Comment by Huongngtm

My hypothesis is that we will first use visual recognition for a few highly relevant fields. But after that we'll expand to almost all objects.