Correlation Cost Layer.

Source: R/layers.R

Correlation Cost Layer.

layer_correlation_cost(
  object,
  kernel_size,
  max_displacement,
  stride_1,
  stride_2,
  pad,
  data_format,
  ...
)

Arguments

object

Model or layer object
kernel_size

An integer specifying the height and width of the patch used to compute the per-patch costs.
max_displacement

An integer specifying the maximum search radius for each position.
stride_1

An integer specifying the stride length in the input.
stride_2

An integer specifying the stride length in the patch.
pad

An integer specifying the paddings in height and width.
data_format

Specifies the data format. Possible values are: "channels_last" float [batch, height, width, channels] "channels_first" float [batch, channels, height, width] Defaults to "channels_last".
...

additional parameters to pass

Value

A tensor

Details

This layer implements the correlation operation from FlowNet Learning Optical Flow with Convolutional Networks (Fischer et al.): https://arxiv.org/abs/1504.06