Exact Feature Distribution Matching for Arbitrary Style Transfer and Domain Generalization
The official codes of our CVPR2022 paper: Exact Feature Distribution Matching for Arbitrary Style Transfer and Domain Generalization
One Sentence Summary: EFDM outperforms AdaIN, which only matches first and second order statistics, by implicitly matching high orders statistics in an efficient manner.
 |
|---|
| Fig.1: AdaIN vs. histogram matching vs. EFDM via Sort-matching. |
A brief introduction:
Many real-world tasks (e.g., Arbitrary Style Transfer and Domain Generalizaiton) can be cast as a feature distribution matching problem.
With the assumption of Gaussian feature distribution, conventional feature distribution matching methods usually match the mean and standard deviation of features.
However, the feature distributions of real-world data are usually much more complicated than Gaussian, which cannot be accurately matched by using only the first-order and