Research map / Label-efficient and robust learning / Recognition and long-tail
A realistic approach to generate masked faces applied on two novel masked face recognition data sets
Tudor-Alexandru Mare, Georgian Duta, Iuliana Georgescu, Adrian Sandru, Bogdan Alexe, Marius Popescu, Radu Tudor Ionescu
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 |
|---|---|---|---|
| fine-tune on CelebA with synthetic masks instead of unmasked images | FaceNet · CelebA-test+masks | verification accuracy: 90.96 -> 93.58 | Improved |
| fine-tune on CelebA with synthetic masks instead of unmasked images | VGG-face · CelebA-test+masks | verification accuracy: 84.56 -> 91.51 | Improved |
| fine-tune on CelebA with synthetic masks instead of unmasked images | ArcFace · CelebA-test+masks | verification accuracy: 91.78 -> 95.43 | Improved |
| fine-tune on CASIA-WebFace with synthetic masks instead of unmasked images | FaceNet · CASIA-WebFace-test+masks | verification accuracy: 84.21 -> 88.06 | Improved |
| fine-tune on CASIA-WebFace with synthetic masks instead of unmasked images | VGG-face · CASIA-WebFace-test+masks | verification accuracy: 79.65 -> 86.85 | Improved |
| fine-tune on CASIA-WebFace with synthetic masks instead of unmasked images | ArcFace · CASIA-WebFace-test+masks | verification accuracy: 87.95 -> 91.47 | Improved |
| train on our masks instead of unmasked CelebA | FaceNet · MFR2 | verification accuracy: 92.20 -> 96.22 | Improved |
| train on our masks instead of unmasked CelebA | VGG-face · MFR2 | verification accuracy: 84.29 -> 91.26 | Improved |
| replace MaskTheFace masks with our SparkAR masks in training (ablation) | VGG-face · MFR2 | verification accuracy: 88.35 -> 91.26 | Improved |
| replace MaskTheFace masks with our SparkAR masks in training (ablation) | FaceNet · MFR2 | verification accuracy: 95.16 -> 96.22 | Improved |
| replace mask texture with black pixels inside mask (ablation) | FaceNet · MFR2 | verification accuracy: 96.22 -> 93.04 | Hurt |
| replace mask texture with black pixels inside mask (ablation) | VGG-face · MFR2 | verification accuracy: 91.26 -> 88.31 | Hurt |
Closest papers by meaning
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · similarity 0.85)
- HyperFace: Generating Synthetic Face Recognition Datasets by Exploring Face Embedding Hypersphere (ICLR 2025 · similarity 0.84)
- CemiFace: Center-based Semi-hard Synthetic Face Generation for Face Recognition (NeurIPS 2024 · similarity 0.82)
- UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class Diversity in Synthetic Face Recognition (ICLR 2025 · similarity 0.82)
- Masked Face Recognition with Generative-to-Discriminative Representations (ICML 2024 · similarity 0.82)
- $\text{ID}^3$: Identity-Preserving-yet-Diversified Diffusion Models for Synthetic Face Recognition (NeurIPS 2024 · similarity 0.82)
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.
