| Copyright | (c) Huw Campbell 2016-2017 |
|---|---|
| License | BSD2 |
| Stability | experimental |
| Safe Haskell | None |
| Language | Haskell98 |
Grenade.Layers.Inception
Description
Export an Inception style type, which can be used to build up complex multiconvolution size networks.
- type Inception rows cols channels chx chy chz = Network '[Concat (D3 rows cols (chx + chy)) (InceptionMini rows cols channels chx chy) (D3 rows cols chz) (Inception7x7 rows cols channels chz)] '[D3 rows cols channels, D3 rows cols ((chx + chy) + chz)]
- type InceptionMini rows cols channels chx chy = Network '[Concat (D3 rows cols chx) (Inception3x3 rows cols channels chx) (D3 rows cols chy) (Inception5x5 rows cols channels chy)] '[D3 rows cols channels, D3 rows cols (chx + chy)]
- type Resnet branch = Merge Trivial branch
Documentation
type Inception rows cols channels chx chy chz = Network '[Concat (D3 rows cols (chx + chy)) (InceptionMini rows cols channels chx chy) (D3 rows cols chz) (Inception7x7 rows cols channels chz)] '[D3 rows cols channels, D3 rows cols ((chx + chy) + chz)] Source #
Type of an inception layer.
It looks like a bit of a handful, but is actually pretty easy to use.
The first three type parameters are the size of the (3D) data the
inception layer will take. It will emit 3D data with the number of
channels being the sum of chx, chy, chz, which are the number
of convolution filters in the 3x3, 5x5, and 7x7 convolutions Layers
respectively.
The network get padded effectively before each convolution filters such that the output dimension is the same x and y as the input.