The BDD100K data set, made up of 100,000 videos recorded onboard autonomous cars, is now available for download from the University of California, Berkeley.
The data: All of the approximately 40-second-long clips are taken from roads in the US. They contain an array of labeled objects, including 1,021,857 cars, 343,777 signs, 129,262 people, and 179 trains. Lane markings and driveable areas are color-coded in the video, and approximate driving paths are indicated.
Why it matters: Releasing more data will help engineers improve autonomous driving more quickly. Opening this information up could be a positive trend for the industry, signaling that it could be valuing safety above the ongoing competition to be the first to market.