The nuScenes dataset has achieved widespread acceptance in academia and industry as a standard dataset for AV perception problems. To advance the state-of-the-art on the problems of interest we propose benchmark challenges to measure the performance on our dataset. At CVPR 2019 we organized the nuScenes detection challenge. The nuScenes tracking challenge is a natural progression to the detection challenge, building on the best known detection algorithms and tracking these across time. 

Note that this page is a simplified version of the official tracking task page.


The tracking task and challenge are a joint work between Aptiv (Holger Caesar, Caglayan Dicle, Oscar Beijbom) and Carnegie Mellon University (Xinshuo Weng, Kris Kitani). They are based upon the nuScenes dataset and the 3D MOT baseline and benchmark.


The nuScenes tracking evaluation server is open all year round for submission. To participate in the challenge, please create an account at EvalAI. Then upload your zipped result file including all of the required meta data. The results will be exported to the nuScenes leaderboard. This is the only way to benchmark your method against the test dataset. Results and winners will be announced at the AI Driving Olympics Workshop (AIDO) at NeurIPS 2019.


To find out more about the submission rules, results format, classes, evaluation metrics, baselines and leaderboard, please refer to the official tracking task page.