The DOI matches its Crossref record; 49 of the 50 references it cites were found in a registry or an index.
Feature Transfer Learning for Face Recognition With Under-Represented Data
Abstract
Despite the large volume of face recognition datasets, there is a significant portion of subjects, of which the samples are insufficient and thus under-represented. Ignoring such significant portion results in insufficient training data. Training with under-represented data leads to biased classifiers in conventionally-trained deep networks. In this paper, we propose a center-based feature transfer framework to augment the feature space of under-represented subjects from the regular subjects that have sufficiently diverse samples. A Gaussian prior of the variance is assumed across all subjects and the variance from regular ones are transferred to the under-represented ones. This encourages the under-represented distribution to be closer to the regular distribution. Further, an alternating training regimen is proposed to simultaneously achieve less biased classifiers and a more discriminative feature representation. We conduct ablative study to mimic the under-represented datasets by varying the portion of under-represented classes on the MS-Celeb-1M dataset. Advantageous results on LFW, IJB-A and MS-Celeb-1M demonstrate the effectiveness of our feature transfer and training strategy, compared to both general baselines and state-of-the-art methods. Moreover, our feature transfer successfully presents smooth visual interpolation, which conducts disentanglement to preserve identity of a class while augmenting its feature space with non-identity variations such as pose and lighting.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“When the data has a long-tail distribution, it can lead to bias in the classifier.”
From An Improved VGG19 Transfer Learning Strip Steel Surface Defect Recognition Deep Neural Network Based on Few Samples and Imbalanced Datasets (2021) · cites this paper as [4]
In the citing paper
“The classifier is more inclined to identify classes with a sufficient sample size and rich diversity [4,5].”
From An Improved VGG19 Transfer Learning Strip Steel Surface Defect Recognition Deep Neural Network Based on Few Samples and Imbalanced Datasets (2021) · cites this paper as [4]
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380
SMOTE synthetic minority over-sampling technique
Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002) · Journal of Artificial Intelligence Research · doi:10.1613/jair.953 · Cited by 32,865
Pan, S. J., & Yang, Q. (2010) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2009.191 · Cited by 23,387
Labaca-Castro, R. (2023) · Springer eBooks · doi:10.1007/978-3-658-40442-0_9 · Cited by 21,119
GAN（Generative Adversarial Nets）
淳司, 柴. (2017) · Journal of Japan Society for Fuzzy Theory and Intelligent Informatics · doi:10.3156/jsoft.29.5_177_2 · Cited by 20,507
FaceNet: A unified embedding for face recognition and clustering
Schroff, F., Kalenichenko, D., & Philbin, J. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298682 · Cited by 15,267
Unsupervised representation learning with deep convolutional generative adversarial networks
Radford, A., Metz, L., & Chintala, S. (2015) · arXiv · doi:10.48550/arxiv.1511.06434 · Cited by 12,235
He, H., & Garcia, E. A. (2009) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2008.239 · Cited by 10,359
Flagged first, then by citations
1 of the 50 references couldn’t be matched to a known paper.
Research works printed but not linked: 5
- van der maaten, 2008, Visualizing high-dimensional data using t-SNE, Journal of Machine Learning Research, 9, 2579
- lecun, 1998, The MNIST database of handwritten digits, Technical Report
- wang, 2017, NormFace L2 Hypersphere Embedding for Face Verification
- horn, 2017, The devial is in the tails Finegrained classification in the wild
- 10.1109/FG.2017.11 · DOI
Cited by
Sorted by citations
Items 301–347 of 347
Camera Style Guided Feature Generation for Person Re-identification
Hu, H., Liu, Y., Lv, K., Zheng, Y., Zhang, W., Ke, W., & Sheng, H. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-59016-1_14
Imbalance Robust Softmax for Deep Embeeding LearningPreprint
Zhu, H., Yuan, Y., Hu, G., Wu, X., & Robertson, N. M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.11155
“or face recognition, identities with few images cannot provide enough intraclass information for the model, but provide inter-class information which is more useful to open-set protocol. Many methods [25,34,35,36,37] have been proposed to solve the data imbalance in face recognition. However,these works do not deeply…”
BackgroundSpotPatch: Parameter-Efficient Transfer Learning for Mobile Object DetectionPreprint
Ye, K., Kovashka, A., Sandler, M., Zhu, M., Howard, A., & Fornoni, M. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2101.01260
“Some transfer and adaptation techniques minimize the discrepancy between tasks or domains in terms of the feature representation [25,26,27,28,29,30,31,32,33,34].”
BackgroundRecognition and Processing of NATOMPreprint
Deng, Y., & Luo, Y. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2105.03314
“Yin[5] proposed a center-based feature transfer framework to expand the feature space of under represented subjects in a sufficiently diverse routine.”
BackgroundRSG: A Simple but Effective Module for Learning Imbalanced DatasetsPreprint
Wang, J., Lukasiewicz, T., Hu, X., Cai, J., & Xu, Z. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.09859
Training Over-parameterized Models with Non-decomposable ObjectivesPreprint
Narasimhan, H., & Menon, A. K. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2107.04641
“[33, 39]), but we may roughly identify three main strands of work: data modification [48, 11, 93, 103, 108], loss modification [107, 18, 10, 88, 37, 79, 97, 63, 21, 45, 94], and prediction modification [28, 43, 76, 60, 13, 40, 110].”
Face Recognition for Examinee Verification
Anand, N., Sharma, T., & Gaurav, K. (2022) · INTERNATIONAL JOURNAL OF NEXT-GENERATION COMPUTING · doi:10.47164/ijngc.v13i3.668
Entity Factor: A Balanced Method for Table Filling in Joint Entity and Relation Extraction
Liu, Z., Tao, M., & Zhou, C. (2022) · Electronics · doi:10.3390/electronics12010121
WeightFace: weight adaptive scaling loss for face recognition
Ren, H., & Yang, X. (2023) · Multimedia Tools and Applications · doi:10.1007/s11042-023-15085-7
Poster: Link between Bias, Node Sensitivity and Long-Tail Distribution in trained DNNs
Naseer, M., & Shafique, M. (2023) · IEEE International Conference on Software Testing, Verification and Validation (ICST) · doi:10.1109/icst57152.2023.00054
The Progress of Face Recognition on Artificial Intelligence
Xiao, H., Wei, L., Li, Z., Gan, G., Wu, D., & Jiang, L. (2022) · International Symposium on Smart and Healthy Cities (ISHC) · doi:10.1109/ishc56805.2022.00048
“Yin et al.[11] introduced a method to address imbalanced training data.”
MethodBalanced complement loss for long-tailed image classification
Hu, L., Yang, Z., Dou, Y., & Li, J. (2023) · Multimedia Tools and Applications · doi:10.1007/s11042-023-17583-0
“Transfer learning [5, 9, 27, 42] is applied as data augmentation technologies to migrate the information from head classes to tail classes, and then the tail classes can generate a new feature space to achieve class balance.”
Towards Representation Alignment and Uniformity in Long-tailed Classification
Zheng, Y., & Meng, Z. (2023) · ACM International Conference on Multimedia in Asia (MMAsia) · doi:10.1145/3595916.3626433
“Knowledge transfer[6, 23, 33, 40] aims to transfer information from head classes to tail classes, leveraging the abundance of majority samples to enhance the diversity of minority samples.”
BackgroundEffect of Smile on Facial Landmark based Face Recognition
Dey, S., Hossain, M. M., Hassan, A. M., Rahman, A., & Tarin, T. (2023) · Conference proceedings/Conference proceedings - IEEE International Conference on Systems, Man, and Cybernetics · doi:10.1109/smc53992.2023.10394510
“Deep convolutional neural networks (CNNs) [7], siamese networks [8], 3D face recognition, transfer learning [9], face recognition with characteristics, and texture-based techniques like Local Binary Patterns [10], Gabor filters [11] and Eigenfaces [12] are examples of some of the mostly used face…”
BackgroundA Query-Specific Representation-Based Sampling Strategy for Imbalanced ClassificationPreprint
Tan, Y., Qu, Y., Yang, L., Liu, H., Shang, C., & Shen, Q. (2024) · doi:10.36227/techrxiv.171073556.66443984/v1
Class-Adaptive Threshold for Class Imbalanced Semi-Supervised Learning
Liu, X., Wei, X., Wang, J., Lu, S., & Xing, W. (2023) · IEEE International Conference on Parallel and Distributed Systems (ICPADS) · doi:10.1109/icpads60453.2023.00181
Self-citationCnn-Based Palm Vein Confirmation In An Integrated Biometrics Id Configuration
Teotia, K., & Bansal, M. (2024) · doi:10.53555/kuey.v30i4.1827
HPO-LGBM-DRI: Dynamic Recognition Interval Estimation for Imbalanced Fraud Call via HPO-LGBM
Liu, X., Zhi, X., Mei, Q., Wang, P., Su, H., & Wang, J. (2024) · Lecture notes in computer science · doi:10.1007/978-981-97-2966-1_24
Self-citationLtb-Solver: Long-Tailed Bias Solver for Image Synthesis of Diffusion ModelsPreprint
He, X., Fu, S., & Hu, H. R. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.4822238
Towards Data-Centric Long-Tailed Image RecognitionPreprint
Ma, Y., Jiao, L., Liu, F., Li, L., Ma, W., Liu, X., Chen, P., & Yang, S. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.4826649
Self-citationMoung, E. G., Dargham, J. A., & Khoo, J. J. D. (2024) · Internet of things · doi:10.1007/978-981-97-1432-2_10
How does Rejecting Low-Confidence Decisions in Face Recognition Affect Fairness?
Saha, P., Sinha, A., & Terhörst, P. (2024) · International Workshop on Biometrics and Forensics (IWBF) · doi:10.1109/iwbf62628.2024.10593911
“The majority of previous works aim at learning less-biased models by feature transfer and training strategies [14], skewness-aware reinforcement learning [1], adversarial learning [5], [15], or margin-based approach [16].”
BackgroundPan, L., Li, X., Wang, Z., Zhang, R., Yang, N., & Shan, W. (2024) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn60899.2024.10650276
“Focal Transformation Learning (FTL) [20] utilizes intra-class variance knowledge from head labels to guide feature augmentation for tail-label samples.”
MethodClass Incremental Aerial Scene Recognition Under Long-Tailed Distribution
Yang, G., Ding, H., Tu, X., Huang, L., Huang, Y., & Ding, X. (2024) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn60899.2024.10650815
“To address this issue, learning from long-tailed distributions methods such as re-sampling [13], [14], re-weighting [15], [16], transfer learning [17], [18], two-stage learning [19], [20], and ensemble learning [21], [22] have been approached.”
BackgroundMethodJalali, A., Lee, S., & Lee, M. (2024) · Applied Soft Computing · doi:10.1016/j.asoc.2024.112394
Trade-Offs between Richness and Bias of Augmented Data in Long-Tailed RecognitionPreprint
Ma, Y., Dai, W., Chen, J., Chen, X., & Li, S. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.5012834
Verma, R., & Das, S. (2024) · Indian Conference on Computer Vision, Graphics and Image Processing · doi:10.1145/3702250.3702285
“Decoupling methods, which address imbalance mainly at the classifier stage, improve the performance of tail classes but often negatively impact head classes [33].”
Maximum Margin-Based Activation Clipping for Posttraining Overfitting Mitigation in DNN Classifiers
Wang, H., Miller, D. J., & Kesidis, G. (2025) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2025.3552686
“Transfer Learning methods [13], [17], [28], [30] transfer knowledge learned from common classes to rare classes.”
A Multimodal Approach to Biometric Authentication
Kozierkiewicz, A., Graczyk, A., & Pimenidis, E. (2025) · Advanced sciences and technologies for security applications · doi:10.1007/978-3-031-82031-1_24 · Cited by 1
Innovative Imaging Solutions for Tuberculosis Discovery by Deep Learning
Rao, B. S. K., & Maheshwary, P. (2025) · Lecture notes in networks and systems · doi:10.1007/978-981-96-1264-2_15
From Expansion to Retraction: Long-tailed Machine Unlearning via Boundary Manipulation
Chen, M., Gao, W., Wang, C., Liu, G., Abdelmoniem, A. M., & Peng, K. (2025) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3711896.3736970
“For instance, Yin et al. [39] generate additional features for tail classes by leveraging variation information from head classes, and then optimize the model with these additional tail features to reduce the decision boundary bias.”
PAIR: Privacy Attack Against Image Recognition With Sample Augmentation
Feng, Y., & Liu, Q. (2025) · IET Image Processing · doi:10.1049/ipr2.70190
DBBA: Diffusion-Based Backdoor Attacks on Open-Set Face Recognition Models
Qi, F., Gao, H., Bo-ling, L., He, G., Zhang, Y., & Luo, J. (2025) · Lecture notes in computer science · doi:10.1007/978-3-032-07884-1_16
Long-Tailed Federated Learning with Fixed Classifier
Li, Y., Li, W., Zheng, X., Fu, H., & Guo, Y. (2025) · IEEE International Conference on Multimedia and Expo (ICME) · doi:10.1109/icme59968.2025.11209954
“As expected, in long-tail learning, due to the larger norm of the vector weights of the head classifier, the hidden layer features are more likely to be classified into head classes [18].”
Mitigating Lipschitz singularities in long-tailed diffusion models via time-step sharing strategy
Leng, Q., Zhou, Z., Song, K., & Yuan, G. (2025) · The Visual Computer · doi:10.1007/s00371-025-04312-x
“Another line of research focuses on information augmentation strategies [20, 21], which aim to improve the learning of tail classes by transferring informative knowl-123 edge from head classes.”
A Comprehensive Survey of Transfer Learning Approaches for Medical Image Deep Analysis
Bhargavi, C. V., & Subhadra, K. (2025) · BENTHAM SCIENCE PUBLISHERS eBooks · doi:10.2174/9798898812072125020005
Calibrating Biased Distribution in VFM-Derived Latent Space via Cross-Domain Geometric Consistency
Ma, Y., Dai, W., Liu, B., Chen, J., Huang, W., Wan, G., Lu, Z., & Yan, J. (2026) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2026.3662389
“For instance, in long-tailed learning, FTL [59] and FDC [15] assume that the feature distributions of the common and UR classes (i.e., rare classes) have the same variance, so the variance from the head classes is used to guide the feature enhancement of the tail classes.”
Learning salient information with neural networks for cosmological analysis
Lin, Q. (2021) · theses.fr (ABES) · doi:10.70675/044e7500zfd93z4c6az8457z4809ad535dae
Mapping Irrigated Areas using Radar and Optical Remote Sensing Data
Bazzi, H. (2021) · theses.fr (ABES) · doi:10.70675/f36bf8dbzfb3dz4b7az8dfez7343c0b5b9ba
Temperature in Cosine-based Softmax Loss
Kobayashi, T. (2025) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.02061
“The regularization could be helpful for improving performance on long-tailed class distributions which induce imbalanced intra-class feature variances across head (majority) and tail (minority) classes as shown in [44].”
Zhou, Z., Leng, Q., Song, K., & Yuan, G. (2026) · The Journal of Supercomputing · doi:10.1007/s11227-026-08558-7
Gradient Distribution Balancing for Dual Imbalanced Long-Tailed Learning
Yang, H., Li, Y., Jiang, X., Zhou, J., & Xu, Y. (2026) · Communications in computer and information science · doi:10.1007/978-981-95-9507-5_5
Heterogeneous Debiasing Contrastive Learning for Graph Recommendation
Sang, L., Cheng, J., Mu, L., Zhang, Y., & Wu, X. (2026) · IEEE Transactions on Systems Man and Cybernetics Systems · doi:10.1109/tsmc.2026.3703824
“Common strategies to mitigate this issue include post hoc decision correction [27], oversampling or downsampling techniques [28], [29], and loss function modification [30].”
A Prototype Network Assisted Training Mechanism for Edge Federated Learning Models
Yuan, X., & Shi, L. (2026) · Lecture notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering · doi:10.1007/978-3-032-22500-9_6
Illumination-Aware Feature Fusion for Human Pose Estimation in Extreme Low Light
Wang, Z. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-31236-5_17
Metadata
- Authors
- Yin, X., Yu, X., Sohn, K., Liu, X., & Chandraker, M.
- Affiliations
- Microsoft (United States) (US)
- NEC (Japan) (JP)
- NEC (United States) (US)
- Michigan State University (US)
- University of California San Diego (US)
- Venue
- 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
- Publisher
- IEEE
- Published
- 2019-06
- Pages
- 5697-5706
- ISSN
- 1063-6919 (ISSN-L)
- OpenAlex
- W2962895364
- Cited by
- 346
- References
- 50
- Citations by year
- 2026: 7 · 2025: 28 · 2024: 63 · 2023: 67 · 2022: 57 · 2021: 69 · 2020: 49 · 2019: 6
- FWCI
- 14.75
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Face recognition and analysis
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Face and Expression Recognition
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Keywords
- face recognition, underrepresented classes, classifier bias, alternating training, feature space augmentation, disentangled representations
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
