Research map / Image generation / GANs
Teacher-Student Adversarial Depth Hallucination To Improve Face Recognition
Hardik Uppal, Alireza Sepas-Moghaddam, Michael Greenspan, Ali Etemad
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 |
|---|---|---|---|
| add the student component to the teacher (ablation) | TS-GAN (VGG-16) · EURECOM KinectFaceDb | rank-1 accuracy: 85.7% -> 88.3% | Improved |
| add the student component to the teacher (ablation) | TS-GAN (SE-ResNet-50) · EURECOM KinectFaceDb | rank-1 accuracy: 91.9% -> 93.2% | Improved |
| add the student component to the teacher (ablation) | TS-GAN (VGG-16) · IIIT-D | rank-1 accuracy: 95.4% -> 95.4% | No clear effect |
| add the teacher's discriminator to the A2B generator (ablation) | Teacher (VGG-16) · EURECOM KinectFaceDb | rank-1 accuracy: 87.8% -> 85.7% | Hurt |
| add the teacher's discriminator to the A2B generator (ablation) | Teacher (VGG-16) · IIIT-D | rank-1 accuracy: 95.1% -> 95.4% | Improved |
| add the student on top of the A2B generator (ablation) | TS-GAN (ResNet-50) · EURECOM KinectFaceDb | rank-1 accuracy: 90.6% -> 92.2% | Improved |
| add hallucinated depth to RGB | VGG-16 · LFW | rank-1 accuracy: 75.3% -> 78.7% | Improved |
| add hallucinated depth to RGB | SE-ResNet-50 · LFW | rank-1 accuracy: 83.2% -> 85.6% | Improved |
| add hallucinated depth to RGB | VGG-16 · EURECOM KinectFaceDb | rank-1 accuracy: 83.6% -> 88.3% | Improved |
| use feature-level fusion instead of score-level fusion | RGB + hallucinated depth (VGG-16) · LFW | rank-1 accuracy: 76.1% -> 78.7% | Improved |
| use score-level fusion of RGB and hallucinated depth | VGG-16 · IIIT-D | rank-1 accuracy: 94.1% -> 94.1% | No clear effect |
| replace ground-truth depth with hallucinated depth | ResNet-50 feature fusion · IIIT-D | rank-1 accuracy: 96.9% -> 97.1% | No clear effect |
Closest papers by meaning
- A 3D GAN for Improved Large-Pose Facial Recognition (CVPR 2021 · similarity 0.88)
- 3D-Aided Dual-Agent GANs for Unconstrained Face Recognition (TPAMI 2018 · similarity 0.87)
- Teachers Do More Than Teach: Compressing Image-to-Image Models (CVPR 2021 · similarity 0.86)
- Adversarial Learning of Portable Student Networks (AAAI 2018 · similarity 0.86)
- 3DFaceGAN: Adversarial Nets for 3D Face Representation, Generation, and Translation (IJCV 2020 · similarity 0.86)
- GANFIT: Generative Adversarial Network Fitting for High Fidelity 3D Face Reconstruction (CVPR 2019 · 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.
