Researchers often package their implementation (PyTorch or TensorFlow code) and pre-trained models in archives for distribution on platforms like GitHub or ResearchGate.
It proposes the Balanced Binary Angular Margin (BBAM) loss. This loss function helps balance the "variance bias" between positive and negative samples by looking at the angle of features in a multi-label setup. BBAM.rar
A more recent 2024 paper titled introduces a different "BBAM". BBAM.rar