Batch processing for a custom python layer

[Alex Ter-Sarkisov]

I've written my Python layer that transforms the ground truth labels. Looking at other custom python layers (e.g. rpn) I didn't quite understand if I need to add the functionality for processing batches size >1 and outputting the per-image loss for the batch as a vector. Could someone explain perhaps?