Reproduce “UniPortrait: A Unified Framework for Identity-Preserving Single- and Multi-Human Image Personalization” 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
Junjie He, Yifeng Geng, Liefeng Bo
UniPortrait introduces a unified framework for identity-preserving single- and multi-human image personalization, utilizing novel ID embedding and ID routing modules to achieve high-fidelity identity, extensive editability, and robust multi-ID generation from free-form text. The system outperforms state-of-the-art methods in both quantitative and qualitative assessments.
Key insights
- Utilizing local features from the penultimate layer of a face recognition backbone better preserves spatial information, leading to higher identity fidelity than global features.
- A decoupling strategy with strong dropping regularization on face structure features prevents overfitting to non-identity attributes, enhancing editability while maintaining detailed identity preservation.
- Explicitly routing identity embeddings to specific spatial locations via a learnable network and regularization loss effectively resolves identity blending issues in multi-human generations without requiring layout masks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.