Question about 'Shared Attention'

[HIT-lcccc]

Thanks for your excellent work! There are some questions about the Shared Attention when I read the paper.

1. In the formulas of Section3.3, the reference task stream r are used to calculate attention, and task stream t gets y^t based on the referred attention. However, the figure 3 illustrates the y^r which is not explained in formulas. Was it used in task stream t? If so, what is the relationship between y^t and y^r ?
2. The released code is about segmentation based on MMSeg. Would you tell me the location about the details of Shared Attention on multi-task learning?
   Best wishes.