![]() Trajectory from live to spoof) to be the same for all domains. Work, instead of constructing a domain-invariant feature space, we encourageĭomain separability while aligning the live-to-spoof transition (i.e., the Test domain, which backfires on the generalizability of the classifier. ![]() The training data, we show that the feature shift still exists in an unseen ![]() Though learning a domain-invariant feature space is viable for Metric learning or adversarial losses to remove them from feature Most prior works regard domain-specific signals as a negative impact, and apply On domain gaps, such as image resolution, blurriness and sensor variations. Download a PDF of the paper titled Rethinking Domain Generalization for Face Anti-spoofing: Separability and Alignment, by Yiyou Sun and 4 other authors Download PDF Abstract: This work studies the generalization issue of face anti-spoofing (FAS) models
0 Comments
Leave a Reply. |