Asking about motion dynamics position embedding

[dreamyou070]

In the paper, you project motion magnitude to the position embedding.
I am little confusing...
motion magnitude means hoe the region (from the gt dataset video) is movable.
is that means, you project the motion magnitude using position embedding,
as if the position axis is not spatial but motion(dynamics)