Giters

cvdfoundation / google-landmark

Dataset with 5 million images depicting human-made and natural landmarks spanning 200 thousand classes.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Few annotated labels in test set

MurrayC7 opened this issue · comments

commented

Hi! I am working on the research based on this valuable dataset. However, I found few annotated labels in the test set as for recognition. In summary, 1973 samples have only 849 landmarks compared to the total 200 000 ones. For the rest, 115605 samples' labels are missing. Thus,

  • I wonder if a large number of samples are of non-landmarks.
  • Anyway, I consider there are too few landmarks for testing. Which public test set did your baseline paper use?

Thanks.