Skip to content

MatrixMultiplicationLayer.md

Latest commit

 

History

History
37 lines (23 loc) · 1.33 KB

MatrixMultiplicationLayer.md

File metadata and controls

37 lines (23 loc) · 1.33 KB

CMatrixMultiplicationLayer Class

This class implements a layer that performs matrix multiplication operation for its two inputs. It treats first three dimesions (BatchLength * BatchWidth * ListSize) as a matrix number, next three (Height * Width * Depth) as row number and last one (Channels) as a number of columns. So this layer implements a bunch of independent matrix multiplications.

Settings

This layer has no settings.

Trainable parameters

This layer has no trainable parameters.

Inputs

This layer has 2 inputs. Each of them accepts the set of matrices:

  • BatchLength * BatchWidth * ListSize - amount of matrices in sets, must be equal between inputs
  • Height * Width * Depth - amount of rows
  • Channels - amount of columns
  • Channels of the first input must be equal to Height * Width * Depth of the second input

Outputs

Single output returns the blob with multiplication results. The size of the blob:

  • BatchLength, BatchWidth, ListSize, Height, Width, Depth are equal to these dimensions of the first input
  • Channels is equal to the Channels of the second input.