkpoint_path
+ self.last_op = None
+
+ # Control the networks that will be updated
+ self.demographic_switch = False
+ self.identity_switch = False
+ self.mine_switch = False
+
+ # Defining the dimension of the MINE model
+ dim = (
+ self.backbone.features.num_features
+ if hasattr(self.backbone, "features")
+ else mine_dimension
+ )
+
+ # Get the backbone
+ self.mine = Mine(T(dim, dim))
+
+ def training_epoch_end(self, training_step_outputs):
+ if self.backbone_checkpoint_path:
+
+ state = self.backbone.state_dict()
+ torch.save(state, self.backbone_checkpoint_path)
+
+ pass
+
+ def training_step(self, batch, batch_idx):
+
+ data = batch["data"]
+ label = batch["label"]
+ demography = batch["demography"]
+
+ embedding = self.backbone(data)
+
+ if self.current_epoch < self.demographic_epochs:
+ ## First we learn the demography classifiers
+
+ # Switching of the backbone
+ if not self.demographic_switch:
+
+ self.demographic_head = switch(
+ self.demographic_head, True
+ ) # Update just he demographic classifier
+
+ self.identity_head = switch(
+ self.identity_head, False
+ ) # Update just he demographic classifier
+
+ # Demographic CLASSIFICATION loss
+ _, demographic_logits = self.demographic_head(embedding, demography)
+
+ loss_demography = self.loss_fn(demographic_logits, demography)
+ self.log("train/loss_demography", loss_demography)
+
+ acc = (
+ sum(
+ np.argmax(demographic_logits.cpu().detach().numpy(), axis=1)
+ == demography.cpu().detach().numpy()
+ )
+ / demography.shape[0]
+ )
+ self.log("train/acc_demography_before_MINE", acc)
+
+ return loss_demography
+
+ if self.current_epoch < self.identity_epochs:
+ ## Second the idenity classifier
+
+ if not self.identity_switch:
+ self.identity_switch = True
+
+ # Switching of the backbone
+ self.backbone = switch(self.backbone, False)
+
+ self.demographic_head = switch(
+ self.demographic_head, False
+ ) # Update just he demographic classifier
+
+ self.identity_head = switch(
+ self.identity_head, True
+ ) # Update just he demographic classifier
+
+ # Identity loss
+ logits_identiy = self.identity_head(embedding, label)
+ loss_identity = self.loss_fn(logits_identiy, label)
+ self.log("train/loss_identity", loss_identity)
+
+ acc = (
+ sum(
+ np.argmax(logits_identiy.cpu().detach().numpy(), axis=1)
+ == label.cpu().detach().numpy()
+ )
+ / label.shape[0]
+ )
+ self.log("train/acc_identity_before_MINE", acc)
+
+ return loss_identity
+
+ #########################################
+ # Now we learn the mine
+ #########################################
+
+ # Switching of the backbone
+ if not self.mine_switch:
+ self.mine_switch = True
+
+ self.backbone = switch(self.backbone, True)
+
+ self.demographic_head = switch(
+ self.demographic_head, False
+ ) # Update just he demographic classifier
+
+ self.identity_head = switch(
+ self.identity_head, True
+ ) # Update just he demographic classifier
+
+ # Identity loss
+ logits_identiy = self.identity_head(embedding, label)
+ loss_identity = self.loss_fn(logits_identiy, label)
+ # self.log("train/loss_identity", loss_identity)
+
+ # Demographic CLASSIFICATION loss
+ demographic_embeding, demographic_logits = self.demographic_head(
+ embedding, demography
+ )
+
+ # Demographic
+ acc = (
+ sum(
+ np.argmax(demographic_logits.cpu().detach().numpy(), axis=1)
+ == demography.cpu().detach().numpy()
+ )
+ / demography.shape[0]
+ )
+ self.log("train/acc_demography_after_MINE", acc)
+
+ # Identity
+ acc = (
+ sum(
+ np.argmax(logits_identiy.cpu().detach().numpy(), axis=1)
+ == label.cpu().detach().numpy()
+ )
+ / label.shape[0]
+ )
+ self.log("train/acc_identity_after_MINE", acc)
+
+ # MUTUAL INFORMATION LOSS
+ #
+ loss_mine = self.mine(embedding, demographic_embeding)
+
+ self.log("train/loss_MINE", loss_mine)
+
+ total_loss = self.identity_factor * loss_identity + self.mine_factor * loss_mine
+
+ # total_loss = loss_identity
+ self.log("train/total_loss", total_loss)
+
+ return total_loss
+
+ def configure_optimizers(self):
+ # optimizer = torch.optim.Adam(self.parameters(), lr=1e-3)
+ return self.optimizer_fn(params=self.parameters())
+
+
+class Mine(nn.Module):
+ """
+ Implements the MINE loss from the paper:
+
+ Mutual Information Neural Estimation (https://arxiv.org/pdf/1801.04062.pdf)
+
+ """
+
+ def __init__(self, T, alpha=0.01, is_mine_f=False):
+ super().__init__()
+ # self.running_mean = 0
+ # self.loss = loss
+ # self.alpha = alpha
+ # self.method = method
+ self.is_mine_f = is_mine_f
+ self.T = T
+
+ def forward(self, x, z, z_marg=None):
+ if z_marg is None:
+ z_marg = z[torch.randperm(x.shape[0])]
+
+ t = self.T(x, z).mean()
+ t_marg = self.T(x, z_marg)
+
+ if self.is_mine_f:
+ second_term = torch.exp(t_marg - 1).mean()
+ else:
+ second_term = torch.logsumexp(t_marg, 0) - math.log(t_marg.shape[0])
+
+ return -t + second_term
+
+ def mi(self, x, z, z_marg=None):
+ if isinstance(x, np.ndarray):
+ x = torch.from_numpy(x).float()
+ if isinstance(z, np.ndarray):
+ z = torch.from_numpy(z).float()
+
+ with torch.no_grad():
+ mi = -self.forward(x, z, z_marg)
+ return mi
+
+
+"""
+class ConcatLayer(nn.Module):
+ def __init__(self, dim=1):
+ super().__init__()
+ self.dim = dim
+
+ def forward(self, x, y):
+ return torch.cat((x, y), self.dim)
+"""
+
+
+class T(nn.Module):
+
+ # Base network for to train the mutial informatuib
+
+ def __init__(self, x_dim, z_dim):
+ super().__init__()
+ self.t = nn.Sequential(
+ nn.Linear(x_dim + z_dim, 100),
+ nn.ReLU(),
+ nn.Linear(100, 100),
+ nn.ReLU(),
+ nn.Linear(100, 1),
+ )
+
+ def forward(self, x, z):
+
+ x_z = torch.cat((x, z), 1)
+ return self.t(x_z)
diff --git a/bob/bio/demographics/regularizers/trainers.py b/bob/bio/demographics/regularizers/trainers.py
new file mode 100644
index 0000000000000000000000000000000000000000..5e84d272a637771840114e30d2495459d4c2c587
--- /dev/null
+++ b/bob/bio/demographics/regularizers/trainers.py
@@ -0,0 +1,239 @@
+from bob.bio.demographics.datasets import MedsTorchDataset, MorphTorchDataset
+
+# https://pytorch.org/docs/stable/data.html
+from torch.utils.data import DataLoader
+import pytest
+from bob.extension import rc
+import os
+from bob.learn.pytorch.architectures.iresnet import iresnet34, iresnet100
+from bob.learn.pytorch.head import ArcFace, Regular
+from bob.bio.demographics.regularizers.independence import (
+ DemographicRegularHead,
+ OrthogonalityModel,
+ MINEModel,
+)
+import pytorch_lightning as pl
+import torch
+from functools import partial
+
+from pytorch_lightning.callbacks import ModelCheckpoint
+from pytorch_lightning.loggers import TensorBoardLogger
+import torchvision.transforms as transforms
+import click
+import yaml
+
+
+def ortogonality_trainer(
+ output_dir,
+ identity_factor,
+ orthogonality_factor,
+ max_epochs,
+ demographic_epochs,
+ identity_epochs,
+ batch_size,
+ train_dataloader,
+ backbone_model,
+ transform,
+):
+
+ """
+ Trains a Pytorch CNN using the ortogonality hypothesis
+ """
+
+ # Defing the variables of the experiment, so we don't get lost
+ os.makedirs(output_dir, exist_ok=True)
+ with open(f"{output_dir}/config.yml", "w") as file:
+ dict_file = dict()
+ dict_file["demographic_epochs"] = demographic_epochs
+ dict_file["identity_epochs"] = identity_epochs
+ dict_file["max_epochs"] = max_epochs
+ dict_file["batch_size"] = batch_size
+ dict_file["identity_factor"] = identity_factor
+ dict_file["orthogonality_factor"] = orthogonality_factor
+ dict_file["hypothesis"] = "Orthogonality"
+ yaml.dump(dict_file, file)
+
+ backbone_checkpoint_path = f"{output_dir}/model.pth"
+ checkpoint_dir = f"{output_dir}/last.ckpt"
+
+ #####################
+ ## IDENTITY
+ num_class = len(list(train_dataloader.dataset.labels.values()))
+ identity_head = ArcFace(
+ feat_dim=backbone_model.features.num_features, num_class=num_class
+ )
+
+ ######################
+ ## DEMOGRAPHIC
+ num_class = len(list(train_dataloader.dataset.demographic_keys.values()))
+ demographic_head = DemographicRegularHead(
+ feat_dim=backbone_model.features.num_features, num_class=num_class
+ )
+
+ ################
+ ## Trainer
+ optimizer = partial(torch.optim.SGD, lr=0.001, momentum=0.9)
+
+ # demographic_epochs = 50
+ # identity_epochs = 200
+ # ortogonality_epochs = 400
+
+ # Preparing lightining model
+ model = OrthogonalityModel(
+ backbone=backbone_model,
+ identity_head=identity_head,
+ demographic_head=demographic_head,
+ loss_fn=torch.nn.CrossEntropyLoss(),
+ optimizer_fn=optimizer,
+ identity_factor=identity_factor,
+ orthogonality_factor=orthogonality_factor,
+ backbone_checkpoint_path=backbone_checkpoint_path,
+ demographic_epochs=demographic_epochs,
+ identity_epochs=identity_epochs,
+ )
+
+ """
+ from bob.bio.face.pytorch.callbacks import VanillaBiometricsCallback
+
+
+ vanilla_callback = VanillaBiometricsCallback(
+ config="/remote/idiap.svm/user.active/tpereira/gitlab/bob/bob.nightlies/src/bob.bio.demographics/bob/bio/demographics/fair_transformers/transformers.py",
+ output_path="./vanilla-callback",
+ )
+ """
+ model_checkpoint = ModelCheckpoint(
+ output_dir, every_n_train_steps=100, save_last=True
+ )
+ logger = TensorBoardLogger(os.path.join(output_dir, "tb_logs"))
+
+ # Be careful with
+ # https://github.com/PyTorchLightning/pytorch-lightning/issues/5325
+ resume_from_checkpoint = checkpoint_dir if os.path.exists(checkpoint_dir) else None
+
+ # TODO: using this code to learn too
+ # so, be nice with my comments
+ # callbacks=[model_checkpoint, vanilla_callback],
+ callbacks = [model_checkpoint]
+ trainer = pl.Trainer(
+ callbacks=callbacks,
+ logger=logger,
+ max_epochs=max_epochs,
+ gpus=-1 if torch.cuda.is_available() else None,
+ resume_from_checkpoint=resume_from_checkpoint,
+ # resume_from_checkpoint=resume_from_checkpoint, #https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#resume-from-checkpoint
+ # debug flags
+ # limit_train_batches=10, # https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#limit-train-batches
+ # limit_val_batches=1,
+ amp_level="00", # https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#amp-level
+ log_every_n_steps=5,
+ )
+
+ trainer.fit(
+ model=model, train_dataloaders=train_dataloader,
+ )
+
+
+def mine_trainer(
+ output_dir,
+ identity_factor,
+ mine_factor,
+ max_epochs,
+ demographic_epochs,
+ identity_epochs,
+ batch_size,
+ train_dataloader,
+ backbone_model,
+ transform,
+):
+
+ """
+ Trains a Pytorch CNN using the ortogonality hypothesis
+ """
+
+ # Defing the variables of the experiment, so we don't get lost
+ os.makedirs(output_dir, exist_ok=True)
+ with open(f"{output_dir}/config.yml", "w") as file:
+ dict_file = dict()
+ dict_file["demographic_epochs"] = demographic_epochs
+ dict_file["identity_epochs"] = identity_epochs
+ dict_file["max_epochs"] = max_epochs
+ dict_file["batch_size"] = batch_size
+ dict_file["identity_factor"] = identity_factor
+ dict_file["mine_factor"] = mine_factor
+ dict_file["hypothesis"] = "MINE"
+ yaml.dump(dict_file, file)
+
+ backbone_checkpoint_path = f"{output_dir}/model.pth"
+ checkpoint_dir = f"{output_dir}/last.ckpt"
+
+ #####################
+ ## IDENTITY
+ num_class = len(list(train_dataloader.dataset.labels.values()))
+ identity_head = ArcFace(
+ feat_dim=backbone_model.features.num_features, num_class=num_class
+ )
+
+ ######################
+ ## DEMOGRAPHIC
+ num_class = len(list(train_dataloader.dataset.demographic_keys.values()))
+ demographic_head = DemographicRegularHead(
+ feat_dim=backbone_model.features.num_features, num_class=num_class
+ )
+
+ ################
+ ## Trainer
+ optimizer = partial(torch.optim.SGD, lr=0.001, momentum=0.9)
+
+ # Preparing lightining model
+ model = MINEModel(
+ backbone=backbone_model,
+ identity_head=identity_head,
+ demographic_head=demographic_head,
+ loss_fn=torch.nn.CrossEntropyLoss(),
+ optimizer_fn=optimizer,
+ identity_factor=identity_factor,
+ mine_factor=mine_factor,
+ backbone_checkpoint_path=backbone_checkpoint_path,
+ demographic_epochs=demographic_epochs,
+ identity_epochs=identity_epochs,
+ )
+
+ """
+ from bob.bio.face.pytorch.callbacks import VanillaBiometricsCallback
+
+
+ vanilla_callback = VanillaBiometricsCallback(
+ config="/remote/idiap.svm/user.active/tpereira/gitlab/bob/bob.nightlies/src/bob.bio.demographics/bob/bio/demographics/fair_transformers/transformers.py",
+ output_path="./vanilla-callback",
+ )
+ """
+ model_checkpoint = ModelCheckpoint(
+ output_dir, every_n_train_steps=100, save_last=True
+ )
+ logger = TensorBoardLogger(os.path.join(output_dir, "tb_logs"))
+
+ # Be careful with
+ # https://github.com/PyTorchLightning/pytorch-lightning/issues/5325
+ resume_from_checkpoint = checkpoint_dir if os.path.exists(checkpoint_dir) else None
+
+ # TODO: using this code to learn too
+ # so, be nice with my comments
+ # callbacks=[model_checkpoint, vanilla_callback],
+ callbacks = [model_checkpoint]
+ trainer = pl.Trainer(
+ callbacks=callbacks,
+ logger=logger,
+ max_epochs=max_epochs,
+ gpus=-1 if torch.cuda.is_available() else None,
+ resume_from_checkpoint=resume_from_checkpoint,
+ # resume_from_checkpoint=resume_from_checkpoint, #https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#resume-from-checkpoint
+ # debug flags
+ # limit_train_batches=10, # https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#limit-train-batches
+ # limit_val_batches=1,
+ amp_level="00", # https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#amp-level
+ log_every_n_steps=5,
+ )
+
+ trainer.fit(
+ model=model, train_dataloaders=train_dataloader,
+ )
diff --git a/bob/bio/demographics/reports.py b/bob/bio/demographics/reports.py
index a6c34552edd6c3a40ab9cedd5b499ae8e51f14b3..2b786a1d8cfedbc914213be87fc1213beadebb29 100644
--- a/bob/bio/demographics/reports.py
+++ b/bob/bio/demographics/reports.py
@@ -72,8 +72,8 @@ def meds_report(
taus = [compute_fmr_thresholds(d, fmr_thresholds) for d in negatives_dev]
fig = plot_fdr(
- negatives_dev,
- positives_dev,
+ negatives_eval,
+ positives_eval,
titles,
variable_suffix,
fmr_thresholds,
@@ -123,7 +123,7 @@ def morph_report(
output_filename,
scores_eval=None,
fmr_thresholds=[10 ** i for i in list(range(-8, 0))],
- percentile=0.01,
+ percentile=0.05,
titles=None,
possible_races=["A", "B", "H", "W"],
genders_considered=["M"],
@@ -178,16 +178,16 @@ def morph_report(
taus = [compute_fmr_thresholds(d, fmr_thresholds) for d in negatives_dev]
- # fig = plot_fdr(
- # negatives_dev,
- # positives_dev,
- # titles,
- # variable_suffix,
- # fmr_thresholds,
- # taus=taus,
- # )
+ fig = plot_fdr(
+ negatives_eval,
+ positives_eval,
+ titles,
+ variable_suffix,
+ fmr_thresholds,
+ taus=taus,
+ )
- # pdf.savefig(fig)
+ pdf.savefig(fig)
for i, (n_dev, p_dev, n_eval, p_eval) in enumerate(
zip(negatives_dev, positives_dev, negatives_eval, positives_eval)
@@ -209,17 +209,17 @@ def morph_report(
pdf.savefig(fig)
#### PLOTTING THE FMR AND FNMR TRADE OFF
- # fig = plot_fmr_fnmr_tradeoff(
- # n_dev,
- # p_dev,
- # variable_suffix=variable_suffix,
- # fmr_thresholds=fmr_thresholds,
- # negatives_eval=n_eval,
- # positives_eval=p_eval,
- # print_fmr_fnmr=True,
- # label_lookup_table=label_lookup_table,
- # )
- # pdf.savefig(fig)
+ fig = plot_fmr_fnmr_tradeoff(
+ n_dev,
+ p_dev,
+ variable_suffix=variable_suffix,
+ fmr_thresholds=fmr_thresholds,
+ negatives_eval=n_eval,
+ positives_eval=p_eval,
+ print_fmr_fnmr=True,
+ label_lookup_table=label_lookup_table,
+ )
+ pdf.savefig(fig)
pdf.close()
diff --git a/bob/bio/demographics/script/commands.py b/bob/bio/demographics/script/commands.py
index 082603a030fe6f13e4a0c492f7c208d44b669b64..626bb6ffe95d4fea7460c2f81fc4ba55df260859 100644
--- a/bob/bio/demographics/script/commands.py
+++ b/bob/bio/demographics/script/commands.py
@@ -81,7 +81,7 @@ def morph(ctx, scores, evaluation, output, titles, percentile, dask_client, **ka
scores_eval,
percentile=percentile,
titles=titles,
- fmr_thresholds=[10 ** i for i in list(range(-6, -2))],
+ fmr_thresholds=[10 ** i for i in list(range(-8, 0))],
)
dask_client.shutdown()
diff --git a/bob/bio/demographics/test/test_mine.py b/bob/bio/demographics/test/test_mine.py
new file mode 100644
index 0000000000000000000000000000000000000000..409700480b804407791780a9221f0601492ae5cb
--- /dev/null
+++ b/bob/bio/demographics/test/test_mine.py
@@ -0,0 +1,45 @@
+import torch
+import torch.nn as nn
+
+
+from bob.bio.demographics.regularizers.independence import Mine, T
+import numpy as np
+
+
+def run_mine(is_mine_f):
+ np.random.seed(10)
+ N = 1000
+ d = 1
+ EPOCHS = 1000
+
+ # Generating
+ X = np.sign(np.random.normal(0.0, 1.0, [N, d]))
+ Z = X + np.random.normal(0.0, np.sqrt(0.2), [N, d])
+
+ from sklearn.feature_selection import mutual_info_regression
+
+ mi_numerical = mutual_info_regression(X.reshape(-1, 1), Z.ravel())[0]
+
+ x_dim = X.shape[1]
+ z_dim = Z.shape[1]
+ t = T(x_dim, z_dim)
+ model = Mine(T=t, is_mine_f=is_mine_f)
+
+ X = torch.tensor(X.astype("float32"))
+ Z = torch.tensor(Z.astype("float32"))
+
+ solver = torch.optim.Adam(model.parameters(), lr=1e-4)
+
+ for i in range(EPOCHS):
+ solver.zero_grad()
+ loss = model.forward(X, Z)
+ loss.backward()
+ solver.step()
+
+ mi = model.mi(X, Z)
+ assert np.allclose(mi, mi_numerical, atol=0.1)
+
+
+def test_mine():
+ run_mine(True)
+ run_mine(False)
diff --git a/bob/bio/demographics/test/test_regularizer.py b/bob/bio/demographics/test/test_regularizer.py
index 54181f12f0e088cd64593853930b58a73b148e69..8f6b73050396cc4464626e1542a33a616367f65c 100644
--- a/bob/bio/demographics/test/test_regularizer.py
+++ b/bob/bio/demographics/test/test_regularizer.py
@@ -10,10 +10,13 @@ from bob.learn.pytorch.head import ArcFace, Regular
from bob.bio.demographics.regularizers.independence import (
DemographicRegularHead,
OrthogonalityModel,
+ MINEModel,
)
import pytorch_lightning as pl
import torch
from functools import partial
+import torchvision.transforms as transforms
+import bob.io.image
@pytest.mark.skipif(
@@ -87,6 +90,90 @@ def test_orthogonality():
model=model, train_dataloaders=train_dataloader,
)
+
+@pytest.mark.skipif(
+ rc.get("bob.bio.demographics.directory") is None,
+ reason="Demographics features directory not available. Please do `bob config set bob.bio.demographics.directory [PATH]` to set the base features path.",
+)
+def test_mine():
+
+ database_path = os.path.join(
+ rc.get("bob.bio.demographics.directory"), "meds", "samplewrapper"
+ )
+
+ transform = transforms.Compose(
+ [
+ lambda x: bob.io.image.to_matplotlib(x.astype("float32")),
+ # transforms.ToPILImage(mode="RGB"),
+ # transforms.RandomHorizontalFlip(p=0.5),
+ # transforms.RandomRotation(degrees=(-3, 3)),
+ # transforms.RandomAutocontrast(p=0.1),
+ transforms.ToTensor(),
+ lambda x: (x - 127.5) / 128.0,
+ ]
+ )
+
+ dataset = MedsTorchDataset(
+ protocol="verification_fold1", database_path=database_path, transform=transform
+ )
+
+ # train_dataloader = DataLoader(
+ # dataset, batch_size=64, shuffle=True, pin_memory=True, num_workers=2
+ # )
+ train_dataloader = DataLoader(dataset, batch_size=64, shuffle=True)
+
+ backbone = iresnet34(
+ pretrained="/idiap/temp/tpereira/bob/data/pytorch/iresnet-91a5de61/iresnet34-5b0d0e90.pth"
+ )
+
+ # list(dataloader.dataset.labels.values())
+ #####################
+ ## IDENTITY
+ num_class = len(list(train_dataloader.dataset.labels.values()))
+ identity_head = ArcFace(
+ feat_dim=backbone.features.num_features, num_class=num_class
+ )
+
+ ######################
+ ## DEMOGRAPHIC
+ num_class = len(list(train_dataloader.dataset.demographic_keys.values()))
+ demographic_head = DemographicRegularHead(
+ feat_dim=backbone.features.num_features, num_class=num_class
+ )
+
+ ################
+ ## Trainer
+ optimizer = partial(torch.optim.SGD, lr=0.1, momentum=0.9)
+
+ # Preparing lightining model
+ model = MINEModel(
+ backbone=backbone,
+ identity_head=identity_head,
+ demographic_head=demographic_head,
+ loss_fn=torch.nn.CrossEntropyLoss(),
+ optimizer_fn=optimizer,
+ demographic_epochs=1,
+ identity_epochs=1,
+ )
+
+ # TODO: using this code to learn too
+ # so, be nice with my comments
+ trainer = pl.Trainer(
+ # callbacks=..... # https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#callbacks
+ # logger=logger,...
+ max_epochs=4,
+ gpus=-1 if torch.cuda.is_available() else None,
+ # resume_from_checkpoint=resume_from_checkpoint, #https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#resume-from-checkpoint
+ # debug flags
+ # limit_train_batches=10, # https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#limit-train-batches
+ # limit_val_batches=1,
+ amp_level="00", # https://pytorch-lightning.readthedocs.io/en/latest/common/trainer.html#amp-level
+ )
+
+ trainer.fit(
+ model=model, train_dataloaders=train_dataloader,
+ )
+
pass
# head = ArcFace(feat_dim=30, num_class=10)
# head = Regular(feat_dim=84, num_class=10)
diff --git a/experiments/regularization/mine/meds/meds_evaluate.sh b/experiments/regularization/mine/meds/meds_evaluate.sh
new file mode 100644
index 0000000000000000000000000000000000000000..042e1270ad98265e522b4e75f46bf395e190dd76
--- /dev/null
+++ b/experiments/regularization/mine/meds/meds_evaluate.sh
@@ -0,0 +1,26 @@
+BASE_PATH=/remote/idiap.svm/user.active/tpereira/gitlab/bob/bob.nightlies
+
+#$BASE_PATH/bin/bob bio pipelines vanilla-biometrics $BASE_PATH/src/bob.bio.demographics/bob/bio/demographics/experiments/evaluation/ortogonality_hypothesis/meds.py --output /idiap/temp/tpereira/2.FRDemographics/regularization/experiments/orthogonality/iresnet100_identity_factor1_orthogonality_factor1 --dask-partition-size 50 -c -g dev -g eval -l sge
+
+#/idiap/temp/tpereira/2.FRDemographics/regularization/experiments/orthogonality/meds/analysis
+
+
+OUTPUT_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/experiments/orthogonality/meds/
+MODELS_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality/meds/
+
+BACKBONE=iresnet100
+
+
+array=( "identity-1.0_orthogonality-1.0" "identity-1.0_orthogonality-5.0" "identity-1.0_orthogonality-10.0" "identity-1.0_orthogonality-100.0" "identity-5.0_orthogonality-1.0" "identity-10.0_orthogonality-1.0")
+for CNN in "${array[@]}"
+do
+ ./bin/python $BASE_PATH/src/bob.bio.demographics/bob/bio/demographics/experiments/evaluation/ortogonality/meds.py $BACKBONE $MODELS_PATH/$BACKBONE/$CNN/model.pth $OUTPUT_PATH/$BACKBONE/$CNN/
+
+done
+
+
+
+
+# -l sge
+
+
diff --git a/experiments/regularization/mine/meds/meds_train.sh b/experiments/regularization/mine/meds/meds_train.sh
new file mode 100644
index 0000000000000000000000000000000000000000..547bf69f03a579f4c1a9c508128c31d6eb7028b3
--- /dev/null
+++ b/experiments/regularization/mine/meds/meds_train.sh
@@ -0,0 +1,49 @@
+BASE_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality/meds/
+
+BACKBONE=iresnet100
+
+COMMAND=./src/bob.bio.demographics/bob/bio/demographics/experiments/train/ortogonality/meds.py
+
+./bin/jman submit --name ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-1.0 \
+ --identity-factor 1. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-5.0 \
+ --identity-factor 1. --orthogonality-factor 5. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-10.0 \
+ --identity-factor 1. --orthogonality-factor 10. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-100.0 \
+ --identity-factor 1. --orthogonality-factor 100. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-5.0_orthogonality-1.0 \
+ --identity-factor 5. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-10.0_orthogonality-1.0 \
+ --identity-factor 10. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+
+
+
+./bin/python ./src/bob.bio.demographics/bob/bio/demographics/experiments/train/mine/meds.py /idiap/temp/tpereira/2.FRDemographics/regularization/models/mine/meds/iresnet100/identity-1.0_mine-1.0 \
+ --identity-factor 1. --mine 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
diff --git a/experiments/regularization/ortogonality/meds/meds_evaluate.sh b/experiments/regularization/ortogonality/meds/meds_evaluate.sh
index 927d0a67a8083a769c83723c1dd4c9715afd1825..042e1270ad98265e522b4e75f46bf395e190dd76 100644
--- a/experiments/regularization/ortogonality/meds/meds_evaluate.sh
+++ b/experiments/regularization/ortogonality/meds/meds_evaluate.sh
@@ -1,5 +1,26 @@
BASE_PATH=/remote/idiap.svm/user.active/tpereira/gitlab/bob/bob.nightlies
-$BASE_PATH/bin/bob bio pipelines vanilla-biometrics $BASE_PATH/src/bob.bio.demographics/bob/bio/demographics/experiments/evaluation/ortogonality_hypothesis/meds.py --output /idiap/temp/tpereira/2.FRDemographics/regularization/experiments/orthogonality/iresnet100_identity_factor1_orthogonality_factor1 --dask-partition-size 50 -c -g dev -g eval -l sge
+#$BASE_PATH/bin/bob bio pipelines vanilla-biometrics $BASE_PATH/src/bob.bio.demographics/bob/bio/demographics/experiments/evaluation/ortogonality_hypothesis/meds.py --output /idiap/temp/tpereira/2.FRDemographics/regularization/experiments/orthogonality/iresnet100_identity_factor1_orthogonality_factor1 --dask-partition-size 50 -c -g dev -g eval -l sge
+
+#/idiap/temp/tpereira/2.FRDemographics/regularization/experiments/orthogonality/meds/analysis
+
+
+OUTPUT_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/experiments/orthogonality/meds/
+MODELS_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality/meds/
+
+BACKBONE=iresnet100
+
+
+array=( "identity-1.0_orthogonality-1.0" "identity-1.0_orthogonality-5.0" "identity-1.0_orthogonality-10.0" "identity-1.0_orthogonality-100.0" "identity-5.0_orthogonality-1.0" "identity-10.0_orthogonality-1.0")
+for CNN in "${array[@]}"
+do
+ ./bin/python $BASE_PATH/src/bob.bio.demographics/bob/bio/demographics/experiments/evaluation/ortogonality/meds.py $BACKBONE $MODELS_PATH/$BACKBONE/$CNN/model.pth $OUTPUT_PATH/$BACKBONE/$CNN/
+
+done
+
+
+
+
+# -l sge
diff --git a/experiments/regularization/ortogonality/meds/meds_evaluate_irenet50.sh b/experiments/regularization/ortogonality/meds/meds_evaluate_irenet50.sh
new file mode 100644
index 0000000000000000000000000000000000000000..01617021493437f307dec588688b7a5eb01f7dda
--- /dev/null
+++ b/experiments/regularization/ortogonality/meds/meds_evaluate_irenet50.sh
@@ -0,0 +1,26 @@
+BASE_PATH=/remote/idiap.svm/user.active/tpereira/gitlab/bob/bob.nightlies
+
+#$BASE_PATH/bin/bob bio pipelines vanilla-biometrics $BASE_PATH/src/bob.bio.demographics/bob/bio/demographics/experiments/evaluation/ortogonality_hypothesis/meds.py --output /idiap/temp/tpereira/2.FRDemographics/regularization/experiments/orthogonality/iresnet100_identity_factor1_orthogonality_factor1 --dask-partition-size 50 -c -g dev -g eval -l sge
+
+#/idiap/temp/tpereira/2.FRDemographics/regularization/experiments/orthogonality/meds/analysis
+
+
+OUTPUT_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/experiments/orthogonality/meds/
+MODELS_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality/meds/
+
+BACKBONE=iresnet50
+
+
+array=( "identity-1.0_orthogonality-1.0" "identity-1.0_orthogonality-5.0" "identity-1.0_orthogonality-10.0" "identity-1.0_orthogonality-100.0" "identity-5.0_orthogonality-1.0" "identity-10.0_orthogonality-1.0")
+for CNN in "${array[@]}"
+do
+ ./bin/python $BASE_PATH/src/bob.bio.demographics/bob/bio/demographics/experiments/evaluation/ortogonality/meds.py $BACKBONE $MODELS_PATH/$BACKBONE/$CNN/model.pth $OUTPUT_PATH/$BACKBONE/$CNN/
+
+done
+
+
+
+
+# -l sge
+
+
diff --git a/experiments/regularization/ortogonality/meds/meds_train.sh b/experiments/regularization/ortogonality/meds/meds_train.sh
index 589f7409679fcef8b47172ef74376761d6022a6e..f2f525dfa2248eeab154bdac27725f21eb60a214 100644
--- a/experiments/regularization/ortogonality/meds/meds_train.sh
+++ b/experiments/regularization/ortogonality/meds/meds_train.sh
@@ -1,18 +1,41 @@
-BASE_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality_hypothesis/meds/
-
-./bin/jman submit --name ORTH -q sgpu -- ./bin/python ./src/bob.bio.demographics/bob/bio/demographics/experiments/train/ortogonality_hypothesis/meds.py $BASE_PATH/identity-1.0_orthogonality-1.0 \
- --identity-factor 1. --orthogonality-factor 1. --max-epochs 1000
-
-./bin/jman submit --name ORTH -q sgpu -- ./bin/python ./src/bob.bio.demographics/bob/bio/demographics/experiments/train/ortogonality_hypothesis/meds.py $BASE_PATH/identity-5.0_orthogonality-1.0 \
- --identity-factor 5. --orthogonality-factor 1. --max-epochs 1000
-
-./bin/jman submit --name ORTH -q sgpu -- ./bin/python ./src/bob.bio.demographics/bob/bio/demographics/experiments/train/ortogonality_hypothesis/meds.py $BASE_PATH/identity-10.0_orthogonality-1.0 \
- --identity-factor 10. --orthogonality-factor 1. --max-epochs 1000
-
-
-./bin/jman submit --name ORTH -q sgpu -- ./bin/python ./src/bob.bio.demographics/bob/bio/demographics/experiments/train/ortogonality_hypothesis/meds.py $BASE_PATH/identity-1.0_orthogonality-5.0 \
- --identity-factor 1. --orthogonality-factor 5. --max-epochs 1000
-
-
-./bin/jman submit --name ORTH -q sgpu -- ./bin/python ./src/bob.bio.demographics/bob/bio/demographics/experiments/train/ortogonality_hypothesis/meds.py $BASE_PATH/identity-1.0_orthogonality-10.0 \
- --identity-factor 1. --orthogonality-factor 10. --max-epochs 1000
+BASE_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality/meds/
+
+BACKBONE=iresnet100
+
+COMMAND=./src/bob.bio.demographics/bob/bio/demographics/experiments/train/ortogonality/meds.py
+
+./bin/jman submit --name ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-1.0 \
+ --identity-factor 1. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-5.0 \
+ --identity-factor 1. --orthogonality-factor 5. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-10.0 \
+ --identity-factor 1. --orthogonality-factor 10. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-100.0 \
+ --identity-factor 1. --orthogonality-factor 100. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-5.0_orthogonality-1.0 \
+ --identity-factor 5. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-10.0_orthogonality-1.0 \
+ --identity-factor 10. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
diff --git a/experiments/regularization/ortogonality/meds/meds_train_iresnet50.sh b/experiments/regularization/ortogonality/meds/meds_train_iresnet50.sh
new file mode 100644
index 0000000000000000000000000000000000000000..bf84881f561259c894fa3b28d10511962661b199
--- /dev/null
+++ b/experiments/regularization/ortogonality/meds/meds_train_iresnet50.sh
@@ -0,0 +1,41 @@
+BASE_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality/meds/
+
+BACKBONE=iresnet50
+
+COMMAND=./src/bob.bio.demographics/bob/bio/demographics/experiments/train/ortogonality/meds.py
+
+./bin/jman submit --name MEDS-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-1.0 \
+ --identity-factor 1. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MEDS-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-5.0 \
+ --identity-factor 1. --orthogonality-factor 5. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MEDS-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-10.0 \
+ --identity-factor 1. --orthogonality-factor 10. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MEDS-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-100.0 \
+ --identity-factor 1. --orthogonality-factor 100. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MEDS-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-5.0_orthogonality-1.0 \
+ --identity-factor 5. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MEDS-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-10.0_orthogonality-1.0 \
+ --identity-factor 10. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
diff --git a/experiments/regularization/ortogonality/morph/morph_evaluate.sh b/experiments/regularization/ortogonality/morph/morph_evaluate.sh
new file mode 100644
index 0000000000000000000000000000000000000000..e0e8491ca141cf6e610189e6fb4baae89c554fb4
--- /dev/null
+++ b/experiments/regularization/ortogonality/morph/morph_evaluate.sh
@@ -0,0 +1,23 @@
+BASE_PATH=/remote/idiap.svm/user.active/tpereira/gitlab/bob/bob.nightlies
+
+
+
+OUTPUT_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/experiments/orthogonality/morph/
+MODELS_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality/morph/
+
+BACKBONE=iresnet100
+
+
+array=( "identity-1.0_orthogonality-1.0" "identity-1.0_orthogonality-5.0" "identity-1.0_orthogonality-10.0" "identity-1.0_orthogonality-100.0" "identity-5.0_orthogonality-1.0" "identity-10.0_orthogonality-1.0")
+for CNN in "${array[@]}"
+do
+ ./bin/python $BASE_PATH/src/bob.bio.demographics/bob/bio/demographics/experiments/evaluation/ortogonality/morph.py $BACKBONE $MODELS_PATH/$BACKBONE/$CNN/model.pth $OUTPUT_PATH/$BACKBONE/$CNN/ -l sge-gpu
+
+done
+
+
+
+
+# -l sge
+
+
diff --git a/experiments/regularization/ortogonality/morph/morph_evaluate_iresnet50.sh b/experiments/regularization/ortogonality/morph/morph_evaluate_iresnet50.sh
new file mode 100644
index 0000000000000000000000000000000000000000..72f0b6b5ba287c2efd5aefca5873a25dec2b0a38
--- /dev/null
+++ b/experiments/regularization/ortogonality/morph/morph_evaluate_iresnet50.sh
@@ -0,0 +1,23 @@
+BASE_PATH=/remote/idiap.svm/user.active/tpereira/gitlab/bob/bob.nightlies
+
+
+
+OUTPUT_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/experiments/orthogonality/morph/
+MODELS_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality/morph/
+
+BACKBONE=iresnet50
+
+
+array=( "identity-1.0_orthogonality-1.0" "identity-1.0_orthogonality-5.0" "identity-1.0_orthogonality-10.0" "identity-1.0_orthogonality-100.0" "identity-5.0_orthogonality-1.0" "identity-10.0_orthogonality-1.0")
+for CNN in "${array[@]}"
+do
+ ./bin/python $BASE_PATH/src/bob.bio.demographics/bob/bio/demographics/experiments/evaluation/ortogonality/morph.py $BACKBONE $MODELS_PATH/$BACKBONE/$CNN/model.pth $OUTPUT_PATH/$BACKBONE/$CNN/ -l sge
+
+done
+
+
+
+
+# -l sge
+
+
diff --git a/experiments/regularization/ortogonality/morph/morph_iresnet50_train.sh b/experiments/regularization/ortogonality/morph/morph_iresnet50_train.sh
new file mode 100644
index 0000000000000000000000000000000000000000..9cd714fcc69c98847ad5ea5e9f707b5ce0b6e05c
--- /dev/null
+++ b/experiments/regularization/ortogonality/morph/morph_iresnet50_train.sh
@@ -0,0 +1,50 @@
+BASE_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality/morph/
+
+BACKBONE=iresnet50
+
+COMMAND=./src/bob.bio.demographics/bob/bio/demographics/experiments/train/ortogonality/morph.py
+
+./bin/jman submit --name MORPH-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-1.0 \
+ --identity-factor 1. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MORPH-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-5.0 \
+ --identity-factor 1. --orthogonality-factor 5. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MORPH-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-10.0 \
+ --identity-factor 1. --orthogonality-factor 10. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MORPH-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-100.0 \
+ --identity-factor 1. --orthogonality-factor 100. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MORPH-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-5.0_orthogonality-1.0 \
+ --identity-factor 5. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MORPH-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-10.0_orthogonality-1.0 \
+ --identity-factor 10. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+
+
+
+#./bin/python ./src/bob.bio.demographics/bob/bio/demographics/experiments/train/ortogonality/morph.py /idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality/morph/iresnet100/identity-1.0_orthogonality-1.0 \
+# --identity-factor 1. --orthogonality-factor 1. --identity-epochs 1 --demographic-epochs 1 \
+# --backbone iresnet100 \
+# --max-epochs 1000
+
diff --git a/experiments/regularization/ortogonality/morph/morph_train.sh b/experiments/regularization/ortogonality/morph/morph_train.sh
new file mode 100644
index 0000000000000000000000000000000000000000..c1a874d22592317e899a2dfd8d089a757fc2a075
--- /dev/null
+++ b/experiments/regularization/ortogonality/morph/morph_train.sh
@@ -0,0 +1,50 @@
+BASE_PATH=/idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality/morph/
+
+BACKBONE=iresnet100
+
+COMMAND=./src/bob.bio.demographics/bob/bio/demographics/experiments/train/ortogonality/morph.py
+
+./bin/jman submit --name MORPH-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-1.0 \
+ --identity-factor 1. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MORPH-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-5.0 \
+ --identity-factor 1. --orthogonality-factor 5. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MORPH-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-10.0 \
+ --identity-factor 1. --orthogonality-factor 10. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MORPH-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-1.0_orthogonality-100.0 \
+ --identity-factor 1. --orthogonality-factor 100. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MORPH-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-5.0_orthogonality-1.0 \
+ --identity-factor 5. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+./bin/jman submit --name MORPH-ORTH -q sgpu -- ./bin/python $COMMAND $BASE_PATH/$BACKBONE/identity-10.0_orthogonality-1.0 \
+ --identity-factor 10. --orthogonality-factor 1. \
+ --backbone $BACKBONE \
+ --max-epochs 1000
+
+
+
+
+
+#./bin/python ./src/bob.bio.demographics/bob/bio/demographics/experiments/train/ortogonality/morph.py /idiap/temp/tpereira/2.FRDemographics/regularization/models/orthogonality/morph/iresnet100/identity-1.0_orthogonality-1.0 \
+# --identity-factor 1. --orthogonality-factor 1. --identity-epochs 1 --demographic-epochs 1 \
+# --backbone iresnet100 \
+# --max-epochs 1000
+