Currently neither video, nor image can have zero weight.
This does not cover the case, when only one type of files makes sense to be used.
For example for action recognition can only be performed by videos, so images are useless and would rather be valued as simply 0 weight.
This issue proposes to add such functionality.