The Annotated Semantic Dataset is composed of $11$ videos, divided in $3$ activity categories: Biking; Driving and Walking, according to their amount of semantic information. The classes are: $0p$, which represents the videos with approximately no semantic information; $25p$, for the videos containing relevant semantic information in ∼$25%$ of its frames ; the same ideia for the classes $50p$ and $75p$, The videos were record using a GoPro Hero 3 camera mounted in a helmet for the Biking and Walking videos and attached to a head strap for the Driving videos.