will the model still perform good from the training images that the same person divided to two different id folders?

[LancasterLiu]

I have added some additional datasets, but I just realized that there may be some images that should be in the same id folder were put in two different id folders. So I wonder can the model learn the interclass features, which means that will the model still perform good from the training images that the same person divided to two different id folders? btw I used Resnet50.

[layumi]

@LancasterLiu Thank you. Sorry for the late response. I think it will be fine if there are only a few such cases.
But it may be good to avoid strong losses such as hard negative triple loss.