Hacker News new | ask | show | jobs
by bskrobisz 972 days ago
There's plenty of knowledge to be gained apart from that, there are top players that explain some of their thinking that have content on Youtube/Twitch. There are other examples along the lines of what you're saying -- traffic markings like bollards, signs, lines, license plates, etc. vary country-by-country; the position of the sun along with the date the google maps van took the image, telephone poles and other infrastructure. What's really interesting to me, though, is that there's some information that feels similar to when ML fits to metaknowledge about your training set -- e.g. there are some areas which can be ruled out entirely by the date the picture was created, there are certain areas where privacy laws alter what maps can provide or must blur, and there are areas where there was some issue with the google maps equipment that caused artifacts in the images (e.g. there are 'rifts' in the sky in some parts of SEA)