About the all-in-one training details

[HUANG HONGZHAN]

Hi, I am confusing about the "In other words, AirNet under AIO is the model trained on the collection of all the datasets that
consists of three corruptions (i.e., noise, rain, and haze) with different degradation levels (i.e., σ = 15, 25, 50)."
Is it true that the three corruptions datasets is mixed up in the trainning procedure ？