lvis-dataset/lvis-api

AP inconsistency

Closed this issue · 2 comments

❓ Questions and Help

The results of eval.AI is inconsistent with results of lvis api. Do you use different evalution code on eval.AI?

We're following up privately on email to get specific information. Will update here with our findings.

Confirmed in the Google Group that the inconsistency was due to using code from before the bug fix in 73142a4.