Research map / 3D vision / Face and shape
A 3D GAN for Improved Large-Pose Facial Recognition
Check 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 |
|---|---|---|---|
| Augment NetScrape with 3D GAN images using the full pose and SH-lighting setup | ResNet-34 trained on NetScrape · CFP-FP | verification accuracy: 93.59% -> 95.29% | Helped |
| Augment NetScrape with 3D GAN images using the full pose and SH-lighting setup | ResNet-34 trained on NetScrape · CPLFW | verification accuracy: 84.56% -> 86.25% | Helped |
| Add 20k FFHQ-generated synthetic identities | ResNet-100 trained on MS1M-V3 · CPLFW | verification accuracy: 93.36% -> 93.53% | Helped |
| Add spherical harmonic lighting augmentation to synthetic images (ablation) | 3D GAN data augmentation for ResNet-34 · CFP-FP | verification accuracy: 94.46% -> 94.89% | Helped |
| Add spherical harmonic lighting augmentation to synthetic images (ablation) | 3D GAN data augmentation for ResNet-34 · CPLFW | verification accuracy: 85.55% -> 86.25% | Helped |
| Generate broad pose variation instead of a narrow pose distribution (ablation) | 3D GAN data augmentation for ResNet-34 · CFP-FP | verification accuracy: 93.76% -> 94.89% | Helped |
| Generate broad pose variation instead of a narrow pose distribution (ablation) | 3D GAN data augmentation for ResNet-34 · CPLFW | verification accuracy: 84.93% -> 86.25% | Helped |
| Increase synthetic identities from 10k to 20k (ablation) | ResNet-34 trained on NetScrape · CFP-FP | verification accuracy: 94.89% -> 95.29% | Helped |
| Increase synthetic identities from 10k to 20k (ablation) | ResNet-34 trained on NetScrape · CPLFW | verification accuracy: 86.25% -> 85.96% | Hurt |
| Increase synthetic identities from 20k to 30k (ablation) | ResNet-34 trained on NetScrape · CFP-FP | verification accuracy: 95.29% -> 94.63% | Hurt |
| Increase synthetic identities from 20k to 30k (ablation) | ResNet-34 trained on NetScrape · CPLFW | verification accuracy: 85.96% -> 85.91% | Hurt |
| Add 20k FFHQ-generated synthetic identities | ResNet-50 trained on CASIA · CFP-FP | verification accuracy: 95.50% -> 96.74% | Helped |
Closest papers by meaning
- GANFIT: Generative Adversarial Network Fitting for High Fidelity 3D Face Reconstruction (CVPR 2019 · similarity 0.91)
- Disentangled Representation Learning GAN for Pose-Invariant Face Recognition (CVPR 2017 · similarity 0.90)
- FaceID-GAN: Learning a Symmetry Three-Player GAN for Identity-Preserving Face Synthesis (CVPR 2018 · similarity 0.89)
- Dual-Agent GANs for Photorealistic and Identity Preserving Profile Face Synthesis (NeurIPS 2017 · similarity 0.89)
- UV-GAN: Adversarial Facial UV Map Completion for Pose-Invariant Face Recognition (CVPR 2018 · similarity 0.89)
- Diffusion-driven GAN Inversion for Multi-Modal Face Image Generation (CVPR 2024 · 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.