Layer

Layer objects are the main objects of the framework.

Inside ACETONE, the neural network is represented by a list of Layer objects, each of then having links to their predecessor(s) and descendant(s).

Notation

Most of the layers presented in this Wiki takes as input and returns a 4D tensor of size $(N,C,W,H)$.

However, the first dimension is always supposed to be a Batch dimension, not used in the framework. Thus, for simplicity, unless stated otherwise, the first dimension will implied in the notation.

Example:

Let T be a 4D tensor of size $(N,C,W,H)$.

For $z < C$, $x < H$ and $y < W$, $T(z,x,y)$ is the element $T[0,z,x,y]$.