Research map / Image generation / Editing and style transfer
DCFace: Synthetic Face Generation With Dual Condition Diffusion Model
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 DCFace instead of DigiFace for the 0.5M synthetic training set | IR-SE50 face recognition training · LFW | verification accuracy: 95.4 -> 98.55 | Helped |
| use DCFace instead of DigiFace for the 0.5M synthetic training set | IR-SE50 face recognition training · CFP-FP | verification accuracy: 87.4 -> 85.33 | Hurt |
| use a 5×5 patch-wise style grid rather than a 1×1 grid (ablation) | DCFace · FR validation average | verification accuracy: 79.28 -> 89.04 | Helped |
| increase style grid size from 5×5 to 7×7 (ablation) | DCFace · FR validation average | verification accuracy: 89.04 -> 50.00 | Hurt |
| use the time-step-dependent ID loss instead of L_naive1 (ablation) | DCFace · FR validation average | verification accuracy: 84.75 -> 89.04 | Helped |
| use the time-step-dependent ID loss instead of L_naive2 (ablation) | DCFace · FR validation average | verification accuracy: 50.00 -> 89.04 | Helped |
| use the larger WebFace4M-pretrained ID-loss backbone (ablation) | DCFace · FR validation average | verification accuracy: 89.04 -> 89.89 | Helped |
| balance the ID gender/ethnicity sampling instead of sampling IDs randomly (ablation) | DCFace · FR validation average | verification accuracy: 89.04 -> 88.77 | Hurt |
| match style-image gender/ethnicity to the sampled ID instead of random style sampling (ablation) | DCFace · FR validation average | verification accuracy: 89.04 -> 89.12 | No clear effect |
| oversample X_id during face-recognition training to include more frontal faces (ablation) | DCFace-generated training set · FR validation average | verification accuracy: 89.04 -> 89.56 | Helped |
| use classifier-free guidance instead of the time-dependent ID loss (ablation) | DCFace · FR validation average | verification accuracy: 89.56 -> 84.05 | Hurt |
| combine DCFace and DigiFace synthetic datasets | IR50 face recognition training · FR validation average | verification accuracy: 91.21 -> 93.06 | Helped |
Closest papers by meaning
- HyperFace: Generating Synthetic Face Recognition Datasets by Exploring Face Embedding Hypersphere (ICLR 2025 · similarity 0.91)
- CemiFace: Center-based Semi-hard Synthetic Face Generation for Face Recognition (NeurIPS 2024 · similarity 0.89)
- Data Synthesis with Diverse Styles for Face Recognition via 3DMM-Guided Diffusion (CVPR 2025 · similarity 0.89)
- UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class Diversity in Synthetic Face Recognition (ICLR 2025 · similarity 0.89)
- $\text{ID}^3$: Identity-Preserving-yet-Diversified Diffusion Models for Synthetic Face Recognition (NeurIPS 2024 · similarity 0.89)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · similarity 0.88)
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.