Developer Guide for Intel® Data Analytics Acceleration Library 2018
The forward two-dimensional (2D) transposed convolution layer computes the tensor Y by applying a set of nKernels 2D kernels K of size m 3 x m 4 to the input tensor X.
The problem is to compute the four-dimensional tensor of values Y ∈ R n 1 x nKernels x n 3 x n 4 such that:
For the notations in this formula, refer to 2D Convolution Backward Layer.
The computation flow in the forward 2D transposed convolution layer is identical to the computation of the gradient in the 2D convolution backward layer, except the following notation changes:
2D Convolution Backward Layer |
2D Transposed Convolution Forward Layer |
---|---|
Input gradient tensor G |
Input tensor X |
Gradient tensor Z |
Result tensor Y |
nKernels |
l 2 |
n 2 |
nKernels |