Research map / Label-efficient and robust learning / Recognition and long-tail
Face Recognition with Contrastive Convolution
Chunrui Han, Shiguang Shan, Meina Kan, Shuzhe Wu, Xilin Chen
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 contrastive convolution instead of conventional convolution (pairwise loss only) | 4-layer CNN · IJB-A TAR@FAR=0.01 | TAR (%): 22.43 -> 52.51 | Improved |
| add contrastive convolution instead of conventional convolution (pairwise loss only) | 4-layer CNN · LFW | mAcc: 91.80 -> 95.20 | Improved |
| add contrastive convolution instead of conventional convolution (with softmax loss) | 4-layer CNN · LFW | mAcc: 97.50 -> 98.20 | Improved |
| add contrastive convolution instead of conventional convolution (with softmax loss) | 4-layer CNN · IJB-A TAR@FAR=0.001 | TAR (%): 52.72 -> 58.04 | Improved |
| add softmax kernel classification loss to pairwise loss (ablation) | Contrastive CNN · IJB-A TAR@FAR=0.001 | TAR (%): 31.37 -> 58.04 | Improved |
| add softmax loss to pairwise loss (ablation) | L-VanillaCNN · LFW | mAcc: 91.80 -> 97.50 | Improved |
| increase sub-generators from 1 to 3 (ablation) | Contrastive CNN · IJB-A TAR@FAR=0.001 | TAR (%): 37.32 -> 58.04 | Improved |
| increase sub-generators from 1 to 3 (ablation) | Contrastive CNN · LFW | mAcc: 97.83 -> 98.20 | Improved |
| increase sub-generators from 2 to 3 (ablation) | Contrastive CNN · IJB-A TAR@FAR=0.01 | TAR (%): 75.08 -> 74.55 | No clear effect |
| deepen base CNN from 4 to 16 layers | Contrastive CNN · LFW | mAcc: 98.20 -> 99.12 | Improved |
| deepen base CNN from 4 to 16 layers | Contrastive CNN · IJB-A TAR@FAR=0.1 | TAR (%): 90.24 -> 95.31 | Improved |
| deepen base CNN from 4 to 16 layers | Contrastive CNN · IJB-A TAR@FAR=0.001 | TAR (%): 58.04 -> 63.91 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SphereFace — SphereFace: Deep Hypersphere Embedding for Face Recognition (CVPR 2017) · better on 0 of 1 shared result
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
- Hybrid Deep Learning for Face Verification (TPAMI 2016 · similarity 0.86)
- Robust RGB-D Face Recognition Using Attribute-Aware Loss (TPAMI 2019 · similarity 0.84)
- Learning Invariant Deep Representation for NIR-VIS Face Recognition (AAAI 2017 · similarity 0.84)
- Detecting Faces Using Inside Cascaded Contextual CNN (ICCV 2017 · similarity 0.84)
- Towards Interpretable Face Recognition (ICCV 2019 · similarity 0.84)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (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.
