Are we allowed to fetch topk of each query


When we generate result, are we allowed to only keep topk of each query (makes 50k * n scores), or must we put all 50k * 100w scores together as the final result?

I have this question because the rule says each reference are independent. I compared the references to get the topk, but I did not compute scores based on other references.

Hi @coin - Is this for the Matching Track? You may submit up to 500K rows. It’s up to you how many rows to submit.

As you mentioned, each individual score should be generated independently.

Submitted individual predictions may not take into account more than one query image or more than one reference image at a time.

Hope that helps!