SCUT-FBP5500: A Diverse Benchmark Dataset for Multi-Paradigm Facial Beauty Prediction

Facial beauty prediction (FBP) is a significant visual recognition problem tomake assessment of facial attractiveness that is consistent to humanperception. To tackle this problem, various data-driven models, especiallystate-of-the-art deep learning techniques, were introduced, and benchmarkdataset become one of the essential elements to achieve FBP. Previous workshave formulated the recognition of facial beauty as a specific supervisedlearning problem of classification, regression or ranking, which indicates thatFBP is intrinsically a computation problem with multiple paradigms. However,most of FBP benchmark datasets were built under specific computationconstrains, which limits the performance and flexibility of the computationalmodel trained on the dataset. In this paper, we argue that FBP is amulti-paradigm computation problem, and propose a new diverse benchmarkdataset, called SCUT-FBP5500, to achieve multi-paradigm facial beautyprediction. The SCUT-FBP5500 dataset has totally 5500 frontal faces withdiverse properties (male/female, Asian/Caucasian, ages) and diverse labels(face landmarks, beauty scores within [1,~5], beauty score distribution), whichallows different computational models with different FBP paradigms, such asappearance-based/shape-based facial beauty classification/regression model formale/female of Asian/Caucasian. We evaluated the SCUT-FBP5500 dataset for FBPusing different combinations of feature and predictor, and various deeplearning methods. The results indicates the improvement of FBP and thepotential applications based on the SCUT-FBP5500.