Who is the genius that says image ID costs dollars per image? A grid of 49 images is presented.ID the top left image (the longest time), The other 48 frames are scanned and ticked if they arethe same as or close to the top left. End click and all the identified images go away and the field is populated and the task repeated. Same or new top left does not matter, you have 500 of these 'tiny Turks' each dealing with 49 images in 3-5 seconds. Duplication can be used to polish it. You can also only use images grabbed on roads at all angles. This can be highly automated and optimised (slows and error makers = assigned other jobs). This will get you down to pennies per image and if an AI is trained into it, even less.