What is feature alignment in domain adaptation for vision models?

Feature alignment reduces distribution gap between source and target domains.

Domain adaptation addresses performance degradation when training and test data distributions differ. Feature alignment methods match feature distributions using techniques like adversarial training, MMD loss, or normalization alignment. This helps models generalize across different environments such as synthetic-to-real transfer.