Reproduce “SynthFaces: Balanced Large Scale Human Dataset” 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
Sergey Karpukhin, Denis Bobkov, Kirill Morozov, Aibek Alanov
SynthFaces uses ID-Flow to sample face-recognition embeddings on a unit hypersphere, conditioned on age, race, and gender, then generates captioned synthetic images. In a 1,000-sample comparison, it achieves lower label KL and higher identity diversity than balanced-prompt FLUX.2. The 150,000-image dataset improves face similarity after tuning on FairBench, but not Context; it offers no formal privacy guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.