If you start with a video you can do this. You'd transform an X x Y x T dataset (X x Y dimensional frames shown over a period 0-T) into a T x Y x X dataset (T x Y dimensional frames shown over a period 0-X).
For a long video you'd get a very wide image that varies over a pretty short duration. A 1920x1080 video clip that's 80 seconds long @24 fps would come out right. It'll still be 1920x1080 and 80 seconds long after transformation.
For a long video you'd get a very wide image that varies over a pretty short duration. A 1920x1080 video clip that's 80 seconds long @24 fps would come out right. It'll still be 1920x1080 and 80 seconds long after transformation.