Skip to content
This repository was archived by the owner on Jul 2, 2024. It is now read-only.
This repository was archived by the owner on Jul 2, 2024. It is now read-only.

MLPs wrong skip connection #27

Description

@SirSykon

Hello,

According to paper, MLPs skip connection is on layer 2 (starting from 0 is the third one) but using nerfactor I've seen strange behaviour so I checked the code. Network .call() in mlp.py does the following:

    x_ = x + 0 # make a copy
    for i, layer in enumerate(self.layers):
        y = layer(x_)
        if i in self.skip_at:
            y = tf.concat((y, x), -1)
        x_ = y
    return y

So the concatenation is applied after calling the layer and therefore the true skip connection is at the next layer (the fourth one).

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions