Research map / Label-efficient and robust learning / Recognition and long-tail
Semi-Siamese Training for Shallow Face Learning
Hang Du, Hailin Shi, Yuchi Liu, Jun Wang, Zhen Lei, Dan Zeng, Tao Mei
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 prototype constraint (ablation) | softmax on shallow data · BLUFR (shallow data) | BLUFR verification rate (%): 27.05 -> 24.87 | Hurt |
| add the prototype constraint (ablation) | Arc-softmax on shallow data · BLUFR (shallow data) | BLUFR verification rate (%): 40.34 -> 40.08 | No clear effect |
| add the network constraint (ablation) | softmax on shallow data · BLUFR (shallow data) | BLUFR verification rate (%): 27.05 -> 30.65 | Improved |
| add the gallery queue (ablation) | softmax on shallow data · BLUFR (shallow data) | BLUFR verification rate (%): 27.05 -> 52.05 | Improved |
| add the gallery queue (ablation) | AM-softmax on shallow data · BLUFR (shallow data) | BLUFR verification rate (%): 33.23 -> 62.07 | Improved |
| combine network constraint and gallery queue (ablation) | softmax on shallow data · BLUFR (shallow data) | BLUFR verification rate (%): 52.05 -> 69.93 | Improved |
| use moving-average Semi-Siamese networks with gallery queue (full SST) (ablation) | softmax on shallow data · BLUFR (shallow data) | BLUFR verification rate (%): 69.93 -> 77.58 | Improved |
| full SST over plain training (ablation) | AM-softmax on shallow data · BLUFR (shallow data) | BLUFR verification rate (%): 33.23 -> 84.95 | Improved |
| full SST over plain training (ablation) | A-softmax on shallow data · BLUFR (shallow data) | BLUFR verification rate (%): 37.16 -> 80.65 | Improved |
| train on shallow instead of deep data (two images per ID) (ablation) | softmax · LFW (MobileFaceNet) | accuracy (%): 99.10 -> 92.64 | Hurt |
| apply SST on deep data | softmax · BLUFR (deep data) | BLUFR verification rate (%): 93.15 -> 94.56 | Improved |
| apply SST on deep data | Arc-softmax · BLUFR (deep data) | BLUFR verification rate (%): 95.29 -> 94.76 | No clear effect |
Closest papers by meaning
- Pre-training Strategies and Datasets for Facial Representation Learning (ECCV 2022 · similarity 0.85)
- Wasserstein CNN: Learning Invariant Features for NIR-VIS Face Recognition (TPAMI 2018 · similarity 0.84)
- NPT-Loss: Demystifying Face Recognition Losses With Nearest Proxies Triplet (TPAMI 2022 · similarity 0.84)
- Pairwise Relational Networks for Face Recognition (ECCV 2018 · similarity 0.84)
- Range Loss for Deep Face Recognition With Long-Tailed Training Data (ICCV 2017 · similarity 0.84)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 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.
