| 30 | 0 | 111 |
| 下载次数 | 被引频次 | 阅读次数 |
在单源域泛化中，往往面临着训练数据单一、缺少新领域数据的问题。现有的单源域泛化模型，大部分是在训练过程中引入数据增强技术来生成多样化的训练样本，以提升模型在未知领域的泛化能力。然而，这样的数据增强技术有时会失效。因此，提出一种基于深度学习的单源域泛化方法，该方法不仅引入数据增强技术，而且对网络结构进行改进。首先，该方法在数据增强阶段引入了风格归一化和恢复模块，以生成更为丰富多样的训练数据；其次，采用多尺度特征提取与融合技术，提取涵盖不同视野域的多尺度信息；最后，采取联合注意力特征融合策略，以增强特征提取过程中通道与空间位置之间的相互依赖性。实验结果表明，本文方法在单源域泛化领域中均优于现有方法。
Abstract：In single-source domain generalization, we often face the problem of single training data and lack of new domain data. Most of the existing single-source domain generalization models introduce data augmentation techniques to generate diverse training samples during the training process in order to enhance the model's generalization ability in unknown domains. However, such data enhancement techniques sometimes fail. Therefore, we propose a single-source domain generalization method based on deep learning, which not only introduces data enhancement techniques but also improves the network structure. Firstly, the method introduces style normalization and recovery modules in the data enhancement phase to generate richer and more diverse training data. Secondly, multi-scale information covering different field of view domains is extracted by multi-scale feature extraction and fusion techniques. Finally, a joint-attention feature fusion strategy is adopted to enhance the inter-dependence between channels and spatial locations in the feature extraction process. The experimental results show that our method outperforms all current methods in single-source domain generalization domains.
[1] Zhou K,Liu Z,Qiao Y,et al.Domain generalization:A survey[J].IEEE transactions on pattern analysis and machine intelligence,2022,45(4):4396-4415.
[2] Ni H,Li Y,Shen H,et al.Part-aware transformer for generalizable person re-identification[J].in 2023 IEEE/CVF International Conference on Computer Vision (ICCV),2023:11246-11255.
[3] Cheng S,Gokhale T,Yang Y.Adversarial Bayesian augmentation for single-source domain generalization[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision.2023:11400-11410.
[4] Hendrycks D,Mu N,Cubuk ED,et al.AugMix:A Simple Data Processing Method to Improve Robustness and Uncertainty[EB/OL].arXiv preprint arXiv:1912.02781.(2019-12-05) [2024-12-31].https://arxiv.org/abs/1912.02781.
[5] Xu Z,Liu D,Yang J,et al.Robust and Generalizable Visual Representation Learning via Random Convolutions[EB/OL].arXiv preprint arXiv:2007.13003.(2020-07-25) [2024-12-31].https://arxiv.org/abs/2007.13003.
[6] Carlucci F M,D’Innocente A,Bucci S,et al.Domain generalization by solving jigsaw puzzles[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.2019:2229-2238.
[7] Qiao F,Zhao L,Peng X.Learning to learn single domain generalization[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.2020:12556-12565.
[8] Li D,Yang Y,Song Y Z,et al.Learning to generalize:Meta-learning for domain generalization[C]//Proceedings of the AAAI conference on artificial intelligence.2018,32(1):3490-3497.
[9] Wan C,Shen X,Zhang Y,et al.Meta convolutional neural networks for single domain generalization[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2022:4682-4691.
[10] Gokhale T,Anirudh R,Thiagarajan J J,et al.Improving diversity with adversarially learned transformations for domain generalization[C] //Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision.2023:434-443.
[11] He K,Zhang X,Ren S,et al.Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.2016:770-778.
[12] Muandet K,Balduzzi D,Sch?lkopf B.Domain generalization via invariant feature representation[C]//International conference on machine learning.PMLR,2013:10-18.
[13] Pan X,Luo P,Shi J,et al.Two at once:Enhancing learning and generalization capacities via ibn-net[C]//Proceedings of the European conference on computer vision (ECCV).2018:464-479.
[14] Huang X,Belongie S.Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization[C].//16th IEEE International Conference on Computer Vision (ICCV).2017:1510-1519.
[15] Jin X,Lan C,Zeng W,et al.Style normalization and restitution for generalizable person re-identification[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.2020:3143-3152.
[16] 于明，李学博，郭迎春.融合注意力机制的域泛化行人再识别[J].控制与决策，2022,37(7):1721-1728.
[17] Zhou K,Yang Y,Cavallaro A,et al.Omni-scale feature learning for person re-identification[C]//Proceedings of the IEEE/CVF international conference on computer vision.2019:3702-3712.
[18] Zhang R,Li J,Sun H,et al.Scan:Self-and-collaborative attention network for video person re-identification[J].IEEE Transactions on Image Processing,2019,28(10):4870-4882.
[19] Xia H,Ma J,Ou J,et al.Pedestrian detection algorithm based on multi-scale feature extraction and attention feature fusion[J].Digital Signal Processing,2022,121:103311.
[20] Volpi R,Namkoong H,Sener O,et al.Generalizing to unseen domains via adversarial data augmentation [C].//32nd Conference on Neural Information Processing Systems (NIPS).2018:5334-5344.
[21] Wang H,He Z,Lipton ZC,Xing EP.Learning Robust Representations by Projecting Superficial Statistics Out [EB/OL].arXiv preprint arXiv:1903.06256.(2019-03-02) [2024-12-31].https://arxiv.org/abs/1903.06256.
[22] Shen W B,Xu D,Zhu Y,et al.Situational fusion of visual representation for visual navigation[C]//Proceedings of the IEEE/CVF international conference on computer vision.2019:2881-2890.
[23] Li D,Yang Y,Song Y Z,et al.Deeper,broader and artier domain generalization[C]//Proceedings of the IEEE international conference on computer vision.2017:5542-5550.
[24] Nam H,Lee H J,Park J,et al.Reducing domain gap by reducing style bias[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2021:8690-8699.
[25] Cheng S,Gokhale T,Yang Y.Adversarial Bayesian augmentation for single-source domain generalization[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision.2023:11400-11410.
[26] Xu Q,Zhang R,Wu Y Y,et al.Simde:A simple domain expansion approach for single-source domain generalization[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2023:4798-4808.
基本信息:
DOI：10.20064/j.cnki.2095-347X.2025.05.002
中图分类号:TP18
引用信息:
[1]郝宇航,梁建安,王莹,等.基于深度学习的单源域泛化方法研究[J].网络新媒体技术,2025,14(05):12-20.DOI:10.20064/j.cnki.2095-347X.2025.05.002.
基金信息:
科技部重点研发计划(编号：2024YFC2208000); 基金委重点项目(编号：62035015); 山西省重点研发项目(编号：202102150101003)