search

xmartlabs / Bender

Easily craft fast Neural Networks on iOS! Use TensorFlow models. Metal under the hood.
https://xmartlabs.github.io/Bender/
MIT License
1.8k stars 90 forks source link

Documentation #10

Closed mats-claassen closed 7 years ago

mats-claassen commented 7 years ago

WIP:

Added basic documentation. This is not complete as there are still some internals undocumented which will make certain things hard to grasp.

bryant1410 commented 7 years ago

which one?

El vie., 2 de jun. de 2017 a la(s) 15:43, Mathias Claassen < notifications@github.com> escribió:

@mats-claassen commented on this pull request.

In Documentation/Supported_Layers.md https://github.com/xmartlabs/Bender/pull/10#discussion_r119926324:

Supported Layers

This is a list of the layers/ nodes that are currently supported. We plan to keep adding more and hope that the community will help with this as well.

- +> In Metal, the basic unit which holds data between layers is the MTLTexture which is contained in an MPSImage. A texture can have an arbitrary amount of feature channels (it is not limited to three or four) + + +The implemented layers are: + Add: Sums two textures element-wise. Both textures have to be of the same size + Convolution: Wrapper for MPSCNNConvolution + ConvTranspose: Represents a transposed convolution + FullyConnected: Represents a Fully connected (or Dense) layer +* InstanceNorm: Instance Normalization as presented in Ulyanov's Instance Normalization

Described in is used in the repo of that stuff

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/xmartlabs/Bender/pull/10#discussion_r119926324, or mute the thread https://github.com/notifications/unsubscribe-auth/ADuX3RccF5_ksfJJ7PoXPY_9e6WbBa7Oks5sAFe3gaJpZM4Nuhev .