imo no. Just use existing image sets to train future models if this becomes too big of a problem.