Open soulredeem opened 3 years ago
For domain generalization, how could you train the model without same classes between only one source domain and target domain, like Market1501 and DukeMTMC-reID ? Could you release the code of person re-identification experiments?
I didn't reimplement the reID part, but the basic idea is replace the CE loss with triplet loss in some way.
For domain generalization, how could you train the model without same classes between only one source domain and target domain, like Market1501 and DukeMTMC-reID ? Could you release the code of person re-identification experiments?