glasses.models.classification.densenet package

Module contents

class glasses.models.classification.densenet.DenseBottleNeckBlock(in_features: int, out_features: int, activation: torch.nn.modules.module.Module = functools.partial(<class 'torch.nn.modules.activation.ReLU'>, inplace=True), expansion: int = 4, **kwargs)[source]

Bases: glasses.models.classification.densenet.DenseNetBasicBlock

Bottleneck block composed by two preactivated layer of convolution. The expensive 3x3 conv is computed after a cheap 1x1 conv donwsample the input resulting in less parameters.

https://github.com/FrancescoSaverioZuppichini/glasses/blob/develop/docs/_static/images/DenseNetBottleneckBlock.png?raw=true
Parameters

  • out_features (int) – Number of input features

  • out_features – Number of output features

  • conv (nn.Module, optional) – [description]. Defaults to nn.Conv2d.

  • activation (nn.Module, optional) – [description]. Defaults to ReLUInPlace.

  • expansion (int, optional) – [description]. Defaults to 4.

Initializes internal Module state, shared by both nn.Module and ScriptModule.

training: bool
class glasses.models.classification.densenet.DenseNet(encoder: torch.nn.modules.module.Module = <class 'glasses.models.classification.densenet.DenseNetEncoder'>, head: torch.nn.modules.module.Module = <class 'glasses.models.classification.resnet.ResNetHead'>, *args, **kwargs)[source]

Bases: glasses.models.classification.base.ClassificationModule

Implementation of DenseNet proposed in Densely Connected Convolutional Networks

Create a default models

DenseNet.densenet121()
DenseNet.densenet161()
DenseNet.densenet169()
DenseNet.densenet201()

Examples

# change activation
DenseNet.densenet121(activation = nn.SELU)
# change number of classes (default is 1000 )
DenseNet.densenet121(n_classes=100)
# pass a different block
DenseNet.densenet121(block=...)
# change the initial convolution
model = DenseNet.densenet121()
model.encoder.gate.conv1 = nn.Conv2d(3, 64, kernel_size=3)
# store each feature
x = torch.rand((1, 3, 224, 224))
model = DenseNet.densenet121()
# first call .features, this will activate the forward hooks and tells the model you'll like to get the features
model.encoder.features
model(torch.randn((1,3,224,224)))
# get the features from the encoder
features = model.encoder.features
print([x.shape for x in features])
# [torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14]), torch.Size([1, 512, 7, 7]), torch.Size([1, 1024, 7, 7])]
Parameters

  • in_channels (int, optional) – Number of channels in the input Image (3 for RGB and 1 for Gray). Defaults to 3.

  • n_classes (int, optional) – Number of classes. Defaults to 1000.

Initializes internal Module state, shared by both nn.Module and ScriptModule.

classmethod densenet121(*args, **kwargs) → glasses.models.classification.densenet.DenseNet[source]

Creates a densenet121 model. Grow rate is set to 32

https://github.com/FrancescoSaverioZuppichini/glasses/blob/develop/docs/_static/images/DenseNet121.png?raw=true
Returns

A densenet121 model

Return type

DenseNet

classmethod densenet161(*args, **kwargs) → glasses.models.classification.densenet.DenseNet[source]

Creates a densenet161 model. Grow rate is set to 48

https://github.com/FrancescoSaverioZuppichini/glasses/blob/develop/docs/_static/images/DenseNet161.png?raw=true
Returns

A densenet161 model

Return type

DenseNet

classmethod densenet169(*args, **kwargs) → glasses.models.classification.densenet.DenseNet[source]

Creates a densenet169 model. Grow rate is set to 32

https://github.com/FrancescoSaverioZuppichini/glasses/blob/develop/docs/_static/images/DenseNet169.png?raw=true
Returns

A densenet169 model

Return type

DenseNet

classmethod densenet201(*args, **kwargs) → glasses.models.classification.densenet.DenseNet[source]

Creates a densenet201 model. Grow rate is set to 32

https://github.com/FrancescoSaverioZuppichini/glasses/blob/develop/docs/_static/images/DenseNet201.png?raw=true
Returns

A densenet201 model

Return type

DenseNet

forward(x: torch.Tensor) → torch.Tensor[source]

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class glasses.models.classification.densenet.DenseNetBasicBlock(in_features: int, out_features: int, activation: torch.nn.modules.module.Module = functools.partial(<class 'torch.nn.modules.activation.ReLU'>, inplace=True), *args, **kwargs)[source]

Bases: torch.nn.modules.module.Module

Basic DenseNet block composed by one 3x3 convs with residual connection. The residual connection is perfomed by concatenate the input and the output.

https://github.com/FrancescoSaverioZuppichini/glasses/blob/develop/docs/_static/images/DenseNetBasicBlock.png?raw=true
Parameters

  • out_features (int) – Number of input features

  • out_features – Number of output features

  • conv (nn.Module, optional) – [description]. Defaults to nn.Conv2d.

  • activation (nn.Module, optional) – [description]. Defaults to ReLUInPlace.

Initializes internal Module state, shared by both nn.Module and ScriptModule.

forward(x: torch.Tensor) → torch.Tensor[source]

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class glasses.models.classification.densenet.DenseNetEncoder(in_channels: int = 3, start_features: int = 64, grow_rate: int = 32, depths: List[int] = [4, 4, 4, 4], activation: torch.nn.modules.module.Module = functools.partial(<class 'torch.nn.modules.activation.ReLU'>, inplace=True), block: torch.nn.modules.module.Module = <class 'glasses.models.classification.densenet.DenseBottleNeckBlock'>, *args, **kwargs)[source]

Bases: glasses.models.classification.resnet.ResNetEncoder

DenseNet encoder composed by multiple DeseNetLayer with increasing features size. The .stem is the same used in ResNet

Parameters

  • in_channels (int, optional) – [description]. Defaults to 3.

  • start_features (int, optional) – [description]. Defaults to 64.

  • grow_rate (int, optional) – [description]. Defaults to 32.

  • depths (List[int], optional) – [description]. Defaults to [4, 4, 4, 4].

  • activation (nn.Module, optional) – [description]. Defaults to ReLUInPlace.

  • block (nn.Module, optional) – [description]. Defaults to DenseNetBasicBlock.

Initializes internal Module state, shared by both nn.Module and ScriptModule.

training: bool
class glasses.models.classification.densenet.DenseNetLayer(in_features: int, grow_rate: int = 32, n: int = 4, block: torch.nn.modules.module.Module = <class 'glasses.models.classification.densenet.DenseBottleNeckBlock'>, transition_block: torch.nn.modules.module.Module = <class 'glasses.models.classification.densenet.TransitionBlock'>, *args, **kwargs)[source]

Bases: torch.nn.modules.container.Sequential

A DenseNet layer is composed by n blocks stacked together followed by a transition to downsample the output features.

https://github.com/FrancescoSaverioZuppichini/glasses/blob/develop/docs/_static/images/DenseNetLayer.png?raw=true
Parameters

  • out_features (int) – Number of input features

  • grow_rate (int, optional) – [description]. Defaults to 32.

  • n (int, optional) – [description]. Defaults to 4.

  • block (nn.Module, optional) – [description]. Defaults to DenseNetBasicBlock.

  • transition_block (nn.Module, optional) – A module applied after the block(s). Defaults to TransitionBlock.

Initializes internal Module state, shared by both nn.Module and ScriptModule.

class glasses.models.classification.densenet.TransitionBlock(in_features: int, factor: int = 2, activation: torch.nn.modules.module.Module = functools.partial(<class 'torch.nn.modules.activation.ReLU'>, inplace=True))[source]

Bases: torch.nn.modules.container.Sequential

A transition block is used to downsample the output using 1x1 conv followed by 2x2 average pooling.

https://github.com/FrancescoSaverioZuppichini/glasses/blob/develop/docs/_static/images/DenseNetTransitionBlock.png?raw=true
Parameters

  • out_features (int) – Number of input features

  • factor (int, optional) – Reduction factor applied on the in_features. Defaults to 2

  • conv (nn.Module, optional) – [description]. Defaults to nn.Conv2d.

  • activation (nn.Module, optional) – [description]. Defaults to ReLUInPlace.

Initializes internal Module state, shared by both nn.Module and ScriptModule.