|
MayaFlux 1.0.0
Digital-First Multimedia Processing Framework
|
One grouped convolution layer over [batch, c_in, t_in], with weights [c_out, c_in / groups, kernel]. More...
#include <Layers.hpp>
Public Attributes | |
| Activation | activation { Activation::None } |
| uint32_t | c_in {} |
| uint32_t | c_out {} |
| uint32_t | groups { 1 } |
| uint32_t | kernel {} |
| uint32_t | pad { 0 } |
| uint32_t | stride { 1 } |
| uint32_t | trim { 0 } |
One grouped convolution layer over [batch, c_in, t_in], with weights [c_out, c_in / groups, kernel].
The channels are split into groups equal runs and each output channel reads only the input channels of its own run. The input is zero padded by pad on both sides and windows advance by stride. trim drops that many outputs from the end.
Definition at line 51 of file Layers.hpp.