Abstract
Due to the domain shift problem, the deep learning models trained on one domain often fail to generalize well on others. Researchers formulated such a realistic-yet-challenging scenario as a new research line, termed single domain generalization (single-DG), which aims to generalize a model trained on single source domain to multiple target domains. The existing single-DG approaches tried to address the problem by generating diverse samples using extra trainable network modules. However, due to the limited amount of medical data, the extra network parameters are difficult to train. The generated samples are often failed to achieve satisfactory effect for improving model generalization. In this paper, we propose a simple-yet-effective Fourier-based approach, which augments data via spontaneous Amplitude SPECTrum diverSification (ASPECTS), for single domain generalization. Concretely, the proposed approach first converts the image into frequency domain using the Fourier transform, and then spontaneously generates diverse samples by editing the amplitude spectrum using a pool of randomization operations. The proposed approach is established upon the assumption that the high-level semantic information (domain-invariant) is embedded in the phase spectrum of images after Fourier transform, while the amplitude spectrum mainly contains the domain-variant information. We evaluate the proposed ASPECTS approach on both publicly available and private multi-domain datasets. Compared to the existing single-DG approaches, our method is much easier to implement (i.e., without training of extra network modules) and yields the superior improvement.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
Notes
- 1.
Note that the hyperparameter \(\beta \) varies in the interval (0, 1) during the network training to increase the diversity of adversarial samples.
- 2.
We notice that there are other ways to fuse the two spectra, e.g., pixel-wise multiplication (\(\mathcal {A}_c^r \times \mathcal {A}_c^s\)) or series concatenation (\(\mathcal {T}_s(\mathcal {A}_c^r)\)). However, these fusion approaches are observed to under-randomize the amplitude spectrum and degrade the diversity of adversarial samples. Hence, we adopt pixel-wise summation in this study.
- 3.
Note that the same data augmentation operations are adopted to train benchmarking algorithms and our ASPECTS.
- 4.
Consistent to [14], DG approaches can access multiple source domains for training.
- 5.
Note that L2D needs to integrate an extra style-complement module into the framework for sample diversification.
References
Aubreville, M., Bertram, C.A., Donovan, T.A., Marzahl, C., Maier, A., Klopfleisch, R.: A completely annotated whole slide image dataset of canine breast cancer to aid human breast cancer research. Sci. Data 7(1), 1–10 (2020)
Bian, C., et al.: Uncertainty-aware domain alignment for anatomical structure segmentation. Med. Image Anal. 64, 101732 (2020)
Du, Y., et al.: Learning to learn with variational information bottleneck for domain generalization. In: Vedaldi, A., Bischof, H., Brox, T., Frahm, J.-M. (eds.) ECCV 2020. LNCS, vol. 12355, pp. 200–216. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-58607-2_12
Hansen, B.C., Hess, R.F.: Structural sparseness and spatial phase alignment in natural scenes. J. Opt. Soc. Am. A 24(7), 1873–1885 (2007)
Hendrycks, D., Dietterich, T.: Benchmarking neural network robustness to common corruptions and perturbations. arXiv preprint arXiv:1903.12261 (2009)
Huang, Z., Wang, H., Xing, E.P., Huang, D.: Self-challenging improves cross-domain generalization. In: Vedaldi, A., Bischof, H., Brox, T., Frahm, J.-M. (eds.) ECCV 2020. LNCS, vol. 12347, pp. 124–140. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-58536-5_8
Krizhevsky, A., Hinton, G., et al.: Learning multiple layers of features from tiny images (2009)
Motiian, S., Piccirilli, M., Adjeroh, D.A., Doretto, G.: Unified deep supervised domain adaptation and generalization. In: IEEE International Conference on Computer Vision (2017)
Oppenheim, A., Lim, J.: The importance of phase in signals. Proc. IEEE 69(5), 529–541 (1981)
Piotrowski, L.N., Campbell, F.W.: A demonstration of the visual importance and flexibility of spatial-frequency amplitude and phase. Perception 11(3), 337–346 (1982)
Qiao, F., Zhao, L., Peng, X.: Learning to learn single domain generalization. In: IEEE Conference on Computer Vision and Pattern Recognition (2020)
Simon, C., Koniusz, P., Nock, R., Harandi, M.: Adaptive subspaces for few-shot learning. In: IEEE Conference on Computer Vision and Pattern Recognition (2020)
Volpi, R., Namkoong, H., Sener, O., Duchi, J.C., Murino, V., Savarese, S.: Generalizing to unseen domains via adversarial data augmentation. In: Advances in Neural Information Processing Systems, vol. 31 (2018)
Wang, Z., Luo, Y., Qiu, R., Huang, Z., Baktashmotlagh, M.: Learning to diversify for single domain generalization. In: IEEE International Conference on Computer Vision (2021)
Xie, X., Chen, J., Li, Y., Shen, L., Ma, K., Zheng, Y.: MI\(^2\)GAN: generative adversarial network for medical image domain adaptation using mutual information constraint. In: Martel, A.L., et al. (eds.) MICCAI 2020. LNCS, vol. 12262, pp. 516–525. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-59713-9_50
Xu, Q., Zhang, R., Zhang, Y., Wang, Y., Tian, Q.: A Fourier-based framework for domain generalization. In: IEEE Conference on Computer Vision and Pattern Recognition (2021)
Xu, X., Zhou, X., Venkatesan, R., Swaminathan, G., Majumder, O.: d-SNE: domain adaptation using stochastic neighborhood embedding. In: IEEE Conference on Computer Vision and Pattern Recognition (2019)
Xue, Y., Feng, S., Zhang, Y., Zhang, X., Wang, Y.: Dual-task self-supervision for cross-modality domain adaptation. In: Martel, A.L., et al. (eds.) MICCAI 2020. LNCS, vol. 12261, pp. 408–417. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-59710-8_40
Yang, Y., Soatto, S.: FDA: Fourier domain adaptation for semantic segmentation. In: IEEE Conference on Computer Vision and Pattern Recognition (2020)
Zagoruyko, S., Komodakis, N.: Wide residual networks. arXiv preprint arXiv:1605.07146 (2017)
Zhang, T., et al.: Noise adaptation generative adversarial network for medical image analysis. IEEE Trans. Med. Imaging 39(4), 1149–1159 (2020)
Zhao, L., Liu, T., Peng, X., Metaxas, D.: Maximum-entropy adversarial data augmentation for improved generalization and robustness. Adv. Neural. Inf. Process. Syst. 33, 14435–14447 (2020)
Zhou, K., Yang, Y., Hospedales, T., Xiang, T.: Learning to generate novel domains for domain generalization. In: Vedaldi, A., Bischof, H., Brox, T., Frahm, J.-M. (eds.) ECCV 2020. LNCS, vol. 12361, pp. 561–578. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-58517-4_33
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Li, Y., He, N., Huang, Y. (2022). Single Domain Generalization via Spontaneous Amplitude Spectrum Diversification. In: Xu, X., Li, X., Mahapatra, D., Cheng, L., Petitjean, C., Fu, H. (eds) Resource-Efficient Medical Image Analysis. REMIA 2022. Lecture Notes in Computer Science, vol 13543. Springer, Cham. https://doi.org/10.1007/978-3-031-16876-5_4
Download citation
DOI: https://doi.org/10.1007/978-3-031-16876-5_4
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-16875-8
Online ISBN: 978-3-031-16876-5
eBook Packages: Computer ScienceComputer Science (R0)
