I’m currently working on a multi-label image classification problem where my dataset is significantly imbalanced. Each image in my dataset can have multiple labels. The labels are provided in a one-hot encoded format.