InstanceNormalization#

InstanceNormalization - 6
InstanceNormalization - 1

InstanceNormalization - 6 #

Version

name: InstanceNormalization (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True

This version of the operator has been available since version 6.

Summary

Carries out instance normalization as described in the paper https://arxiv.org/abs/1607.08022.

y = scale * (x - mean) / sqrt(variance + epsilon) + B, where mean and variance are computed per instance per channel.

Attributes

epsilon: The epsilon value to use to avoid division by zero. Default value is 9.999999747378752e-06.

Inputs

input (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
scale (heterogeneous) - T: The input 1-dimensional scale tensor of size C.
B (heterogeneous) - T: The input 1-dimensional bias tensor of size C.

Outputs

output (heterogeneous) - T: The output tensor of the same shape as input.

Type Constraints

T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.

Examples

Differences

`0`	`0`	`Carries out instance normalization as described in the paper`	`Carries out instance normalization as described in the paper`
`1`	`1`	`https://arxiv.org/abs/1607.08022.`	`https://arxiv.org/abs/1607.08022.`
`2`	`2`
`3`	`3`	`y = scale * (x - mean) / sqrt(variance + epsilon) + B,`	`y = scale * (x - mean) / sqrt(variance + epsilon) + B,`
`4`	`4`	`where mean and variance are computed per instance per channel.`	`where mean and variance are computed per instance per channel.`
`5`	`5`
`6`	`6`	`Attributes`	`Attributes`
`7`	`7`
`8`		`* consumed_inputs:`
`9`		`legacy optimization attribute.`
`10`	`8`	`* epsilon:`	`* epsilon:`
`11`	`9`	`The epsilon value to use to avoid division by zero, default is`	`The epsilon value to use to avoid division by zero. Default value is 9.999999747378752e-06.`
`12`		`1e-5f. Default value is 9.999999747378752e-06.`
`13`	`10`
`14`	`11`	`Inputs`	`Inputs`
`15`	`12`
`16`	`13`	`* input (heterogeneous) - T:`	`* input (heterogeneous) - T:`
`17`		`The input 4-dimensional tensor of shape NCHW.`
	`14`		`Input data tensor from the previous operator; dimensions for image`
	`15`		`case are (N x C x H x W), where N is the batch size, C is the number`
	`16`		`of channels, and H and W are the height and the width of the data.`
	`17`		`For non image case, the dimensions are in the form of (N x C x D1 x`
	`18`		`D2 ... Dn), where N is the batch size.`
`18`	`19`	`* scale (heterogeneous) - T:`	`* scale (heterogeneous) - T:`
`19`	`20`	`The input 1-dimensional scale tensor of size C.`	`The input 1-dimensional scale tensor of size C.`
`20`	`21`	`* B (heterogeneous) - T:`	`* B (heterogeneous) - T:`
`21`	`22`	`The input 1-dimensional bias tensor of size C.`	`The input 1-dimensional bias tensor of size C.`
`22`	`23`
`23`	`24`	`Outputs`	`Outputs`
`24`	`25`
`25`	`26`	`* output (heterogeneous) - T:`	`* output (heterogeneous) - T:`
`26`	`27`	`The output 4-dimensional tensor of the same shape as input.`	`The output tensor of the same shape as input.`
`27`	`28`
`28`	`29`	`Type Constraints`	`Type Constraints`
`29`	`30`
`30`	`31`	`* T in (`	`* T in (`
`31`	`32`	`tensor(double),`	`tensor(double),`
`32`	`33`	`tensor(float),`	`tensor(float),`
`33`	`34`	`tensor(float16)`	`tensor(float16)`
`34`	`35`	`):`	`):`
`35`	`36`	`Constrain input and output types to float tensors.`	`Constrain input and output types to float tensors.`

InstanceNormalization - 1 #