Reproduce “FFHQ-Makeup: Paired Synthetic Makeup Dataset with Facial Consistency Across Multiple Styles” 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
Xingchao Yang, Shiori Ueda, Yuantian Huang, Tomoya Akiyama, Takafumi Taketomi
FFHQ-Makeup presents a large-scale paired synthetic makeup dataset, comprising 90,000 bare–makeup image pairs, generated using an improved diffusion-based transfer method. This dataset aims to address the scarcity of high-quality paired data by prioritizing facial consistency and makeup realism for computer vision applications in beauty.
Key insights
- The 3DMM-based makeup residual representation effectively disentangles makeup appearance from facial identity, preventing identity leakage during the transfer process.
- An augmentation strategy involving re-rendering makeup residuals on diverse 3D facial geometries is crucial for enabling the network to learn makeup features that are invariant to specific facial structures.
- Combining a diffusion model with structural controls (ControlNet) and a specialized makeup residual encoder can achieve both high makeup realism and strong facial consistency in synthetic data generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.