Geometric prior guided hybrid deep neural network for facial beauty analysis

Facial beauty analysis is an important topic in human society. It may be used as a guidance for face beautification applications such cosmetic surgery. Deep neural networks (DNNs) have recently been adopted facial and achieved remarkable performance. However, most existing DNN-based models regard normal classification task. They ignore prior knowledge traditional machine learning which illustrate the significant contribution of geometric features analysis. To specific, landmarks whole organs are introduced to extract make decision. Inspired by this, we introduce novel dual-branch network analysis: one branch takes Swin Transformer backbone model full global patterns, another focuses on masked with residual local patterns certain parts. Additionally, designed multi-scale feature fusion module can further facilitate our learn complementary semantic information between two branches. In optimisation, propose hybrid loss function, where especially regulation regressing it force extracted convey features. Experiments performed SCUT-FBP5500 dataset SCUT-FBP demonstrate that outperforms state-of-the-art convolutional models, proves effectiveness proposed regularisation structure network. best knowledge, this first study Vision into