How can I use prediction targets on an intermediate layer?

I'm trying to implement ArcFace on Matlab for image classification using a custom layer before the softmax loss. In that custom layer I need to do a matrix multiplication between the calculated logits and the targets matrix, but I can't find a way to import them. I can't implement it on the final classification layer because this layer needs trainable parameters and in the template there is no option to define them. Any ideas?

Answer 1

[cui,xingxing]

please reference here:

How to set the input as a two-dimensional feature matrix in the custom Intermediate-layer of deep learning?