Doubt about results in Table.2

[Samuel-Recovery-123]

Could you please provide the code of evaluation protocol? Subset of 20% is one part of the Market-1501 training dataset, but PCDMs trains on the whole training. There seems a data leakage, since you know the whole training set distribution after training PCDMs. It's unfair to compare with other data augmentation methods. Do other methods you compare in Table.3 adopt the same strategy? I find that the results of other methods are from PIDM, but PIDM doesn't release the code .The Standard method reports 76.7% mAP on full dataset, and it's much lower than the BoT strong baseline (88% mAP). Which baseline code do you refer to? Did you train BoT without initializing weights pretrained on ImageNet. I am much confused with it.