Research map / Image generation / Editing and style transfer
Face Reconstruction from Facial Templates by Learning Latent Space of a Generator Network
Hatef Otroshi Shahreza, Sébastien Marcel
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 WGAN critic training on the mapping network (ablation) | Ours · MOBIO, attack 1 vs ArcFace, FMR=1e-3 | SAR (%): 0 -> 92.38 | Improved |
| add WGAN critic training on the mapping network (ablation) | Ours · LFW, attack 1 vs ArcFace, FMR=1e-3 | SAR (%): 0.02 -> 86.82 | Improved |
| add pixel loss to ID loss (ablation) | Ours · MOBIO, attack 1 vs ArcFace, FMR=1e-3 | SAR (%): 82.38 -> 92.38 | Improved |
| add pixel loss to ID loss (ablation) | Ours · LFW, attack 1 vs ArcFace, FMR=1e-3 | SAR (%): 80.74 -> 86.82 | Improved |
| use ID loss instead of pixel loss only (ablation) | Ours · MOBIO, attack 1 vs ArcFace, FMR=1e-3 | SAR (%): 0 -> 82.38 | Improved |
| use ArcFace instead of ElasticFace as Floss (ablation) | Ours · MOBIO, attack 3 vs HRNet, FMR=1e-3 | SAR (%): 82.38 -> 86.19 | Improved |
| use ArcFace instead of ElasticFace as Floss (ablation) | Ours · LFW, attack 3 vs HRNet, FMR=1e-3 | SAR (%): 43.71 -> 48.22 | Improved |
| add noise input to the mapping network (ablation) | Ours · MOBIO, attack 1 vs ArcFace, FMR=1e-3 | SAR (%): 74.76 -> 92.38 | Improved |
| add noise input to the mapping network (ablation) | Ours · LFW, attack 1 vs ArcFace, FMR=1e-3 | SAR (%): 77.72 -> 86.82 | Improved |
| map templates to intermediate latent W instead of input latent Z (ablation) | Ours · MOBIO, attack 1 vs ArcFace, FMR=1e-3 | SAR (%): 41.42 -> 92.38 | Improved |
| map templates to intermediate latent W instead of input latent Z (ablation) | Ours · LFW, attack 1 vs ArcFace, FMR=1e-3 | SAR (%): 57.18 -> 86.82 | Improved |
| move from same-system (attack 1) to transfer target Swin | Ours, ArcFace templates · MOBIO, FMR=1e-3 | SAR (%): 92.38 -> 74.29 | Hurt |
Closest papers by meaning
- Comprehensive Vulnerability Evaluation of Face Recognition Systems to Template Inversion Attacks via 3D Face Reconstruction (TPAMI 2023 · similarity 0.95)
- Template Inversion Attack against Face Recognition Systems using 3D Face Reconstruction (ICCV 2023 · similarity 0.94)
- On the Reconstruction of Face Images from Deep Face Templates (TPAMI 2018 · similarity 0.91)
- Recoverable Facial Identity Protection via Adaptive Makeup Transfer Adversarial Attacks (AAAI 2025 · similarity 0.87)
- Non-Adaptive Adversarial Face Generation (NeurIPS 2025 · similarity 0.86)
- Diffusion-driven GAN Inversion for Multi-Modal Face Image Generation (CVPR 2024 · similarity 0.86)
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.
