Enhancing Model Performance of Person Re-Indentification on Unknown Target Domain

Person re-identification(ReID) is the task that aims at retrieving the same person from the images taken across different cameras. Benefiting from the improvement of deep learning algorithms and the appearance of large datasets, the performance of ReID models has been greatly improved. However, most ReID models focus on a single dataset and their performance will drop dramatically when the train-set and test-set are from different datasets. To improve the generalization ability of the ReID model, this paper proposes a method that takes the advantage of triplet loss and multi-dataset training. And the experiment results show that this method can enhance the model performace in cross dataset usage.