Deep Learning for embedded vision requires large datasets. Indeed the more varied training data is, the more accurate the trained network. But, acquiring and accurately annotating datasets costs time and money. This talk ...