Research map / Label-efficient and robust learning / Recognition and long-tail
NPT-Loss: Demystifying Face Recognition Losses With Nearest Proxies Triplet
Syed Safwan Khalid, Muhammad Awais, Zhen-Hua Feng, Chi-Ho Chan, Ammarah Farooq, Ali Akbari, Josef Kittler
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 |
|---|---|---|---|
| replace Curricular-Face with NPT loss | ResNet-50 trained on CASIA · MegaFace 1M distractors | rank-1 accuracy (%): 91.52 -> 92.52 | Improved |
| replace ArcFace with NPT loss | ResNet-50 trained on CASIA · MegaFace 1M distractors | rank-1 accuracy (%): 89.01 -> 92.52 | Improved |
| replace naive proxy-triplet with NPT loss (implicit hard-negative mining) | ResNet-50 trained on CASIA · MegaFace 1M distractors | rank-1 accuracy (%): 35.85 -> 92.52 | Improved |
| replace ArcFace with NPT loss | ResNet-50 trained on CASIA · CFP-FP | accuracy (%): 95.3 -> 96.44 | Improved |
| replace ArcFace with NPT loss | ResNet-50 trained on CASIA · LFW | accuracy (%): 99.3 -> 99.49 | Improved |
| replace Curricular-Face with NPT loss | ResNet-50 trained on MS1Mv2 · MegaFace | rank-1 accuracy (%): 96.98 -> 97.67 | Improved |
| replace ArcFace with NPT loss | ResNet-50 trained on MS1Mv2 · MegaFace | rank-1 accuracy (%): 97.58 -> 97.67 | Improved |
| replace ArcFace with NPT loss | ResNet-50 trained on MS1Mv2 · IJB-C | TAR @ FAR=1e-4 (%): 94.79 -> 94.97 | Improved |
| replace ArcFace with NPT loss | ResNet-50 trained on MS1Mv2 · IJB-B | TAR @ FAR=1e-4 (%): 93.15 -> 93.39 | Improved |
| replace Huang et al. with NPT loss on low-resolution training | ResNet-50 · SCFace | average accuracy (%): 94.46 -> 96.61 | Improved |
| vary margin-like hyper-parameter Delta above r^2/2 (ablation) | NPT loss, CASIA · MegaFace (10K distractors) | rank-1 accuracy | No clear effect |
| replace Curricular-Face with NPT loss at very low FAR | ResNet-50 trained on CASIA · MegaFace ROC | TAR | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Curricular-Face — CurricularFace: Adaptive Curriculum Learning Loss for Deep Face Recognition (CVPR 2020) · better on 8 of 8 shared results
- ArcFace — ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CosFace: Large Margin Cosine Loss for Deep Face Recognition (CVPR 2018 · similarity 0.90)
- Mis-Classified Vector Guided Softmax Loss for Face Recognition (AAAI 2020 · similarity 0.90)
- UniTSFace: Unified Threshold Integrated Sample-to-Sample Loss for Face Recognition (NeurIPS 2023 · similarity 0.88)
- Robust RGB-D Face Recognition Using Attribute-Aware Loss (TPAMI 2019 · similarity 0.88)
- Cross-Domain Similarity Learning for Face Recognition in Unseen Domains (CVPR 2021 · similarity 0.88)
- Masked Face Recognition with Generative-to-Discriminative Representations (ICML 2024 · 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.
