Research map / Label-efficient and robust learning / Recognition and long-tail
Large-Scale Bisample Learning on ID Versus Spot Face Recognition
Xiangyu Zhu, Hao Liu, Zhen Lei, Hailin Shi, Fan Yang, Dong Yi, Guo-Jun Qi, Stan Ziqing Li
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 verification transfer stage on IvS data after wild pre-training (ablation) | CVC · CASIA-IvS-Test | VR@FAR=1e-5 (%): 51.90 -> 83.23 | Improved |
| add final large-scale DP-softmax classification stage (ablation) | CV# · CASIA-IvS-Test | VR@FAR=1e-5 (%): 83.23 -> 91.92 | Improved |
| add verification stage between the two classification stages (ablation) | C#C · CASIA-IvS-Test | VR@FAR=1e-5 (%): 72.35 -> 91.92 | Improved |
| use sequential verification then classification instead of joint (ablation) | C(VC) · CASIA-IvS-Test | VR@FAR=1e-5 (%): 82.80 -> 91.92 | Improved |
| add wild pre-training before large-scale classification (ablation) | ##C · CASIA-IvS-Test | VR@FAR=1e-5 (%): not converge -> 72.35 | Improved |
| use dominant prototypes instead of random prototypes (100k) (ablation) | LBL · CASIA-IvS-Test | VR@FAR=1e-5 (%): 90.34 -> 91.92 | Improved |
| increase random prototypes from 20k to 100k (ablation) | RP-softmax · CASIA-IvS-Test | VR@FAR=1e-5 (%): 87.84 -> 90.34 | Improved |
| increase dominant queue size from 10 to 100 (ablation) | DP-softmax · CASIA-IvS-Test | VR@FAR=1e-5 (%): 85.12 -> 91.92 | Improved |
| increase dominant queue size from 100 to 300 (ablation) | DP-softmax · CASIA-IvS-Test | VR@FAR=1e-5 (%): 91.92 -> 92.01 | No clear effect |
| add dominant queue updating (ablation) | DP-softmax · CASIA-IvS-Test | VR@FAR=1e-5 (%): 90.77 -> 91.92 | Improved |
| use A-softmax instead of softmax (ablation) | DP-softmax · CASIA-IvS-Test | VR@FAR=1e-5 (%): 89.55 -> 91.92 | Improved |
| use ID-prototype instead of avg-prototype with softmax (ablation) | random prototype softmax · CASIA-IvS-Test | VR@FAR=1e-5 (%): 87.13 -> 89.55 | Improved |
Closest papers by meaning
- An Efficient Training Approach for Very Large Scale Face Recognition (CVPR 2022 · similarity 0.84)
- Killing Two Birds With One Stone: Efficient and Robust Training of Face Recognition CNNs by Partial FC (CVPR 2022 · similarity 0.83)
- SphereFace2: Binary Classification is All You Need for Deep Face Recognition (ICLR 2022 · similarity 0.83)
- CosFace: Large Margin Cosine Loss for Deep Face Recognition (CVPR 2018 · similarity 0.83)
- Virtual Fully-Connected Layer: Training a Large-Scale Face Recognition Dataset With Limited Computational Resources (CVPR 2021 · similarity 0.82)
- UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class Diversity in Synthetic Face Recognition (ICLR 2025 · similarity 0.81)
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.
