Reproduce “SCUT-FBP5500: A Diverse Benchmark Dataset for Multi-Paradigm Facial Beauty Prediction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lingyu Liang, Luojun Lin, Lianwen Jin, Duorui Xie, Mengru Li
Researchers at South China University of Technology introduced SCUT-FBP5500, a benchmark dataset of 5500 diverse faces (Asian/Caucasian, male/female) with multi-paradigm labels including beauty scores and 86 facial landmarks, supporting robust facial beauty prediction research. Deep learning models achieved a Pearson Correlation of 0.8997 using this dataset.
Key insights
- FBP is a multi-paradigm computational problem, and a comprehensive dataset with diverse labels (scores, distributions, landmarks) is essential for developing versatile and robust models.
- Diversity in facial properties (race, gender) is crucial for developing FBP models that generalize well across different populations, supporting specialized studies on cross-cultural perception.
- Face shape, captured by facial landmarks, significantly contributes to facial beauty perception, consistent with psychological and computer vision findings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.