Open lihuiliullh opened 1 year ago
Hi,
The general idea is that increasing the diversity of the dataset can improve the generalization of the learned representations.
In open domain generalization, different domains have different distributions and label sets, which means that each domain contains distinct knowledge but lacks domain knowledge and class knowledge of other domains. This motivates us to design augmentation across domains.
May I know why Domain-Augmentation can improve Open Domain Generalization? What's the idea behind it?