Contents

BNNSDataLayoutConvolutionWeightsOIHrWr

A constant that represents a 4D array of rotated convolution weights.

Declaration

var BNNSDataLayoutConvolutionWeightsOIHrWr: BNNSDataLayout { get }

Discussion

The value (kx, ky, InChannel, OutChannel) is at index:

(kw - 1 - kx) * stride[0] + (kh - 1 - ky) * stride[1] + OutChannel * stride[2] + InChannel * stride[3].

  • size[0] is the convolution kernel width in pixels.

  • size[1] is the convolution kernel height in pixels.

  • size[2] is the number of output channels.

  • size[3] is the number of input channels.

  • kw is size[0] and kx is between 0 to kw-1.

  • kh is size[1] and ky is between 0 to kh-1.

See Also

4D Data Layouts