Research map / Label-efficient and robust learning / Recognition and long-tail
R3 Adversarial Network for Cross Model Face 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 |
|---|---|---|---|
| Replace naive FC feature mapping with R3AN | MobileNetV2(T=6) → ResNet-101 feature transformation · MegaFace, 1M distractors | Top1 Accuracy: 83.92 -> 95.97 | Helped |
| Add the regression path to the reconstruction and representation paths (ablation) | R3AN · MegaFace, MobileNetV2(T=6) → ResNet-101 | Top1 Accuracy: 85.65 -> 95.97 | Helped |
| Add naive L2 reconstruction to the regression path (ablation) | R3AN · MegaFace, MobileNetV2(T=6) → ResNet-101 | Top1 Accuracy: 94.05 -> 94.80 | Helped |
| Add adversarial reconstruction to the regression path (ablation) | R3AN · MegaFace, MobileNetV2(T=6) → ResNet-101 | Top1 Accuracy: 94.05 -> 94.21 | Helped |
| Add the representation path to adversarial reconstruction and regression (ablation) | R3AN · MegaFace, MobileNetV2(T=6) → ResNet-101 | Top1 Accuracy: 94.21 -> 94.93 | Helped |
| Add adversarial reconstruction to the L2 reconstruction and representation paths (ablation) | R3AN · MegaFace, MobileNetV2(T=6) → ResNet-101 | Top1 Accuracy: 85.65 -> 95.97 | Helped |
| Omit the reconstruction path (ablation) | R3AN · MegaFace, MobileNetV2(T=6) → ResNet-101 | Top1 Accuracy: 95.97 -> NAN | Hurt |
| Replace naive FC mapping with R3AN | MobileNetV2(T=6) → MobileNetV2(T=16) feature transformation · MegaFace, 1M distractors | Top-1 face identification accuracy: 83.54 -> 94.18 | Helped |
| Replace naive FC gallery transformation with R3AN | PolyNetE → MobileNetV2(T=6) · MegaFace, 1M distractors | Top-1 face identification accuracy: 80.79 -> 94.16 | Helped |
| Train R3AN on IMDb-Face rather than MS-Celeb-1M and VGG2-Face | Cross-model feature transformation · MegaFace, MobileNetV2(T=6) → ResNet-101 | Top-1 face identification accuracy: 95.97 -> 94.84 | Hurt |
| Transform the gallery with R3AN instead of naive FC | ResNet-50 → PolyNetE · MegaFace, 1M distractors | Top-1 face identification accuracy: 88.67 -> 98.40 | Helped |
| Transform the gallery with R3AN rather than use the target model's original features | PolyNetE → MobileNetV2(T=6) · MegaFace, 1M distractors | Top-1 face identification accuracy: 92.84 -> 94.16 | Helped |
Closest papers by meaning
- r-BTN: Cross-Domain Face Composite and Synthesis From Limited Facial Patches (AAAI 2018 · similarity 0.86)
- Adversarial Pose Regression Network for Pose-Invariant Face Recognitions (AAAI 2021 · similarity 0.86)
- Regressing Robust and Discriminative 3D Morphable Models With a Very Deep Neural Network (CVPR 2017 · similarity 0.85)
- Adversarial Discriminative Heterogeneous Face Recognition (AAAI 2018 · similarity 0.85)
- Improving the Transferability of Adversarial Attacks on Face Recognition with Diverse Parameters Augmentation (CVPR 2025 · similarity 0.84)
- OSDFace: One-Step Diffusion Model for Face Restoration (CVPR 2025 · similarity 0.83)
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.