Out of curiosity I looked up some of the damage images in the satellite view of Google Maps by their latitude and longitude coordinates. Not only are those images available, but they are in general also higher resolution and better quality than the ones in the lab data. It occurred to me that this would be a way to provide ‘no damage’ images corresponding to ‘damage’ images if the former are not available. They may correspond to before or after images of the damage ones, but either should work.
Hi @jlammens,
I believe the dataset used in the lab was with the intent that it should not be memory heavy (as Coursera’s platform is limited in resources). The notebook’s intention is to teach and show how things are done, which I believe, it fulfils. But yes, when working professionally, the higher resolution images are better.
Best,
Mubsi