This is the leaderboard of the INTERPRET Challenge trajectory prediction for the ICCV Stage. There are four tracks, including Single-Agent Track, Conditional Single-Agent Track, Multi-Agent Track, and Conditional Multi-Agent Track.

Participants can click on the header of each metric to re-rank all methods based on the selected metric. The button "scene_type" can change the test data composition. "Regualar" means that the scores are obtained by using a portion of the test data which has the same locations and scenarios as those of the train/val data, while "Generalizability" indicates scores obtained by using the other portion of the test data with different locations and scenarios. "All" means using all the test data, and all of the tracks are ranked with "All" by default.

Single-agent and conditional single-agent tracks' metrics can be found in the INTERPRET's Github Repo. The final ranking of these two tracks is based on the MR.

Multi-agent and conditional multi-agent tracks' metrics can be found in the INTERPRET's Github Repo. The final ranking of these two tracks is based on the Consistent-minJointMR.