简介

简介#

备注

转置卷积层(transposed convolutional layers,也被称为分数步长卷积层(fractionally strided convolutional layers))

神经网络的中心思想是 仿射变换 (affine transformations):\(Y = XW + b\)

对任何卷积运算 \(i, k, p\)\(s\),

\[ o = \lfloor \frac{i - k + 2p}{s} \rfloor + 1 \]