4 ms·
Where did META get the Segment Anything training dataset from? Or, if they made it, how did they generate such a massive dataset of accurate masks?
by 55555 4y ago
Where did META get the Segment Anything training dataset from? Or, if they made it, how did they generate such a massive dataset of accurate masks?
- xiphias2 4y agoActually it is quite small dataset by current standards (millions of images), so I guess it was using humans in the loop.
- fkarg 4y agothe relevant part is the masks, and so far it's the biggest dataset in existence, with about 400x more masks than the next smaller one. : excluding non-public ones
- fkarg 4y agothey described it in their paper: they created it themselves. they had/have a loop of having humans annotate images their system was uncertain about, but the released one was fully automated.