Binary class segmentation

Hi, I saw that you are not using the background class during training and testing. I am wondering how would you use the model to predict the binary mask for one certain class? For example, given a text query 'apple', the model predicts the mask for the apple. 