How to handle missing masks in Multi task model keras

Question

I ran into a problem when I tried to implement a Multi-Task Model based on Unet. This Model includes:

1. Segmentation branch
2. Reconstruction branch
3. Cl