Research map / Image generation / Editing and style transfer
On the Reconstruction of Face Images from Deep Face Templates
Guangcan Mai, Kai Cao, Pong Chi Yuen, Anil K. Jain
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 NbNet-B blocks instead of plain de-convolution (ablation) | VGG-trained, perceptual loss · LFW type-II | TAR @ FAR 0.1% (%): 45.63 -> 58.05 | Improved |
| use NbNet-B blocks instead of plain de-convolution (ablation) | VGG-trained, pixel loss · LFW type-I | TAR @ FAR 0.1% (%): 70.22 -> 89.52 | Improved |
| use NbNet-B blocks instead of plain de-convolution (ablation) | MPIE-trained, pixel loss · LFW type-I | TAR @ FAR 0.1% (%): 37.11 -> 67.86 | Improved |
| use NbNet-A blocks instead of plain de-convolution (ablation) | MPIE-trained, perceptual loss · LFW type-I | TAR @ FAR 0.1% (%): 85.34 -> 80.33 | Hurt |
| use NbNet-B blocks instead of plain de-convolution (ablation) | MPIE-trained, perceptual loss · FRGC v2.0 type-I | TAR @ FAR 0.1% (%): 55.22 -> 73.76 | Improved |
| add perceptual loss instead of pixel difference loss (ablation) | NbNet-B trained on VGG-Face · LFW type-I | TAR @ FAR 0.1% (%): 89.52 -> 94.37 | Improved |
| add perceptual loss instead of pixel difference loss (ablation) | NbNet-B trained on Multi-PIE · LFW type-II | TAR @ FAR 0.1% (%): 24.00 -> 37.30 | Improved |
| add DCGAN-generated image augmentation (ablation) | NbNet-B trained on VGG-Face · LFW type-I | TAR @ FAR 0.1% (%): 72.53 -> 89.52 | Improved |
| add DCGAN-generated image augmentation (ablation) | NbNet-B trained on Multi-PIE · LFW type-I | TAR @ FAR 0.1% (%): 34.87 -> 67.86 | Improved |
| add DCGAN-generated image augmentation (ablation) | NbNet-B trained on Multi-PIE · color FERET type-I rank-1 | rank-1 (%): 63.88 -> 85.61 | Improved |
| train on mixed raw datasets instead of raw VGG-Face (ablation) | NbNet-B pixel loss · LFW type-I | TAR @ FAR 0.1% (%): 72.53 -> 71.62 | Hurt |
| use NbNet with DCGAN augmentation instead of RBF regression | NbNet-B perceptual loss · LFW type-I | TAR @ FAR 0.1% (%): 19.76 -> 94.37 | Improved |
Closest papers by meaning
- Face Reconstruction from Facial Templates by Learning Latent Space of a Generator Network (NeurIPS 2023 · similarity 0.91)
- Template Inversion Attack against Face Recognition Systems using 3D Face Reconstruction (ICCV 2023 · similarity 0.90)
- Comprehensive Vulnerability Evaluation of Face Recognition Systems to Template Inversion Attacks via 3D Face Reconstruction (TPAMI 2023 · similarity 0.89)
- Face Reconstruction Transfer Attack as Out-of-Distribution Generalization (ECCV 2024 · similarity 0.88)
- Recoverable Facial Identity Protection via Adaptive Makeup Transfer Adversarial Attacks (AAAI 2025 · similarity 0.84)
- CryptoFace: End-to-End Encrypted Face Recognition (CVPR 2025 · similarity 0.84)
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.
