Each example shows: input video, first frame with trajectory, and edited video.
The full pipeline editing is applied to the input video, incorporating both cross-view motion transformation and video resynthesis.
Even though our method is trained on single object editing, it can be applied to multi object editing with zero-shot manner.
The red point is marked to highlight a reference point for easy to draw correspondance.
We compare solely against resynthesis baselines, conditioned on the object boxes from the input video.