rank-m problem

[Phoebe-star]

rank-m problem

[zhunzhong07]

In our code, we use a new protocol splits for CUHK03 dataset, this protocokl split the dataset into training set and testing set similar to that of Market-1501, which consist of 767 identities and 700 identities respectively.

Our new protocol is different from the one in open-reid.

[Phoebe-star]

I know you have new protocol , but I use your IDE features ( fixed features ) to compute CMC .
I use the two cmc code , they have different value

[zhunzhong07]

In open re-id, you should use the same testing protocol as the new protocol, and use the 'Market-1501' metric in the ope re-id.

[Phoebe-star]

1.you say the testing protocol of open re-id, is used by everyone ( old )
and your protocol is the new
so use cmc method is not match
so open re-id cmc method 's protocal is the same as your , because rank-1 are 38.1%
Person-Reid cmc method 's protocal is not the same as your , because rank-1 are 63% and 38.1%

2. do you know how to choose cuhk03 dataset for training and cmc ( validation cmc ) ? (general)

[zhunzhong07]

The open re-id is the general way to split CUHK03 dataset, which have 1000+ identities for training, and 100 identities for testing, and needs to training/testing 20 runs.