Fig. 1: Leaderboard scores as a function of team ranking.

Private and public leaderboard (mean average precision) scores as a function of the performance ranking of the teams. A and B show the scores on the private test set and C and D show scores on the public one. B and D zoom in on the top 15 ranked teams. The top 5 models in the private leaderboard all appeared in the top 8 entries in the public leaderboard, with some variations in the order, as indicated by the color in the zoomed-in figures (right panels). Source data are provided as a Source Data file71.