torchlayers.normalization module¶
-
class
torchlayers.normalization.
BatchNorm
(num_features: int, eps: float = 1e-05, momentum: float = 0.1, affine: bool = True, track_running_stats: bool = True)[source]¶ Apply Batch Normalization over inferred dimension (2D up to 5D).
Based on input shape it either creates
1D
,2D
or3D
batch normalization for inputs of shape2D/3D
,4D
,5D
respectively (including batch as first dimension).Otherwise works like standard PyTorch’s BatchNorm.
- Parameters
num_features (int) – (number of channels in input) from an expected input. Can be number of outputs of previous linear layer as well
eps (float, optional) – Value added to the denominator for numerical stability. Default:
1e-5
momentum (float, optional) – Value used for the
running_mean
andrunning_var
computation. Can be set toNone
for cumulative moving average (i.e. simple average). Default:0.1
affine (bool, optional) – If
True
, this module has learnable affine parameters. Default:True
track_running_stats (bool, optional) – If
True
, this module tracks the running mean and variance, and when set toFalse
, this module does not track such statistics and always uses batch statistics in both training and eval modes. Default:True
-
class
torchlayers.normalization.
GroupNorm
(num_channels: int, num_groups: int, eps: float = 1e-05, affine: bool = True)[source]¶ Apply Group Normalization over a mini-batch of inputs.
Works exactly like PyTorch’s counterpart, but
num_channels
is used as first argument so it can be inferred during first forward pass.- Parameters
num_channels (int) – Number of channels expected in input
num_groups (int) – Number of groups to separate the channels into
eps (float, optional) – Value added to the denominator for numerical stability. Default:
1e-5
affine (bool, optional) – If
True
, this module has learnable affine parameters. Default:True
-
class
torchlayers.normalization.
InstanceNorm
(num_features: int, eps: float = 1e-05, momentum: float = 0.1, affine: bool = False, track_running_stats: bool = False)[source]¶ Apply Instance Normalization over inferred dimension (3D up to 5D).
Based on input shape it either creates 1D, 2D or 3D instance normalization for inputs of shape 3D, 4D, 5D respectively (including batch as first dimension).
Otherwise works like standard PyTorch’s InstanceNorm
- Parameters
num_features (int) – (number of channels in input) from an expected input. Can be number of outputs of previous linear layer as well
eps (float, optional) – Value added to the denominator for numerical stability. Default:
1e-5
momentum (float, optional) – Value used for the
running_mean
andrunning_var
computation. Default:0.1
affine (bool, optional) – If
True
, this module has learnable affine parameters, initialized just like in batch normalization. Default:False
track_running_stats (bool, optional) – If
True
, this module tracks the running mean and variance, and when set toFalse
, this module does not track such statistics and always uses batch statistics in both training and eval modes. Default:False