On your statement about including mask loss in bipartite matching

[MLDeS]

I have a question on the inclusion of mask loss in bipartite matching. You mentioned in the paper in page 5 last sentence that "The participation of mask loss may affect the global matching between the object queries and ground-truths.". Could you please say if this is an empirical finding? Basically, I want to know why do you think this would be true?

Thanks a lot in advance!