This is possible only in very specific and controlled cases. We would need the original uncompressed footage, before Mike encoded it and published it as compressed video. It would also need to have been 16 bits per sample or greater, to provide enough precision for a good deconvolution. And we would need to know the blur radius that was used.

If we don't have all those things, it would be like asking: If the average of two numbers is 10, what were the original numbers?