5 ms·
We recently overlaid the Microsoft buildings over Boulder Colorado aerial maps and in some places they were pretty good. In others, where there was just the ri
by linsomniac 8y ago
We recently overlaid the Microsoft buildings over Boulder Colorado aerial maps and in some places they were pretty good. In others, where there was just the right amount of tree cover, it was pretty bad. Houses cut up into 2 or 3 or 4 chunks, for example.
It's a great data set, but it'll probably need some curation.
- maxerickson 8y agoFor sure, it will need curation. But even if it were just information about where buildings probably existed it would be hugely useful. An example use case is improving building coverage in a rural area.
- linsomniac 8y agoSure! I wonder if the ML system could make use of information such as lot size and shape (from the county), building sqft, number of levels, information about outbuildings (from the MLS or property tax), and could get smarter. And if it can tell a tree from grass. "This small lot has 3 200sqft buildings separated by a tree on it, but is supposed to have a 1200sqft ranch house... Maybe those 3 buildings define the bounding box of the house?"
- jimmyrocks 8y agoThere is also a good buildings dataset from DRCOG that we're (OSM Colorado) importing into OSM. Reach out to us, or come to our next meetup if you're interested.
- rmc 8y ago> It's a great data set, but it'll probably need some curation. Some sort of crowd sourced, geo database... ;)