Let the Gallery Learn Itself: Training-Free Iterative Side-View Face Recognition
Organizations: Shanghai Jiao Tong University
Abstract
What if a face recognition system could learn from every encounter without a single gradient update? We present Gallery Self-Evolution (GSE), a training-free framework where the gallery itself adapts at test time. The core mechanism---dedup-FIFO curation---archives novel appearance samples after successful recognition while permanently locking the enrollment template, bounded at 30 embeddings per identity. On CFP-FP and IJB-C benchmarks, GSE closes the side-view gap by +23 points at 45° yaw and boosts TAR@FAR=0.001 by +9.4 points, all at CPU-level cost. No retraining. No GPU. Just a gallery that grows smarter every time it meets you.
Figures & tables
| Yaw | Baseline (1 samp.) | Ours (4 samp.) | |
|---|---|---|---|
| 0 ∘ | 0.912 | 0.918 | |
| 15 ∘ | 0.856 | 0.932 | |
| 30 ∘ | 0.784 | 0.951 | |
| 45 ∘ | 0.712 | 0.945 | |
| 60 ∘ | 0.658 | 0.928 |
| Baseline | Ours | ||
|---|---|---|---|
| 0.00 | 0.784 | 0.832 | |
| 0.05 | 0.712 | 0.895 | |
| 0.10 | 0.628 | 0.912 | |
| 0.15 | 0.542 | 0.856 | |
| 0.20 | 0.476 | 0.708 |
| Method | Retrain | GPU | R1@45 ∘ | Edge |
|---|---|---|---|---|
| GAN Frontalization [ 8 ] | Yes | Yes | 0.78 | No |
| Pose-Invariant [ 3 ] | Yes | Yes | 0.82 | No |
| 3D Reconstruction [ 7 ] | No | Yes | 0.85 | No |
| Continual Learning [ 10 ] | Yes | Yes | 0.88 | No |
| GSE (Ours) | No | No | 0.945 | Yes |
Face Recognition
Face recognition matches or identifies people from facial images, videos, and biometric representations across changing conditions.
See all 116 papers