We are excited to announce the opening of our Traffic4cast competition on June 15, 2021. Traffic4cast pushes the most advanced methods in machine learning to model complex geo-spatial processes, contributing to addressing the most pressing problems of humanity. Building on its success at NeurIPS 2019 and NeurIPS 2020, Traffic4cast is going into its third year offering new challenges and opportunities. Going beyond the goals of the past competitions, Traffic4cast 2021 will explore the models that adapt to domain shifts in space and time.
The 2021 competition comprises two transfer learning challenges of increasing complexity. In the core challenge, the models are required to handle a drastic temporal domain shift due to Covid-19 pandemic for four cities, based on only few examples from the new domain. In the extended challenge, the models must predict traffic flows in two entirely new cities based on few examples. The training data are available for four different cities before and during the pandemic.
The competition thus brings together a range of highly active fields in machine learning, from few shot learning and transfer learning to video frame prediction and graph based modeling.
This year our Traffic Map Movies have been derived from more than 1012 data points and cover 10 diverse cities around the world from 2019 through 2020, providing an order of magnitude more data compared to last year. The Traffic Map Movies represent a high resolution privacy preserving comprehensive view of urban traffic, aggregating the GPS trajectories from a large fleet of vehicles into spatiotemporal cells. Each cell contains eight dynamic channels encoding traffic speed and volume per direction and nine static channels encoding the properties of the road maps.
Traffic4cast 2021 community forums are now open! We have already published several posts discussing the challenges and the data. Join the discussion.