nn.functional.instance_norm¶

lucid.nn.functional.instance_norm(input_: Tensor, running_mean: Tensor | None = None, running_var: Tensor | None = None, weight: Tensor | None = None, bias: Tensor | None = None, training: bool = True, momentum: float = 0.1, eps: float = 1e-05) → Tensor¶

The instance_norm function performs instance normalization on the input tensor, commonly used in style transfer and image processing applications.

Function Signature¶

def instance_norm(
    input_: Tensor,
    running_mean: Tensor,
    running_var: Tensor,
    weight: Tensor | None = None,
    bias: Tensor | None = None,
    training: bool = True,
    momentum: float = 0.1,
    eps: float = 1e-5,
) -> Tensor

Parameters¶

input_ (Tensor):
The input tensor of shape (N, C, …) where N is the batch size, C is the number of channels, and … represents any number of additional dimensions.
running_mean (Tensor):
The running mean tensor of shape (C,). Used for normalization in evaluation mode.
running_var (Tensor):
The running variance tensor of shape (C,). Used for normalization in evaluation mode.
weight (Tensor | None, optional):
The learnable scale parameter of shape (C,). Default: None.
bias (Tensor | None, optional):
The learnable shift parameter of shape (C,). Default: None.
training (bool, optional):
If True, the layer uses batch statistics computed from the input. If False, it uses running statistics. Default: True.
momentum (float, optional):
The value used for the running mean and variance computation. Default: 0.1.
eps (float, optional):
A small constant added to the denominator for numerical stability. Default: 1e-5.

Returns¶

Tensor:
The normalized tensor of the same shape as input_, with normalized values for each instance.

Examples¶

Performing instance normalization during training:

>>> import lucid.nn.functional as F
>>> input_ = Tensor([[[1.0, 2.0], [3.0, 4.0]]])  # Shape: (1, 2, 2)
>>> running_mean = Tensor([0.0, 0.0])  # Shape: (2,)
>>> running_var = Tensor([1.0, 1.0])  # Shape: (2,)
>>> weight = Tensor([1.0, 1.0])  # Shape: (2,)
>>> bias = Tensor([0.0, 0.0])  # Shape: (2,)
>>> out = F.instance_norm(input_, running_mean, running_var, weight, bias, training=True)
>>> print(out)
Tensor(...)  # Normalized values