Brief

Researchers have developed a new one-class face anti-spoofing method called UFDANet. This technique disentangles liveness and domain features to improve robustness against unseen attacks. UFDANet also synthesizes novel liveness and domain features to enhance generalization capabilities, achieving performance comparable to existing two-class methods. AI

Researchers have developed a new method for accurately measuring heart rate from very short video clips, specifically 2-second segments. The approach addresses challenges like limited heartbeat cycles and spectral leakage by enforcing periodicity between estimated and ground truth signals. A generator is also included to reconstruct longer signals from the short ones while maintaining periodic consistency, leading to improved heart rate measurement. Experiments on benchmark datasets show this method achieves state-of-the-art performance in ultra-short video heart rate estimation. AI