classification

BaseNetworkBuilder(features_extractor, pre_classifier=None, classifier=None, freeze=True, hyperspherical=False, flatten_features=True)

Bases: Module

Baseline Feature Extractor, with the possibility to map features to an hypersphere. If hypershperical is True the classifier is ignored.

Parameters:

• features_extractor (Module) –

  Feature extractor as a toch.nn.Module.

• pre_classifier (Module | None, default: None ) –

  Pre classifier as a torch.nn.Module. Defaults to nn.Identity() if None.

• classifier (Module | None, default: None ) –

  Classifier as a torch.nn.Module. Defaults to nn.Identity() if None.

• freeze (bool, default: True ) –

  Whether to freeze the feature extractor. Defaults to True.

• hyperspherical (bool, default: False ) –

  Whether to map features to an hypersphere. Defaults to False.

• flatten_features (bool, default: True ) –

  Whether to flatten the features before the pre_classifier. May be required if your model is outputting a feature map rather than a vector. Defaults to True.

Source code in quadra/models/classification/base.py

def __init__(
    self,
    features_extractor: nn.Module,
    pre_classifier: nn.Module | None = None,
    classifier: nn.Module | None = None,
    freeze: bool = True,
    hyperspherical: bool = False,
    flatten_features: bool = True,
):
    super().__init__()
    if pre_classifier is None:
        pre_classifier = nn.Identity()

    if classifier is None:
        classifier = nn.Identity()

    self.features_extractor = features_extractor
    self.freeze = freeze
    self.hyperspherical = hyperspherical
    self.pre_classifier = pre_classifier
    self.classifier = classifier
    self.flatten: bool = False
    self._hyperspherical: bool = False
    self.l2: L2Norm | None = None
    self.flatten_features = flatten_features

    self.freeze = freeze
    self.hyperspherical = hyperspherical

    if self.freeze:
        for p in self.features_extractor.parameters():
            p.requires_grad = False

freeze: bool property writable

Whether to freeze the feature extractor.

hyperspherical: bool property writable

Whether to map the extracted features into an hypersphere.

TimmNetworkBuilder(model_name, pretrained=True, pre_classifier=None, classifier=None, freeze=True, hyperspherical=False, flatten_features=True, **timm_kwargs)

Bases: BaseNetworkBuilder

Torchvision feature extractor, with the possibility to map features to an hypersphere.

Parameters:

• model_name (str) –

  Timm model name

• pretrained (bool, default: True ) –

  Whether to load the pretrained weights for the model.

• pre_classifier (Module | None, default: None ) –

  Pre classifier as a torch.nn.Module. Defaults to nn.Identity() if None.

• classifier (Module | None, default: None ) –

  Classifier as a torch.nn.Module. Defaults to nn.Identity() if None.

• freeze (bool, default: True ) –

  Whether to freeze the feature extractor. Defaults to True.

• hyperspherical (bool, default: False ) –

  Whether to map features to an hypersphere. Defaults to False.

• flatten_features (bool, default: True ) –

  Whether to flatten the features before the pre_classifier. Defaults to True.

• **timm_kwargs (Any, default: {} ) –

  Additional arguments to pass to timm.create_model

Source code in quadra/models/classification/backbones.py

def __init__(
    self,
    model_name: str,
    pretrained: bool = True,
    pre_classifier: nn.Module | None = None,
    classifier: nn.Module | None = None,
    freeze: bool = True,
    hyperspherical: bool = False,
    flatten_features: bool = True,
    **timm_kwargs: Any,
):
    self.pretrained = pretrained
    features_extractor = timm.create_model(model_name, pretrained=self.pretrained, num_classes=0, **timm_kwargs)

    super().__init__(
        features_extractor=features_extractor,
        pre_classifier=pre_classifier,
        classifier=classifier,
        freeze=freeze,
        hyperspherical=hyperspherical,
        flatten_features=flatten_features,
    )

TorchHubNetworkBuilder(repo_or_dir, model_name, pretrained=True, pre_classifier=None, classifier=None, freeze=True, hyperspherical=False, flatten_features=True, **torch_hub_kwargs)

Bases: BaseNetworkBuilder

TorchHub feature extractor, with the possibility to map features to an hypersphere.

Parameters:

• repo_or_dir (str) –

  The name of the repository or the path to the directory containing the model.

• model_name (str) –

  The name of the model within the repository.

• pretrained (bool, default: True ) –

  Whether to load the pretrained weights for the model.

• pre_classifier (Module | None, default: None ) –

  Pre classifier as a torch.nn.Module. Defaults to nn.Identity() if None.

• classifier (Module | None, default: None ) –

  Classifier as a torch.nn.Module. Defaults to nn.Identity() if None.

• freeze (bool, default: True ) –

  Whether to freeze the feature extractor. Defaults to True.

• hyperspherical (bool, default: False ) –

  Whether to map features to an hypersphere. Defaults to False.

• flatten_features (bool, default: True ) –

  Whether to flatten the features before the pre_classifier. Defaults to True.

• **torch_hub_kwargs (Any, default: {} ) –

  Additional arguments to pass to torch.hub.load

Source code in quadra/models/classification/backbones.py

def __init__(
    self,
    repo_or_dir: str,
    model_name: str,
    pretrained: bool = True,
    pre_classifier: nn.Module | None = None,
    classifier: nn.Module | None = None,
    freeze: bool = True,
    hyperspherical: bool = False,
    flatten_features: bool = True,
    **torch_hub_kwargs: Any,
):
    self.pretrained = pretrained
    features_extractor = torch.hub.load(
        repo_or_dir=repo_or_dir, model=model_name, pretrained=self.pretrained, **torch_hub_kwargs
    )
    super().__init__(
        features_extractor=features_extractor,
        pre_classifier=pre_classifier,
        classifier=classifier,
        freeze=freeze,
        hyperspherical=hyperspherical,
        flatten_features=flatten_features,
    )

TorchVisionNetworkBuilder(model_name, pretrained=True, pre_classifier=None, classifier=None, freeze=True, hyperspherical=False, flatten_features=True, **torchvision_kwargs)

Bases: BaseNetworkBuilder

Torchvision feature extractor, with the possibility to map features to an hypersphere.

Parameters:

• model_name (str) –

  Torchvision model function that will be evaluated, for example: torchvision.models.resnet18.

• pretrained (bool, default: True ) –

  Whether to load the pretrained weights for the model.

• pre_classifier (Module | None, default: None ) –

  Pre classifier as a torch.nn.Module. Defaults to nn.Identity() if None.

• classifier (Module | None, default: None ) –

  Classifier as a torch.nn.Module. Defaults to nn.Identity() if None.

• freeze (bool, default: True ) –

  Whether to freeze the feature extractor. Defaults to True.

• hyperspherical (bool, default: False ) –

  Whether to map features to an hypersphere. Defaults to False.

• flatten_features (bool, default: True ) –

  Whether to flatten the features before the pre_classifier. Defaults to True.

• **torchvision_kwargs (Any, default: {} ) –

  Additional arguments to pass to the model function.

Source code in quadra/models/classification/backbones.py

def __init__(
    self,
    model_name: str,
    pretrained: bool = True,
    pre_classifier: nn.Module | None = None,
    classifier: nn.Module | None = None,
    freeze: bool = True,
    hyperspherical: bool = False,
    flatten_features: bool = True,
    **torchvision_kwargs: Any,
):
    self.pretrained = pretrained
    model_function = models.__dict__[model_name]
    features_extractor = model_function(pretrained=self.pretrained, progress=True, **torchvision_kwargs)
    # Remove classifier
    features_extractor.classifier = nn.Identity()
    super().__init__(
        features_extractor=features_extractor,
        pre_classifier=pre_classifier,
        classifier=classifier,
        freeze=freeze,
        hyperspherical=hyperspherical,
        flatten_features=flatten_features,
    )