Pretty sure it's a small set of images of each type as well. It's probably easier for a dumb heuristics-based bot to beat it than it is for most people.
Right. It says select tiles with cars, bikes or buses and I fail because I recognized one far away. Traffic lights and stairs, if a tile has just a tiny bit of the thing, does that count? I don't think they train on this, they've got their answers and they just want people to feel dumb, I guess.