Face Shape Classification Using Swin Transformer Model

Abstract

Face shape classification plays a crucial role in various applications, including facial recognition and personalized beauty rec- ommendations like hairstyles. In some cases, it is tricky to establish one's genuine facial shape, making it difficult to further process this information. This paper attempts to harness the advantages of the Swin Transformer to classify face shapes with higher accuracy. We also compared the test results with augmentation to evaluate the improvement. Our proposed method obtained 86.34% accuracy with augmentation, which is decent for further processing in various applications that require face shape recognition.