Zero values for compactness loss

[NivC]

Hi,

Thank you very much for the code and interesting paper!

We are trying to run your method on our dataset, but seem to often get zero values for compactness loss.
This is specifically a problem when we try to normalize the "feature_distance_list" during inference.

Could you please advise?

Also, if you have any suggestions how to scale the number of epochs with the size of the dataset, please share.

Niv.