Discussion of the 1st-ranked solution

[BB88Lee]

I previewed the top-ranked report (FB-OCC), and appreciate what they shared and how many interesting and innovative attempts!

I noticed that they use SAM to generate semantic labels in their solution (I noticed that they have different stages of training).
A general question is whether the use of unpublic labels conflicts with the first challenge rule: "We allow using annotations provided in the nuScenes dataset."