gsnn.models.GroupLayerNorm

class gsnn.models.GroupLayerNorm(*args: Any, **kwargs: Any)

Bases: Module

Layer normalization computed separately within each channel group.

__init__(channel_groups, eps=0.1, affine=True)

Parameters:

to; (channel_groups tensor specifies which group a channel belongs) – for instance given: [0,0, 1,1, 2,2] specifies 3 groups with 2 channels in each; the first two channels are assigned to group 0.

Methods

__init__(channel_groups[, eps, affine])	param channel_groups tensor specifies which group a channel belongs to;: for instance given: [0,0, 1,1, 2,2] specifies 3 groups with 2 channels in each;
forward(x)	Normalize x with shape (B, C, 1) or squeezed (B, C); returns (B, C, 1).

forward(x)

Normalize x with shape (B, C, 1) or squeezed (B, C); returns (B, C, 1).