How to perform weighted resampling properly on a highly imbalanced multilabel dataset?

I have a highly imbalanced multilabel dataset of images. Each image can be assigned one or more labels. There are 15 labels in total. I encoded the labels as binary vectors. {1,0}^15. I tried to do weighted resampling using sample weights method, but I realised it accepts weights as a one dim vector ,i.e, [w1(y1), w2(y2),…, wn(yn)]. This is suitable for multi-class classification but not for multi label classification.
Could someone guide me about how to proceed ?