how to use this model in to video?

[pengfeidip]

inference every frame ,then combine them into a video is a common methond, but this will cause incontinuity。

please, any method get a better result in video?