Different quantitive results between the paper and the leaderboard on SMIYC dataset

[cuttle-fish-my]

Hi, thanks to your effort on this fantastic work! I notice that the quantitive results of SMIYC dataset in paper is as following:

But in the leaderboard of SMIYC, the results of Rba has been boosted significantly(see the following figures):


I'm wondering where did the improvements come from? Have you updated the results with RbA + Mapillary + COCO Outlier Supervision?

Thanks in advance and looks forward to your replay!