Perception software in assistive devices is usually tested on scenes recorded by cameras on cars and phones. KF-LV is recorded where blind and partially sighted people most often find that software fails them, and every white cane, guide dog and harness in it is annotated by hand.
The scenes were recorded in Chiba, Osaka, Nairobi and Lalitpur by volunteers, some of them with low vision, with the consent of everyone identifiable in them. Faces are blurred. The set is available to researchers under a data use agreement that forbids any attempt to identify anyone.
Blackwall will use KF-LV in its future evaluations of perception models in medical implants, including the next version of SHIKAI.