tztztztztz/eqlv2

About lvis dataset

U-Help opened this issue · 1 comments

U-Help commented

Hello, I'm sorry to bother you but I get into trouble these days. As far as I know, AP is the average of per-category precision. That's same for APr, APc, APf. Therefore, we should have the following equation:
image
However, the statistics are against it. Do you know the reason?

tztztztztz commented

Hi, The overall AP is computed by

we do not group all categories into three groups (r, c, f) then calculate the average. Instead, the AP is directly aggregated on all categories.

Actually, the computation is more complicated, you can refer to the implementation details