
The performance of face recognition models trained on synthetic datasets is still inferior to those trained on real-world datasets, indicating a gap in the diversity and realism of synthetic data. This study aims to understand this gap by analyzing the distribution of soft-biometric attributes across real and synthetic face datasets.


coremsg

assessing-the-diversity-and-limitations-of-synthetic-face-datasets-compared-to-real-world-datasets


Assessing the Diversity and Limitations of Synthetic Face Datasets Compared to Real-World Datasets
