InstanceNorm

Applies Instance Normalization over a 4D input (a mini-batch of 2D inputs with additional channel dimension) as described in the paper Instance Normalization: The Missing Ingredient for Fast Stylization (opens in a new tab).

Learn more about InstanceNorm (opens in a new tab)

Usage

from tinygrad.nn import InstanceNorm 
 
N, C, H, W = 20, 5, 10, 10
 
layer = InstanceNorm(C)

Arguments

num_features

The number of channels or the number of features of the input tensor.

eps (default: 1e-5)

The epsilon value to avoid division by zero.

affine (default: True)

The flag indicating whether to apply the affine transformation or not.