Traffic4cast 2021 competition is now open. The competition data are available for download.
The competition dataset covers 10 diverse cities around the world in a time span of 2 years. The data include dynamic and static information. The dynamic data are derived from the GPS trajectories of a large fleet of probe vehicles, aggregated into spatiotemporal cells. Each cell corresponds to the area of approximately 100m x 100m and the time interval of 5 minutes. The dynamic information is encoded in 8 channels with the traffic volume and average speed per heading direction. The static information describes the properties of the road maps. It is encoded in 9 channels and is split only into spatial cells.
Traffic4cast 2021 comprises two transfer learning challenges. For all 10 cities, we provide the static information on road geometry in two resolutions. For 4 cities, we provide dynamic data for training spanning 6 months from 2019 and 6 months from 2020. The core challenge focuses on temporal domain shift in traffic due to COVID-19. For 4 different cities, we provide 6 months of training data from 2019 plus 100 1-hour test slots from 2020. The challenge is to predict the dynamic traffic states 5, 10, 15, 30, 45 and 60 minutes into the future after each test slot for each city. An extended challenge presents a few shot learning challenge with a spatiotemporal domain shift. Dynamic traffic states need to be predicted for 2 new cities looking 5, 10, 15, 30, 45 and 60 minutes into the future after 100 1-hour test slots. The test slots are randomly chosen between 2019 and 2020 and are the only dynamic data provided for these cities. Learn more…
