Computer vision startups need a lot of data to be successful in the long run. However, I've talked a to few startups trying to go from zero data to one billion datapoints instead of taking the first steps from zero to a hundred pieces of labeled data.