Closed YurongYou closed 6 years ago
Hi,
What you suggest is actually correct for evaluation. It accords precisely to what I described in my paper.
But in the current evaluation, I would rather keep the persons in the same query camera in the database. It helps simplify the code and enlarges the actual database.
I think both practices are acceptable.
Yes, I agree. Thanks for your prompt reply!
Sure. I would suggest you use the provided evaluation package so that your results are comparable with other people.
Hi!
From the evaluation protocol part of your paper, it is said that
But from https://github.com/liangzheng06/MARS-evaluation/blob/3a91bbc762fad7629fecde8c53e170742d3700f9/utils/evaluation_mars.m#L12-L18
it does not mark the feature of persons of a different label but with the same camera as ``junk'', which will make the computed CMC lower than using the setting of cross-camera search.