deepboof.backward

Interface DSpatialPadding2D<T extends deepboof.Tensor<T>>

All Superinterfaces:

deepboof.ITensor, deepboof.forward.SpatialPadding2D<T>, deepboof.VTensor

All Known Subinterfaces:

DSpatialPadding2D_F64

All Known Implementing Classes:

DClippedPadding2D_F64, DConstantPadding2D_F64

public interface DSpatialPadding2D<T extends deepboof.Tensor<T>>
extends deepboof.forward.SpatialPadding2D<T>

Interface for computing the gradient of a padded spatial tensor. Spatial tensors have the shape of (N, C, H, W) where N is the number of mini-batches, C channels, H height and W width.

Method Summary

Modifier and Type	Method and Description
void	backwardsChannel(T gradientPadded, int batch, int channel, T gradientInput) Compute the gradient of the input image from the gradient of the padded image for a specific mini-batch and channel of the input tensor.
void	backwardsImage(T gradientPadded, int batch, T gradientInput) Compute the gradient of the input image from the gradient of the padded image for a specific mini-batch and channel of the input tensor.

Methods inherited from interface deepboof.forward.SpatialPadding2D

getClippingOffsetCol, getClippingOffsetRow, getPaddingCol0, getPaddingCol1, getPaddingRow0, getPaddingRow1, getTensorType, isClipped, setInput, shapeGivenInput

Methods inherited from interface deepboof.ITensor

createCoor, getDataType, getDimension, getShape, isShape, length

Method Detail

backwardsChannel

void backwardsChannel(T gradientPadded,
                      int batch,
                      int channel,
                      T gradientInput)

Compute the gradient of the input image from the gradient of the padded image for a specific mini-batch and channel of the input tensor.

Parameters:

gradientPadded - (Input) Gradient of padded image at specific mini-batch and channel. (H, W)

batch - (Input) mini-batch.

channel - (Input) Channel.

gradientInput - (Output) Output 4D spatial tensor. Only elements used to compute the channel are modified. (N, C, H, W)

backwardsImage

void backwardsImage(T gradientPadded,
                    int batch,
                    T gradientInput)

Compute the gradient of the input image from the gradient of the padded image for a specific mini-batch and channel of the input tensor.

Parameters:

gradientPadded - (Input) Gradient of padded image at specific mini-batch. (C, H, W)

batch - (Input) mini-batch.

gradientInput - (Output) Output 4D spatial tensor. Only elements used to compute the image are modified. (N, C, H, W)