Dataset labels and predicted vectors from segmentation

[aabbas-cnc]

Hi, thanks for your work,
I converted my own STEP files using your script, and then I've added predict_step to run predictions using the trainer locally (to conveniently transform the input data as you have in the test_step), and the output for one point is a tensor of shape (20,16).

Am I correct in saying that corresponds to 20 faces in the input, and the 16 categories correspond to the 16 categories in the MFCAD dataset?

What is the correspondence between a .bin file (and the faces within) and a label (and the faces in those files)? And so what is the correspondence between the prediction of the model and the bin graph?

Thanks