Detail of augmenting data by translating a single frame over time

[lianqiyan]

I can't understand how the video data were processed in this way, You mentioned in your paper that it is benefcial to improve temporal coherence of the output.