Research map / Image generation / Editing and style transfer
Representation Learning by Rotating Your Faces
Luan Tran, Xi Peng Yin, Xiaoming Liu
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 pose classification component D^p to the discriminator (ablation) | single-image DR-GAN · IJB-A | Rank-1 identification: 87.5 -> 89.0 | Improved |
| add real/fake discriminator component D^r (ablation) | single-image DR-GAN · IJB-A | Rank-1 identification: 88.7 -> 89.0 | Improved |
| train with 6 images per subject instead of 1 (ablation) | DR-GAN · CFP frontal-profile | accuracy: 90.82 -> 93.41 | Improved |
| train with 6 instead of 4 images per subject (ablation) | DR-GAN · CFP frontal-frontal | accuracy: 97.86 -> 97.84 | No clear effect |
| repeatedly replace D^d with the improved G_enc (model switch) (ablation) | DR-GAN · Multi-PIE | identification rate: 79.7 -> 90.4 | Improved |
| fuse template features with omega-fusion instead of feature mean (ablation) | multi-image DR-GAN · IJB-A | Rank-1 identification: 89.6 -> 91.0 | Improved |
| remove images with coefficient below 0.5 before fusion (ablation) | multi-image DR-GAN · IJB-A | Rank-1 identification: 91.0 -> 86.5 | Hurt |
| remove images with coefficient below 0.1 before fusion (ablation) | multi-image DR-GAN · IJB-A | Rank-1 identification: 91.0 -> 91.3 | No clear effect |
| use DR-GAN framework instead of plain softmax loss | G_enc · IJB-A | Rank-1 identification: 87.8 -> 89.0 | Improved |
| replace softmax of D_d with additive-margin softmax | DR-GAN · IJB-A | Rank-1 identification: 90.2 -> 91.0 | Improved |
| separate D^r (split) from earlier DR-GAN (ablation) | DR-GAN · IJB-A | Rank-1 identification: 85.5 -> 91.0 | Improved |
| add representation interpolation to split model (ablation) | DR-GAN split · IJB-A | TAR@FAR=0.001: 72.6 -> 75.1 | Improved |
Closest papers by meaning
- Disentangled Representation Learning GAN for Pose-Invariant Face Recognition (CVPR 2017 · similarity 0.97)
- 3D-Aided Dual-Agent GANs for Unconstrained Face Recognition (TPAMI 2018 · similarity 0.91)
- A 3D GAN for Improved Large-Pose Facial Recognition (CVPR 2021 · similarity 0.90)
- Dual-Agent GANs for Photorealistic and Identity Preserving Profile Face Synthesis (NeurIPS 2017 · similarity 0.89)
- Towards Pose Invariant Face Recognition in the Wild (CVPR 2018 · similarity 0.89)
- Pose-Guided Photorealistic Face Rotation (CVPR 2018 · similarity 0.88)
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.
