GhostFaceNet++: boosting efficiency and accuracy via CSP bottlenecks and Channel Attention | Publicación