Research map / 3D vision / Face and shape
UV-GAN: Adversarial Facial UV Map Completion for Pose-Invariant Face Recognition
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 |
|---|---|---|---|
| Add UV-GAN pose augmentation during training to CASIA-sm | ResNet-27 face recognition model · CFP frontal-profile | verification accuracy (%): 87.74% to 93.09% | Helped |
| Add UV-GAN pose augmentation during training to CASIA-sm | ResNet-27 face recognition model · CFP frontal-frontal | verification accuracy (%): 98.59% to 98.83% | Helped |
| Add generated view templates for both faces during testing (ablation) | CASIA-sm-augUV · CFP frontal-profile | verification accuracy (%): 93.09% to 94.05% | Helped |
| Synthesize frontal faces from profile faces during testing (ablation) | CASIA-sm-augUV · CFP frontal-profile | verification accuracy (%): 93.09% to 93.55% | Helped |
| Synthesize profile faces from frontal faces during testing (ablation) | CASIA-sm-augUV · CFP frontal-profile | verification accuracy (%): 93.09% to 93.72% | Helped |
| Add flipped-face training examples (ablation) | UV completion model using Lgen · UVDB profile faces | PSNR: 25.8 -> 26.3 (3dMD); 25.2 -> 25.7 (MultiPIE); 22.3 -> 22.8 (WildUV) | Helped |
| Add global adversarial loss (ablation) | Flip + Lgen UV completion model · UVDB profile faces | PSNR: 26.3 -> 25.2 (3dMD); 25.7 -> 24.6 (MultiPIE); 22.8 -> 22.0 (WildUV) | Hurt |
| Add local adversarial loss alongside global adversarial loss (ablation) | Flip + Lgen + Ladv_g UV completion model · UVDB profile faces | PSNR: 25.2 -> 25.7 (3dMD); 24.6 -> 25.2 (MultiPIE); 22.0 -> 22.5 (WildUV) | Helped |
| Add the identity-preserving module to reconstruction, flip augmentation, and both adversarial losses (ablation) | UV completion model · UVDB profile faces | SSIM: 0.886 -> 0.898 (3dMD); 0.873 -> 0.886 (MultiPIE); 0.868 -> 0.887 (WildUV) | Helped |
| Use UV-GAN rather than Context Encoder | Profile UV completion · 3dMD | PSNR: 25.1 -> 26.5 | Helped |
| Increase yaw angle and corresponding self-occlusion (ablation) | UV-GAN · UVDB | PSNR: 28.3 -> 26.5 (3dMD); 27.5 -> 25.8 (MPIE); 26.3 -> 22.9 (WildUV), from 0° to ±90° | Hurt |
| Train with UV-GAN pose augmentation rather than 3D shape identity-parameter perturbation augmentation | CASIA-sm face recognition model · CFP frontal-profile | verification accuracy (%): 90.14 -> 93.09 | Helped |
Closest papers by meaning
- GANFIT: Generative Adversarial Network Fitting for High Fidelity 3D Face Reconstruction (CVPR 2019 · similarity 0.89)
- A 3D GAN for Improved Large-Pose Facial Recognition (CVPR 2021 · similarity 0.89)
- Towards High-Fidelity 3D Face Reconstruction From In-the-Wild Images Using Graph Convolutional Networks (CVPR 2020 · similarity 0.88)
- OSTeC: One-Shot Texture Completion (CVPR 2021 · similarity 0.88)
- FFHQ-UV: Normalized Facial UV-Texture Dataset for 3D Face Reconstruction (CVPR 2023 · similarity 0.87)
- FreeUV: Ground-Truth-Free Realistic Facial UV Texture Recovery via Cross-Assembly Inference Strategy (CVPR 2025 · 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.