BDD-X, short for Berkeley Deep Drive-X (eXplanation), consists of 6,970 driving videos with a total length of more than 77 hours. These videos were shot under different driving conditions, such as day/night, highway/city/countryside, summer/winter, etc.

Each video is 40 seconds long on average and contains about 3-4 actions, such as acceleration, deceleration, right turn, etc., all of which are described and explained. The dataset contains 26K activities over 8.4M frames.