Why do we need flow = predict_flow2 * 20.0 in flownet_s.py

[OscarWang114]

@sampepose Thank you for creating this amazing repository!

Main question: I would like to know why do we need flow = predict_flow2 * 20.0 (line 107) in flownet_s.py? In the loss function you wrote flow = flow * 0.05. I assume this is for rescaling since we multiplied the flow by 20.0 before, but why do we need to multiply it by 20.0 in the first place? Thank you in advance!