Research map / Label-efficient and robust learning / Recognition and long-tail
Reconstruction-Based Disentanglement for Pose-Invariant Face Recognition
Xi Peng, Xiang Yu, Kihyuk Sohn, Dimitris N. Metaxas, Manmohan Chandraker
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 Siamese reconstruction regularization (ablation) | MSMT · MultiPIE (P1, avg) | rank-1 accuracy: 0.882 -> 0.905 | Improved |
| add Siamese reconstruction regularization (ablation) | MSMT · 300WLP (avg) | rank-1 accuracy: 0.971 -> 0.980 | Improved |
| replace direct L2 identity distance by reconstruction regularization (ablation) | MSMT+L2 · MultiPIE (P1, avg) | rank-1 accuracy: 0.900 -> 0.905 | Improved |
| add direct L2 distance between identity features of same subject (ablation) | MSMT · MultiPIE (P1, avg) | rank-1 accuracy: 0.882 -> 0.900 | Improved |
| add multi-source multi-task training instead of single-source fine-tuning (ablation) | SS-FT · MultiPIE (P1, avg) | rank-1 accuracy: 0.860 -> 0.882 | Improved |
| fine-tune on target dataset (ablation) | SS · MultiPIE (P1, avg) | rank-1 accuracy: 0.690 -> 0.860 | Improved |
| use N-pair loss instead of cross entropy (ablation) | MSMT · MultiPIE (P1, avg) | rank-1 accuracy: 0.882 -> 0.927 | Improved |
| add Siamese reconstruction on top of N-pair loss (ablation) | MSMT-dagger · MultiPIE (P1, avg) | rank-1 accuracy: 0.927 -> 0.929 | No clear effect |
| add Siamese reconstruction on top of N-pair loss (ablation) | MSMT-dagger · 300WLP (avg) | rank-1 accuracy: 0.977 -> 0.986 | Improved |
| add reconstruction regularization, cross-database train 300WLP test MultiPIE | MSMT · MultiPIE (cross-database, P1 avg) | rank-1 accuracy: 0.788 -> 0.808 | Improved |
| add reconstruction regularization, cross-database train MultiPIE test 300WLP | MSMT · 300WLP (cross-database, avg) | rank-1 accuracy: 0.904 -> 0.910 | Improved |
| add reconstruction regularization with P2 protocol (ablation) | MSMT · MultiPIE (P2, avg) | rank-1 accuracy: 0.939 -> 0.954 | Improved |
Closest papers by meaning
- Towards Pose Invariant Face Recognition in the Wild (CVPR 2018 · similarity 0.89)
- Pose-Invariant Face Recognition via Adaptive Angular Distillation (AAAI 2022 · similarity 0.89)
- Disentangled Representation Learning GAN for Pose-Invariant Face Recognition (CVPR 2017 · similarity 0.88)
- Representation Learning by Rotating Your Faces (TPAMI 2018 · similarity 0.88)
- Reinforced Disentanglement for Face Swapping without Skip Connection (ICCV 2023 · similarity 0.88)
- Pose-Disentangled Contrastive Learning for Self-Supervised Facial Representation (CVPR 2023 · similarity 0.87)
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.
