Calculation of masked classification score

[pmgautam]

I'm not able to understand how the masked classification score is computed. In the paper it says input images are masked by the corresponding segmentations. How can the images be classified if they are masked? Can anyone help me understand this? Thanks in advance!