update pseudo label iteratively during self-training

[kyrawu327]

Hi, have you tried to update pseudo label in an iterative manner?
let's say, the possible procedure goes like this:
teacher model M0 trained with labeled data ---> generated pseudo label ---->semi-supervised training, got model M1 ---->update pseudo label using M1---> semi-supervised training, got model M2 .........................
it is a common practice in semi-supervised classification task, so I wonder if you have tried this iterative approach in object detection? would you mind sharing your insight on this? Thanks a lot.