Adaptive Cross-Cultural Beauty Fusion Network:A Hybrid Deep Learning Framework for Culturally-Aware Facial Beauty Prediction

In this paper, we propose an adaptive cross-cultural beauty fusion network that addresses the limitations of monocultural beauty assessment models. Our method introduces a dynamic architecture capable of disentangling and reorganizing culturally salient facial attributes through facial region-aware feature extraction, cross-cultural contrastive alignment, and metalearning gating. Our system achieves interpretability through visual attention masks that highlight culturally relevant facial regions and provide transparency to the prediction results. Extensive experiments are conducted on SCUT-FBP5500, CFD and MEBeauty datasets, and the results show that the proposed method outperforms other baseline methods, especially in crosscultural evaluation scenarios. Our work considers different cultural diversity for modeling, advancing the fairness of human facial beauty prediction while maintaining computational efficiency and interpretability.