pith. sign in

arxiv: 2607.02271 · v1 · pith:P3NMWJ4Knew · submitted 2026-07-02 · 💻 cs.CV

AGVBench: A Reliability-Oriented Benchmark of Data Augmentation for Vein Recognition

Haiyang Li , Yuming Fu , Qun Song , Hongchao Liao , Jing Chen , Mounim A.EI-Yacoubi , Xin Jin This is my paper

Pith reviewed 2026-07-03 15:41 UTC · model grok-4.3

classification 💻 cs.CV
keywords vein recognitiondata augmentationbenchmarkadversarial robustnessbiometric securitycalibrationpalm veinfinger vein
0
0 comments X

The pith

Mixing data augmentations improve vein recognition but increase vulnerability to adversarial attacks.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents AGVBench to evaluate thirty data augmentation strategies for vein recognition on five datasets using seven different model architectures. It finds that methods that mix multiple images tend to give the best accuracy on normal images. These same methods, however, tend to be poorly calibrated and easily disrupted by adversarial examples. The work shows that judging augmentations only by accuracy misses important reliability problems in biometric systems.

Core claim

The central claim is that multi-image mixing methods such as MixUp, PuzzleMix, and StarMixup deliver the strongest recognition performance on palm- and finger-vein data, yet they are typically poorly calibrated and vulnerable to adversarial perturbations, which reveals an inconsistency between clean accuracy and adversarial security that makes accuracy-centric evaluation insufficient for biometric augmentation.

What carries the argument

AGVBench, the benchmark that applies thirty augmentation strategies to five public vein datasets across seven backbone architectures while measuring recognition accuracy, calibration quality, and adversarial robustness.

If this is right

  • Multi-image mixing methods generally provide the strongest recognition performance.
  • These methods are often poorly calibrated.
  • They are vulnerable to adversarial perturbations.
  • Severe geometric transformations frequently degrade recognition performance.
  • Augmentation effectiveness varies across palm and finger vein datasets.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Designers of vein recognition systems should consider adversarial security when choosing augmentations rather than accuracy alone.
  • The accuracy-security inconsistency may apply to other recognition tasks that depend on fine topological features.
  • The evaluation protocols could be extended to additional biometric modalities to test the generality of the findings.

Load-bearing premise

The five public palm- and finger-vein datasets together with the seven backbone architectures are sufficiently representative to support general conclusions about augmentation effectiveness and the accuracy-security trade-off across vein recognition.

What would settle it

Running the same evaluation protocol on new independent vein datasets where the top mixing methods achieve high accuracy without increased adversarial vulnerability or calibration problems would falsify the claimed inconsistency.

Figures

Figures reproduced from arXiv: 2607.02271 by Haiyang Li, Hongchao Liao, Jing Chen, Mounim A.EI-Yacoubi, Qun Song, Xin Jin, Yuming Fu.

Figure 1
Figure 1. Figure 1: Comprehensive performance evaluation of various data augmentation and regularization methods. (a) Trade-off of Accuracy [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Visualization of various data augmentation techniques applied to a sample vein image. The original (Vanilla) image is shown alongside standard [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the AGVBench codebase framework. Built upon PyTorch and MMCV, the framework is structured into four functional modules: (1) [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Experimental pipeline in AGVBench codebase. The workflow [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Receiver Operating Characteristic (ROC) curves of various data augmentation methods across five vein datasets using the ResNet18 backbone. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The confidence plots of different augmentations on the VERA220 dataset using ResNet18. The red line indicates the expected prediction tendency. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Top-1 Accuracy under varying occlusion ratios (0% [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Receiver Operating Characteristic (ROC) curves of various data augmentation methods across two palm-vein datasets using different backbones. The [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Receiver Operating Characteristic (ROC) curves of various data augmentation methods across two palm-vein datasets using different backbones. The [PITH_FULL_IMAGE:figures/full_fig_p018_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Receiver Operating Characteristic (ROC) curves of various data augmentation methods across SDUMLA-HMT datasets using different backbones. [PITH_FULL_IMAGE:figures/full_fig_p019_10.png] view at source ↗
read the original abstract

Vein recognition is a secure biometric technology often constrained by limited annotated data and imaging variations. While data augmentation mitigates this, strategies designed for natural images may disrupt the fine-grained topology and textures essential for identity discrimination. We present AGVBench, which evaluates 30 representative augmentation strategies on five public palm- and finger-vein datasets with seven backbone architectures, covering classic CNNs, vision transformers, and vein-specific recognition models. Our results show that multi-image mixing methods (e.g., MixUp, PuzzleMix, StarMixup) generally provide the strongest recognition performance. However, they are often poorly calibrated and vulnerable to adversarial perturbations, revealing a clear inconsistency between clean accuracy and adversarial security. We also find that severe geometric transformations frequently degrade recognition, which is potentially due to feature misalignment or spatial cropping, and that augmentation effectiveness varies across palm and finger vein datasets. These findings prove that accuracy-centric evaluation is insufficient for biometric augmentation. AGVBench provides standardized protocols to support reproducible research and guide the design of reliable, secure, and robust vein recognition systems. Our codebase is available at https://github.com/Advance-VeinTech-Innovators/AGVBench.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper presents AGVBench, an empirical benchmark evaluating 30 data augmentation strategies across five public palm- and finger-vein datasets and seven backbone architectures (CNNs, vision transformers, and vein-specific models). It reports that multi-image mixing methods (MixUp, PuzzleMix, StarMixup) achieve the highest clean recognition accuracy but exhibit poor calibration and high vulnerability to adversarial perturbations, demonstrating an inconsistency between clean accuracy and adversarial security. Additional findings include degradation from severe geometric transformations and variation in augmentation effectiveness between palm and finger vein data. The work provides standardized protocols and open code to support reliability-oriented evaluation in vein recognition.

Significance. If the empirical results hold under broader conditions, the benchmark supplies concrete evidence that accuracy-centric augmentation evaluation is insufficient for biometric systems and identifies a systematic accuracy-security trade-off in mixing-based methods. The release of reproducible code and protocols is a clear strength that enables follow-on work.

major comments (2)
  1. [§3] §3 (Datasets and Models): The central claim of a 'clear inconsistency between clean accuracy and adversarial security' for multi-image mixing methods rests on results from only five public datasets and seven backbones. The manuscript does not demonstrate that these collections adequately sample the range of sensor resolutions, cross-session variability, and demographic factors typical in vein imaging; if they under-represent low-contrast or cross-session regimes, the observed trade-off may not generalize.
  2. [§4.3] §4.3 (Adversarial Evaluation): The vulnerability findings for mixing methods are load-bearing for the headline inconsistency result, yet the text provides no explicit values for attack parameters (e.g., PGD step size, number of iterations, or ε bounds) or confirmation that the same attack configuration was applied uniformly across all augmentation strategies and models.
minor comments (2)
  1. [Abstract] The abstract states that 'severe geometric transformations frequently degrade recognition' but does not quantify the degradation (e.g., accuracy drop percentages) or link it to specific tables/figures.
  2. [Tables] Table captions and axis labels should explicitly state the number of runs or random seeds used to compute reported means and standard deviations.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We respond to each major comment below and note the planned revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [§3] §3 (Datasets and Models): The central claim of a 'clear inconsistency between clean accuracy and adversarial security' for multi-image mixing methods rests on results from only five public datasets and seven backbones. The manuscript does not demonstrate that these collections adequately sample the range of sensor resolutions, cross-session variability, and demographic factors typical in vein imaging; if they under-represent low-contrast or cross-session regimes, the observed trade-off may not generalize.

    Authors: We agree that the five public datasets used represent the standard, reproducible collections available in the literature rather than a comprehensive sampling of all possible vein imaging conditions. While these datasets span palm and finger veins with documented variations in resolution and acquisition, they do not explicitly cover the full range of demographic factors or cross-session variability. We will add a dedicated Limitations subsection to the discussion that states this scope explicitly and notes that the released codebase is designed to support extension to new datasets. This revision will qualify the generalizability of the observed accuracy-security trade-off without altering the reported empirical results. revision: partial

  2. Referee: [§4.3] §4.3 (Adversarial Evaluation): The vulnerability findings for mixing methods are load-bearing for the headline inconsistency result, yet the text provides no explicit values for attack parameters (e.g., PGD step size, number of iterations, or ε bounds) or confirmation that the same attack configuration was applied uniformly across all augmentation strategies and models.

    Authors: We thank the referee for identifying this omission. The adversarial evaluations used PGD with ε = 0.03, step size 0.01, and 20 iterations, applied uniformly to every augmentation strategy and backbone. We will revise §4.3 to state these parameters explicitly and add a sentence confirming uniform application across all experiments. revision: yes

Circularity Check

0 steps flagged

Pure empirical benchmark with no derivations or self-referential predictions

full rationale

The paper conducts a direct empirical evaluation of 30 augmentation strategies across five public datasets and seven standard backbones, reporting observed accuracy, calibration, and adversarial robustness metrics. No equations, fitted parameters, uniqueness theorems, or ansatzes are introduced; all claims reduce to tabulated experimental outcomes on external data rather than any internal construction or self-citation chain. This is the expected non-circular outcome for a benchmark study.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The evaluation rests on the domain assumption that the selected public datasets capture typical imaging variations in vein biometrics and that the chosen backbones represent current practice.

axioms (1)
  • domain assumption The five public palm- and finger-vein datasets are representative of real-world vein recognition conditions.
    Invoked to generalize findings from the benchmark results.

pith-pipeline@v0.9.1-grok · 5761 in / 1093 out tokens · 31248 ms · 2026-07-03T15:41:51.023775+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

56 extracted references · 7 canonical work pages · 3 internal anchors

  1. [1]

    El-Yacoubi, and Dexing Zhong

    Huafeng Qin, Yuming Fu, Jing Chen, Qun Song, Yantao Li, Mounim A. El-Yacoubi, and Dexing Zhong. Wtxgrn: Wavelet transform-based extended gated recurrent network for palm vein recognition.IEEE Transactions on Information Forensics and Security, 20:7911–7926, 2025

    2025

  2. [2]

    Palm vein recognition under unconstrained and weak-cooperative con- ditions.IEEE Transactions on Information Forensics and Security, 19:4601–4614, 2024

    Dacan Luo, Yitao Qiao, Di Xie, Shifeng Zhang, and Wenxiong Kang. Palm vein recognition under unconstrained and weak-cooperative con- ditions.IEEE Transactions on Information Forensics and Security, 19:4601–4614, 2024

    2024

  3. [3]

    Multi-scale and multi-direction gan for cnn-based single palm-vein identification.IEEE Transactions on Information Forensics and Security, 16:2652–2666, 2021

    Huafeng Qin, Mounim A El-Yacoubi, Yantao Li, and Chongwen Liu. Multi-scale and multi-direction gan for cnn-based single palm-vein identification.IEEE Transactions on Information Forensics and Security, 16:2652–2666, 2021

    2021

  4. [4]

    A new filter generation method in pcanet for finger vein recognition.IEEE Access, 7:132966–132978, 2019

    Nurul Maisarah Kamaruddin and Bakhtiar Affendi Rosdi. A new filter generation method in pcanet for finger vein recognition.IEEE Access, 7:132966–132978, 2019

    2019

  5. [5]

    Autoaugment: Learning augmentation strategies from data

    Ekin D Cubuk, Barret Zoph, Dandelion Mane, Vijay Vasudevan, and Quoc V Le. Autoaugment: Learning augmentation strategies from data. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 113–123, 2019

    2019

  6. [6]

    Randaugment: Practical automated data augmentation with a reduced search space

    Ekin D Cubuk, Barret Zoph, Jonathon Shlens, and Quoc V Le. Randaugment: Practical automated data augmentation with a reduced search space. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, pages 702–703, 2020

    2020

  7. [7]

    Improved Regularization of Convolutional Neural Networks with Cutout

    Terrance DeVries and Graham W Taylor. Improved regularization of con- volutional neural networks with cutout.arXiv preprint arXiv:1708.04552, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  8. [8]

    mixup: Beyond empirical risk minimization

    Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. mixup: Beyond empirical risk minimization. InInternational Conference on Learning Representations, 2018

    2018

  9. [9]

    Starmixup: A more suitable mixup method for palm-vein identification

    Xin Jin, Hongyu Zhu, Simon Fong, João Alexandre Lobo Marques, Huafeng Qin, and Yun Jiang. Starmixup: A more suitable mixup method for palm-vein identification. In2025 7th International Symposium on Computational and Business Intelligence (ISCBI), pages 83–87. IEEE, 2025

    2025

  10. [10]

    Automix: Unveiling the power of mixup for stronger classifiers

    Zicheng Liu, Siyuan Li, Di Wu, Zihan Liu, Zhiyuan Chen, Lirong Wu, and Stan Z Li. Automix: Unveiling the power of mixup for stronger classifiers. InEuropean Conference on Computer Vision, pages 441–458. Springer, 2022

    2022

  11. [11]

    Adversarial automixup

    Huafeng Qin, Xin Jin, Yun Jiang, Mounîm El-Yacoubi, and Xinbo Gao. Adversarial automixup. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  12. [12]

    Guidedmixup: an efficient mixup strategy guided by saliency maps

    Minsoo Kang and Suhyun Kim. Guidedmixup: an efficient mixup strategy guided by saliency maps. InProceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 1096–1104, 2023

    2023

  13. [13]

    Trivialaugment: Tuning-free yet state-of-the-art data augmentation

    Samuel G Müller and Frank Hutter. Trivialaugment: Tuning-free yet state-of-the-art data augmentation. InProceedings of the IEEE/CVF international conference on computer vision, pages 774–782, 2021

    2021

  14. [14]

    Teachaugment: Data augmentation optimization using teacher knowledge

    Teppei Suzuki. Teachaugment: Data augmentation optimization using teacher knowledge. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10904–10914, 2022

    2022

  15. [15]

    Cutmix: Regularization strategy to train strong classifiers with localizable features

    Sangdoo Yun, Dongyoon Han, Seong Joon Oh, Sanghyuk Chun, Junsuk Choe, and Youngjoon Yoo. Cutmix: Regularization strategy to train strong classifiers with localizable features. InInternational Conference on Computer Vision (ICCV), pages 6023–6032, 2019

    2019

  16. [16]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778, 2016

    2016

  17. [17]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weis- senborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. InInternational Conference on Learning Represen- tations (ICLR), 2021

    2021

  18. [18]

    Random erasing data augmentation

    Zhun Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, and Yi Yang. Random erasing data augmentation. InProceedings of the AAAI conference on artificial intelligence, volume 34, pages 13001–13008, 2020

    2020

  19. [19]

    Keepaugment: A simple information-preserving data augmentation approach

    Chengyue Gong, Dilin Wang, Meng Li, Vikas Chandra, and Qiang Liu. Keepaugment: A simple information-preserving data augmentation approach. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1055–1064, 2021

    2021

  20. [20]

    Gridmix: Strong regularization through local context mapping.Pattern Recognition, 109:107594, 2021

    Kyungjune Baek, Duhyeon Bang, and Hyunjung Shim. Gridmix: Strong regularization through local context mapping.Pattern Recognition, 109:107594, 2021

    2021

  21. [21]

    Data augmen- tation using random image cropping and patching for deep cnns.IEEE Transactions on Circuits and Systems for Video Technology, 30(9):2917– 2931, 2019

    Ryo Takahashi, Takashi Matsubara, and Kuniaki Uehara. Data augmen- tation using random image cropping and patching for deep cnns.IEEE Transactions on Circuits and Systems for Video Technology, 30(9):2917– 2931, 2019

    2019

  22. [22]

    Saliencymix: A saliency guided data augmen- tation strategy for better regularization

    AFM Shahab Uddin, Mst Sirazam Monira, Wheemyung Shin, TaeChoong Chung, and Sung-Ho Bae. Saliencymix: A saliency guided data augmen- tation strategy for better regularization. InInternational Conference on Learning Representations, 2020

    2020

  23. [23]

    Puzzle mix: Exploiting saliency and local statistics for optimal mixup

    Jang-Hyun Kim, Wonho Choo, and Hyun Oh Song. Puzzle mix: Exploiting saliency and local statistics for optimal mixup. InInternational Conference on Machine Learning, pages 5275–5285. PMLR, 2020

    2020

  24. [24]

    Rethinking the inception architecture for computer vision

    Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 2818–2826, 2016

    2016

  25. [25]

    Regularizing neural networks by penalizing confident output distributions

    Gabriel Pereyra, George Tucker, Jan Chorowski, Łukasz Kaiser, and Geof- frey Hinton. Regularizing neural networks by penalizing confident output distributions. InInternational Conference on Learning Representations, 2017

    2017

  26. [26]

    Delving deep into label smoothing

    Chang-Bin Zhang, Peng-Tao Jiang, Qibin Hou, Yunchao Wei, Qi Han, Zhen Li, and Ming-Ming Cheng. Delving deep into label smoothing. IEEE Transactions on Image Processing, 30:5984–5996, 2021

    2021

  27. [27]

    Self-progressing robust training

    Minhao Cheng, Pin-Yu Chen, Sijia Liu, Shiyu Chang, Cho-Jui Hsieh, and Payel Das. Self-progressing robust training. InProceedings of the AAAI conference on artificial intelligence, volume 35, pages 7107–7115, 2021

    2021

  28. [28]

    Enhancement of finger-vein image by vein line tracking and adaptive gabor filtering for finger-vein recognition.Applied Mechanics and Materials, 145:219–223, 2012

    So Ra Cho, Young Ho Park, Gi Pyo Nam, Kwang Youg Shin, Hyeon Chang Lee, Kang Ryoung Park, Sung Min Kim, and Ho Chul AGVBENCH: A RELIABILITY-ORIENTED BENCHMARK OF DATA AUGMENTATION FOR VEIN RECOGNITION 14 Kim. Enhancement of finger-vein image by vein line tracking and adaptive gabor filtering for finger-vein recognition.Applied Mechanics and Materials, 145...

    2012

  29. [29]

    Contactless palm vein recognition using a mutual foreground-based local binary pattern.IEEE transactions on Information Forensics and Security, 9(11):1974–1985, 2014

    Wenxiong Kang and Qiuxia Wu. Contactless palm vein recognition using a mutual foreground-based local binary pattern.IEEE transactions on Information Forensics and Security, 9(11):1974–1985, 2014

    1974

  30. [30]

    Enhanced maximum curvature descriptors for finger vein verification

    Munalih Ahmad Syarif, Thian Song Ong, Andrew BJ Teoh, and Connie Tee. Enhanced maximum curvature descriptors for finger vein verification. Multimedia Tools and Applications, 76(5):6859–6887, 2017

    2017

  31. [31]

    A palm vein recognition system based on a support vector machine.IEIE Transactions on Smart Processing & Computing, 8(1):1–7, 2019

    Vijayakumar Ponnusamy, Abhijit Sridhar, Arun Baalaaji, and M Sangeetha. A palm vein recognition system based on a support vector machine.IEIE Transactions on Smart Processing & Computing, 8(1):1–7, 2019

    2019

  32. [32]

    Finger vein recognition using mutual sparse representation classification.IET biometrics, 8(1):49– 58, 2019

    Shazeeda Shazeeda and Bakhtiar Affendi Rosdi. Finger vein recognition using mutual sparse representation classification.IET biometrics, 8(1):49– 58, 2019

    2019

  33. [33]

    Fvras-net: An embedded finger-vein recognition and antispoofing system using a unified cnn.IEEE Transactions on Instrumentation and Measurement, 69(11):8690–8701, 2020

    Weili Yang, Wei Luo, Wenxiong Kang, Zhixing Huang, and Qiuxia Wu. Fvras-net: An embedded finger-vein recognition and antispoofing system using a unified cnn.IEEE Transactions on Instrumentation and Measurement, 69(11):8690–8701, 2020

    2020

  34. [34]

    Finger vein recognition algorithm based on lightweight deep convolutional neural network.IEEE Transactions on Instrumentation and Measurement, 71:1–13, 2021

    Jiaquan Shen, Ningzhong Liu, Chenglu Xu, Han Sun, Yushun Xiao, Deguang Li, and Yongxin Zhang. Finger vein recognition algorithm based on lightweight deep convolutional neural network.IEEE Transactions on Instrumentation and Measurement, 71:1–13, 2021

    2021

  35. [35]

    Saad Shakeel, and Wenxiong Kang

    Dacan Luo, Junduan Huang, Weili Yang, M. Saad Shakeel, and Wenxiong Kang. Rsnet: Region-specific network for contactless palm vein authentication.IEEE Transactions on Information Forensics and Security, 20:2734–2747, 2025

    2025

  36. [36]

    Gridmask data augmentation.arXiv preprint arXiv:2001.04086, 2020

    Pengguang Chen, Shu Liu, Hengshuang Zhao, Xingquan Wang, and Jiaya Jia. Gridmask data augmentation.arXiv preprint arXiv:2001.04086, 2020

    work page arXiv 2001

  37. [37]

    You only cut once: Boosting data augmentation with a single cut

    Junlin Han, Pengfei Fang, Weihao Li, Jie Hong, Mohammad Ali Armin, Ian Reid, Lars Petersson, and Hongdong Li. You only cut once: Boosting data augmentation with a single cut. InInternational Conference on Machine Learning, pages 8196–8212. PMLR, 2022

    2022

  38. [38]

    Soft augmentation for image classification

    Yang Liu, Shen Yan, Laura Leal-Taixé, James Hays, and Deva Ramanan. Soft augmentation for image classification. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16241–16250, 2023

    2023

  39. [39]

    Randomized quantization: A generic augmentation for data agnostic self-supervised learning

    Huimin Wu, Chenyang Lei, Xiao Sun, Peng-Shuai Wang, Qifeng Chen, Kwang-Ting Cheng, Stephen Lin, and Zhirong Wu. Randomized quantization: A generic augmentation for data agnostic self-supervised learning. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 16305–16316, 2023

    2023

  40. [40]

    Resizemix: Mixing data with preserved object information and true labels.arXiv preprint arXiv:2012.11101, 2020

    Jie Qin, Jiemin Fang, Qian Zhang, Wenyu Liu, Xingang Wang, and Xinggang Wang. Resizemix: Mixing data with preserved object information and true labels.arXiv preprint arXiv:2012.11101, 2020

    work page arXiv 2012

  41. [41]

    Fmix: Enhancing mixed sample data augmentation.arXiv preprint arXiv:2002.12047, 2020

    Ethan Harris, Antonia Marcu, Matthew Painter, Mahesan Niranjan, and Adam Prügel-Bennett Jonathon Hare. Fmix: Enhancing mixed sample data augmentation.arXiv preprint arXiv:2002.12047, 2020

    work page arXiv 2002

  42. [42]

    Training Deep Neural Networks on Noisy Labels with Bootstrapping

    Scott Reed, Honglak Lee, Dragomir Anguelov, Christian Szegedy, Dumitru Erhan, and Andrew Rabinovich. Training deep neural networks on noisy labels with bootstrapping.arXiv preprint arXiv:1412.6596, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  43. [43]

    Mobilenetv2: Inverted residuals and linear bottlenecks

    Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. In2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4510–4520, 2018

    2018

  44. [44]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. InInternational Conference on Computer Vision (ICCV), 2021

    2021

  45. [45]

    Palmprint and palmvein recognition based on dcnn and a new large-scale contactless palmvein dataset.Symmetry, 10(4):78, 2018

    Lin Zhang, Zaixi Cheng, Ying Shen, and Dongqing Wang. Palmprint and palmvein recognition based on dcnn and a new large-scale contactless palmvein dataset.Symmetry, 10(4):78, 2018

    2018

  46. [46]

    On the vulnerability of palm vein recognition to spoofing attacks

    Pedro Tome and Sébastien Marcel. On the vulnerability of palm vein recognition to spoofing attacks. In2015 International Conference on Biometrics (ICB), pages 319–325. IEEE, 2015

    2015

  47. [47]

    Fusion of band limited phase only correlation and width centroid contour distance for finger based biometrics.Expert Systems with Applications, 41(7):3367–3382, 2014

    Mohd Shahrimie Mohd Asaari, Shahrel A Suandi, and Bakhtiar Affendi Rosdi. Fusion of band limited phase only correlation and width centroid contour distance for finger based biometrics.Expert Systems with Applications, 41(7):3367–3382, 2014

    2014

  48. [48]

    Sdumla-hmt: A multimodal biometric database

    Yilong Yin, Lili Liu, and Xiwei Sun. Sdumla-hmt: A multimodal biometric database. InChinese conference on biometric recognition, pages 260–268. Springer, 2011

    2011

  49. [49]

    Cubuk, Barret Zoph, Justin Gilmer, and Balaji Lakshminarayanan

    Dan Hendrycks, Norman Mu, Ekin D. Cubuk, Barret Zoph, Justin Gilmer, and Balaji Lakshminarayanan. AugMix: A simple data processing method to improve robustness and uncertainty.Proceedings of the International Conference on Learning Representations (ICLR), 2020

    2020

  50. [50]

    Benchmarking neural network robustness to common corruptions and perturbations.Proceedings of the International Conference on Learning Representations, 2019

    Dan Hendrycks and Thomas Dietterich. Benchmarking neural network robustness to common corruptions and perturbations.Proceedings of the International Conference on Learning Representations, 2019

    2019

  51. [51]

    Explaining and harnessing adversarial examples

    Ian J Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adversarial examples. InInternational Conference on Learning Representations, 2015

    2015

  52. [52]

    Towards deep learning models resistant to adversarial attacks

    Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu. Towards deep learning models resistant to adversarial attacks. InInternational Conference on Learning Representations, 2018

    2018

  53. [53]

    SGDR: Stochastic Gradient Descent with Warm Restarts

    Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts.arXiv preprint arXiv:1608.03983, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  54. [54]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. InInternational Conference on Learning Representations, 2019

    2019

  55. [55]

    Manifold mixup: Better representations by interpolating hidden states

    Vikas Verma, Alex Lamb, Christopher Beckham, Amir Najafi, Ioannis Mitliagkas, David Lopez-Paz, and Yoshua Bengio. Manifold mixup: Better representations by interpolating hidden states. InInternational Conference on Machine Learning, pages 6438–6447, 2019

    2019

  56. [56]

    Diffusemix: Label-preserving data augmentation with diffusion models

    Khawar Islam, Muhammad Zaigham Zaheer, Arif Mahmood, and Karthik Nandakumar. Diffusemix: Label-preserving data augmentation with diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 27621–27630, 2024. APPENDIX A. Additional Full Results This appendix reports the more complete experimental tables that...

    work page arXiv 2024