Keras : Custom weighted binary crossentropy loss for multi-label classification

Question

I am working on an Imbalanced Multilabel classification problem

These are the amount of images present for each class :

This is how the positive and negative w