Research map / Image generation / Editing and style transfer
StyleFace: Towards Identity-Disentangled Face Generation on Megapixels
Yuchen Luo, Junwei Zhu, Keke He, Wenqing Chu, Ying Tai, Chengjie Wang, Junchi Yan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use StyleFace instead of InfoSwap for swapping | InfoSwap · Identity swapping (FF++ / CelebAMaskHQ) | FID (CelebAMaskHQ): 3.39 -> 2.04 | Improved |
| use StyleFace instead of InfoSwap for swapping | InfoSwap · Identity swapping (FF++ / CelebAMaskHQ) | ID retrieval: 98.70 -> 96.34 | Hurt |
| use StyleFace instead of InfoSwap for swapping | InfoSwap · Identity swapping (FF++ / CelebAMaskHQ) | pose error: 3.11 -> 2.52 | Improved |
| use StyleFace instead of InfoSwap for swapping | InfoSwap · Identity swapping (FF++ / CelebAMaskHQ) | inference time (ms per 1024^2 image): 329 -> 86 | Improved |
| use StyleFace instead of InfoSwap for swapping | InfoSwap · Identity swapping (FF++ / CelebAMaskHQ) | user study preference (%): 18.7% -> 74.6% | Improved |
| use StyleFace instead of CIAGAN for de-identification | CIAGAN · De-identification (LFW, TAR, VGGFace2 FaceNet) | true acceptance rate: 0.034 -> 0.013 | Improved |
| use StyleFace instead of CIAGAN for de-identification | CIAGAN · De-identification (LFW, TAR, CASIA FaceNet) | true acceptance rate: 0.019 -> 0.012 | Improved |
| expand latent space Z to Z+ (hierarchical codes) (ablation) | StyleFace · Identity swapping (qualitative) | identity similarity | Improved |
| add contrastive constraint on W+ codes (ablation) | StyleFace · Virtual identity sampling (t-SNE) | feasibility of sampled identities | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CIAGAN — CIAGAN: Conditional Identity Anonymization Generative Adversarial Networks (CVPR 2020) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- StyleIPSB: Identity-Preserving Semantic Basis of StyleGAN for High Fidelity Face Swapping (CVPR 2023 · similarity 0.90)
- StyleSwap: Style-Based Generator Empowers Robust Face Swapping (ECCV 2022 · similarity 0.90)
- BlendFace: Re-designing Identity Encoders for Face-Swapping (ICCV 2023 · similarity 0.89)
- When StyleGAN Meets Stable Diffusion: a W+ Adapter for Personalized Image Generation (CVPR 2024 · similarity 0.89)
- Enhancing Identity-Deformation Disentanglement in StyleGAN for One-Shot Face Video Re-Enactment (AAAI 2025 · similarity 0.88)
- Data Synthesis with Diverse Styles for Face Recognition via 3DMM-Guided Diffusion (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
