deepboof.forward

Interface SpatialBatchNorm<T extends Tensor<T>>

All Superinterfaces:

BatchNorm, Function<T>

All Known Implementing Classes:

SpatialBatchNorm_F32, SpatialBatchNorm_F64

public interface SpatialBatchNorm<T extends Tensor<T>>
extends Function<T>, BatchNorm

Spatial Batch Normalization seeks to maintain the convolutional property, "that different elements of the same feature map, at different locations, are normalized in the same way." [1] Thus the input tensor (N,C,H,W) is "reshaped" such that it is (N*H*W,C) and it's treated like a mini-batch with N*H*W elements.

See BatchNorm for a general discussion of Batch Normalization

Method Summary

Modifier and Type	Method and Description
void	forward(T input, T output) Performs batch norm on spatial data.
void	setParameters(java.util.List<T> parameters) See forward(T, T) for a description of parameters.

Methods inherited from interface deepboof.Function

getOutputShape, getParameters, getParameterShapes, getTensorType, initialize

Methods inherited from interface deepboof.forward.BatchNorm

getEPS, hasGammaBeta, setEPS

Method Detail

forward

void forward(T input,
             T output)

Performs batch norm on spatial data.

 Summary Table
 -------------------------------------------------
 Input   shape = (N, C, H,  W)
 Output  shape = (N, C, H,  W)
 Params  shape = (C, M)
 -------------------------------------------------
 N   = Size of mini-batch
 C   = Number of channels in input image
 H   = Height of input image
 W   = With of input image
 M   = Number of parameters.  2 or 4 if gamma-beta is being used.
       in order of: mean, stdev  OR mean, stdev, gamma, beta
 

Specified by:

forward in interface Function<T extends Tensor<T>>

Parameters:

input - Input tensor = (N,C,H,W)

output - Output tensor = (N,C,H,W). Modified.

setParameters

void setParameters(java.util.List<T> parameters)

See forward(T, T) for a description of parameters.

Specified by:

setParameters in interface Function<T extends Tensor<T>>

Parameters:

parameters - Variable tensor. (C, M), where M is 2 or 4. Not modified.