StyleDiff: Attribute comparison between unlabeled datasets in latent disentangled space

Keisuke Kawano, Takuro Kutsuna, Ryoko Tokuhisa, Akihiro Nakamura, Yasushi Esaki

Abstract: Highlights•StyleDiff presents the differences between datasets in a human-understandable manner.•StyleDiff extracts latent attributes with different distributions between datasets.•StyleDiff is computed fast enough for large datasets containing tens of thousands.•StyleDiff outperforms the existing methods in the quantitative evaluations.