I had the same question and was happy to see an ML article that actually answered this :)
FTA
"The team trained the machine learning program...by providing it about 370,000 images... labelled as either having or not having a solar panel present"
So it seems like the Stanford team that did the work built a large ground truth dataset
2 comments
[ 2.6 ms ] story [ 14.5 ms ] threadFTA "The team trained the machine learning program...by providing it about 370,000 images... labelled as either having or not having a solar panel present"
So it seems like the Stanford team that did the work built a large ground truth dataset