Understanding of Paper

[Davidyao99]

Hello, sorry to disturb you again but I wanted to ask about this particular table. Could you elaborate a little on how you used motion supervision to train/test the other models in the table? It states here that the model receives the optical flow predicted by RAFT for supervision, but how does the model convert this optical flow into segmentation labels for supervised learning? Are we using the same method explained in the appendix that converts optical flow to segmentation labels here to supervise the other models?

Thank you for your help!! :)))