Research map / Image generation / Editing and style transfer
TextFace: Compositional Text-Guided Identity Preserving Face Synthesis for Face Recognition
Junzhe Yang, Mingjie He, Shiguang Shan
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 |
|---|---|---|---|
| add attribute-library captions and training-stage regularization (ablation) | TextFace base · Ablation (0.5M synthetic) | LPIPS: 0.397 -> 0.439 | Improved |
| add attribute-library captions and training-stage regularization (ablation) | TextFace base · Ablation (0.5M synthetic) | FR avg accuracy: 93.75 -> 93.84 | Improved |
| add attribute-library captions and training-stage regularization (ablation) | TextFace base · Ablation (0.5M synthetic) | valid face rate: 95.19 -> 93.31 | Hurt |
| add dual-conditioned CFG (ablation) | TextFace + T-enhance · Ablation (0.5M synthetic) | FR avg accuracy: 93.84 -> 93.24 | Hurt |
| add dual-conditioned CFG (ablation) | TextFace + T-enhance · Ablation (0.5M synthetic) | valid face rate: 93.31 -> 94.76 | Improved |
| add dual-conditioned CFG (ablation) | TextFace + T-enhance · Ablation (0.5M synthetic) | intra-identity cos sim: 0.364 -> 0.532 | Hurt |
| add identity blending (ablation) | TextFace + T-enhance + CFG · Ablation (0.5M synthetic) | FR avg accuracy: 93.24 -> 93.91 | Improved |
| add identity blending (ablation) | TextFace + T-enhance + CFG · Ablation (0.5M synthetic) | intra-identity LPIPS: 0.445 -> 0.452 | Improved |
| add identity blending (ablation) | TextFace + T-enhance + CFG · Ablation (0.5M synthetic) | intra-identity cos sim: 0.532 -> 0.420 | Improved |
| add identity blending (ablation) | TextFace + T-enhance + CFG · Ablation (0.5M synthetic) | valid face rate: 94.76 -> 94.18 | Hurt |
| add full TextFace (compared to VIGFace) | TextFace · CFP-FP (0.5M synthetic) | verification accuracy: 95.09 -> 93.80 | Hurt |
| scale synthetic data from 0.5M to 6.0M | TextFace · FR avg accuracy | verification accuracy: 93.91 -> 95.50 | 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.
- UIFace — UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class Diversity in Synthetic Face Recognition (ICLR 2025) · 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
- $\text{ID}^3$: Identity-Preserving-yet-Diversified Diffusion Models for Synthetic Face Recognition (NeurIPS 2024 · similarity 0.92)
- Data Synthesis with Diverse Styles for Face Recognition via 3DMM-Guided Diffusion (CVPR 2025 · similarity 0.90)
- UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class Diversity in Synthetic Face Recognition (ICLR 2025 · similarity 0.90)
- When StyleGAN Meets Stable Diffusion: a W+ Adapter for Personalized Image Generation (CVPR 2024 · similarity 0.89)
- FaithfulFaces: Pose-Faithful Facial Identity Preservation for Text-to-Video Generation (NeurIPS 2026 · similarity 0.87)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 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.
