Domain-invariant Feature Refinement Model for Generalizable Person Re-Identification

In normal generalization work, domain-variant information is always ignored by the most method. In this paper, we propose a domain-invariant feature refinement model to solve this issue, a two-branch framework for dealing with domain generalization problems, which uses the instance normalization layer to split the global feature into domain invariant feature and domain variant feature. This method can make effective use of Domain-invariant features and further optimize the performance of the task in person re-identification. This paper is also performed several ablation experiments to verify the improvement of model performance, Through the experimental results, we can see that Feature splitting is the key to the effectiveness of our method.