Do you use cuhk03 evaluation metric or market1501 evaluation metric? about person_reid_baseline_pytorch HOT 4 CLOSED

The evaluation metric is wrote for Market1501, DukeMTMC-reID and CUHK03-NP.
It is not proper for CUHK03. It needs training 20 models (based on different training set) and testing 20 models (different testing set). It reports the average rank-1, rank-5 and rank-10 score.
If you want to run CUHK03, you may refer to the caffe evaluation code (You may google it).

from person_reid_baseline_pytorch.

Maybe I do not clarify the point. The cuhk03 and market evaluation metric I mentioned above are two different methods that compute the cmc curve. See the url: https://cysu.github.io/open-reid/notes/evaluation_metrics.html.

I wonder which metric you use in these experiments. ( I guess it is the market evaluation metric.)

from person_reid_baseline_pytorch.

Hi @Wanggcong ,
Since the code runs on Market1501, we use the evaluation metric based on Market1501.
I converted the original code(Matlab version) to python version.
I checked the result. It achieves the same Rank1 and mAP when evaluating on the original Matlab code.

Note that the evaluation code in Open re-id may be slightly different from me.

from person_reid_baseline_pytorch.

Thanks for your help! @layumi

from person_reid_baseline_pytorch.