artificial intelligence - Tensorflow: How to setup a dataset with label subclasses

Question

I just created a TF model to detect geometric shapes and alphanumeric characters in an image, but I also need to detect the cardinal orientation of those alphanumerics. How do I set up my dataset to classify the orientation as a subclass of characters?

Currently, I just have each shape and letter as a separate label (Square, A, B, Triangle, Rectangle, C, etc). Using Tensorflow and .xml files, how would I add a subclass to a label? Is that where the "pose" field comes into play?

Example .xml I'm using to train

question from:https://stackoverflow.com/questions/65926487/tensorflow-how-to-setup-a-dataset-with-label-subclasses

Answer 1

Waitting for answers