To bridge this gap, our research analyzes the primary sources of label noise in real-world settings, which include a) instantiated identities, b) blurry infrared images, and c) annotators’ errors. In ...