Also when only a tiny part of the wheel is in the corner of a square, does it count? What about the motorcycle’s mirror? Apparently it doesn’t most of the time. But sometimes it does.
Also the absolute worst are those “click all the squares that have a <object> in it until they’re all gone” Then when you click on them the image disappears and sometimes you have to wait way too long for another image to finally appear in its place. You know when the images take too long to appear like that it will fail you no matter what, probably because you “took too long” to solve the captcha.
its likely just used for training data for Waymo, hence the traffic related stuff. It’s likely one of the several reasons why Waymo as an automated driving car company is further in development than many other companies.
Also when only a tiny part of the wheel is in the corner of a square, does it count? What about the motorcycle’s mirror? Apparently it doesn’t most of the time. But sometimes it does.
Also the absolute worst are those “click all the squares that have a <object> in it until they’re all gone” Then when you click on them the image disappears and sometimes you have to wait way too long for another image to finally appear in its place. You know when the images take too long to appear like that it will fail you no matter what, probably because you “took too long” to solve the captcha.
i approach it as training data for vision systems. Answer it in the following way:
“Select the reasonable amount of the object such that that object would be identified by a machine”
so frames with very little parts will typically not count
Where do we want the machine to start “seeing” a car or cyclist? I’d like them to pay attention to egdes.
its likely just used for training data for Waymo, hence the traffic related stuff. It’s likely one of the several reasons why Waymo as an automated driving car company is further in development than many other companies.