About the last layer of GNN

[f-GAN]

Thanks for sharing the code, I have a question about the last layer of GNN.
As the code, the output dim of the last layer of GNN is the number of training category C (such as C-way-K-shot). I think the last layer of GNN is equivalent to the classification layer (such as the last fc layer of a classification network). However, this layer is fixed when testing. Shouldn't there be different classification layer for different tasks? I am very confused why you fixed the last layer of GNN in your network.

[vgsatorras]

The last layer outputs the predicted label with respect to the inputted labels of the support set. Let me explain it with an example:

We have a support set of 3 images, each one from a different class (3-Way 1-shot), with the following labels:
x1, y1 = [1, 0, 0]
x2, y2 = [0, 1, 0]
x3, y3 = [0, 0, 1]
Now we predict the label for an image "x" which is from the same class than x2, hence we predict the label [0, 1, 0].

If we run another episode with the same images but swapping the format of the labels:
x1, y1 = [0, 1, 0]
x2, y2 = [1, 0, 0]
x3, y3 = [0, 0, 1]
Now the predicted label for the image "x" will be [1, 0, 0], because it is predicted with respect to the labels of the support set.

Does that answer your question?

[f-GAN]

Thanks, I got it.