Microsoft and City, University of London have released ORBIT, a dataset of annotated videos depicting various objects. ORBIT is designed to train AI applications for individuals in the blind or low vision community with only a few examples. It contains 3,822 annotated videos of 486 objects recorded by individuals who are blind or have low vision. Each video was shot on a smartphone in varied environments to replicate real-world conditions and better train future object recognition systems.
Image credit: Flickr user Bill Smith