About evaluation metric about person_reid_baseline_pytorch HOT 4 CLOSED

Hi @kentangSJTU ,
Market-1501 Dataset followed the mAP calculation in Oxford 5k dataset.
You may refer to http://www.robots.ox.ac.uk/~vgg/data/oxbuildings/compute_ap.cpp

from person_reid_baseline_pytorch.

Thanks a lot for your timely reply! It's very kind of you~

from person_reid_baseline_pytorch.

@kentangSJTU @layumi
hello, in compute_mAP, why do we ignore the junk_index? the code like below:
mask = np.in1d(index, junk_index, invert=True)
index = index[mask]
and why do like this:
cmc[rows_good[0]:] = 1
instead of
cmc[rows_good] = 1,
could you please more explanation about this? thanks a lot.

from person_reid_baseline_pytorch.

Hi @venuszhou ,

The physical meaning of CMC is whether the true matches appear before Rank-K.
Therefore, if the Rank 6th image is true match, CMC[5:] = 1. All CMC after 5 is set to 1. (We start from zero)

from person_reid_baseline_pytorch.