Struct GroupNorm 

pub struct GroupNorm<D: Dtype, IT, OT, const RANK: usize> {
    pub num_groups: usize,
    pub num_channels: usize,
    pub eps: f64,
    pub affine: bool,
    /* private fields */
}

Group normalisation over num_groups groups of channels.

Input shape: [N, C, *] — normalises over (C/num_groups, *) per sample. Output shape: same as input.

Parameters:

• num_groups — number of groups to divide C into (must divide C evenly)
• num_channels — number of channels C
• eps — numerical stability constant (default 1e-5)
• affine — if true, learns per-channel γ and β (default true)

Fields

num_groups: usize

Number of groups to divide the channels into.

num_channels: usize

Number of channels.

eps: f64

Numerical stability constant added to the variance.

affine: bool

Whether to learn per-channel scale (gamma) and shift (beta) parameters.

Implementations

impl<D: Dtype, IT, OT, const RANK: usize> GroupNorm<D, IT, OT, RANK>

pub fn new(num_groups: usize, num_channels: usize) -> Self

Creates a new layer with default eps, affine on.

pub fn with_eps(self, eps: f64) -> Self

Sets the numerical stability constant.

pub fn with_affine(self, affine: bool) -> Self

Sets whether to learn per-channel scale/shift parameters.

Trait Implementations

impl<D: Dtype, IT: Tensor<D, RANK> + EagerTensor, OT: Tensor<D, RANK>, const RANK: usize> Layer<IT> for GroupNorm<D, IT, OT, RANK>

type Output = OT

This layer’s output type.

fn call(&self, _input: IT) -> Self::Output

Applies this layer to input.

impl<D: Dtype, const RANK: usize> Layer<SymTensor> for GroupNorm<D, SymTensor, SymTensor, RANK>

type Output = SymTensor

This layer’s output type.

fn call(&self, input: SymTensor) -> SymTensor

Applies this layer to input.