Abstract
This paper proposes a texture-based domain-specific data augmentation technique applicable when training on small datasets for deep learning classification tasks. Our method focuses on label-preservation to improve generalization and optimization robustness over data-dependent augmentation methods using textures. We generate a small perturbation in an image based on a randomly sampled texture image. The textures we use are naturally occurring and domain-independent of the training dataset: regular, near regular, irregular, near stochastic and stochastic classes. Our method uses the textures to apply sparse, patterned occlusion to images and a penalty regularization term during training to help ensure label preservation. We evaluate our method against the competitive soft-label Mixup and RICAP data augmentation methods with the ResNet-50 architecture using the unambiguous “Bird or Bicyle” and Oxford-IIT-Pet datasets, as well as a random sampling of the Open Images dataset. We experimentally validate the importance of label-preservation and improved generalization by using out-of-distribution examples and show that our method improves over competitive methods.
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
Notes
- 1.
pytorch.org.
- 2.
numpy.org.
References
Bergstra, J., et al.: Making a science of model search: Hyperparameter optimization in hundreds of dimensions for vision architectures. In: ICML, pp. 115–23. PMLR (2013)
Brown, T.B., et al.: Unrestricted adversarial examples. arXiv preprint arXiv:1809.08352 (2018)
Cimpoi, M., et al.: Describing textures in the wild. In: CVPR, pp. 3606–13 (2014)
DeVries, T., Taylor, G.W.: Improved regularization of convolutional neural networks with cutout. arXiv preprint arXiv:1708.04552 (2017)
Guo, H., et al.: MixUp as locally linear out-of-manifold regularization. In: AAAI, vol. 33, pp. 3714–22 (2019)
He, K., et al.: Deep residual learning for image recognition. In: CVPR, pp. 770–8 (2016)
Hernández-García, A., König, P.: Data augmentation instead of explicit regularization. arXiv preprint arXiv:1806.03852 (2018)
Hinton, G., et al.: Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531 (2015)
Krizhevsky, A.: Learning multiple layers of features from tiny images. Master’s thesis, University of Tront (2009)
Kuznetsova, A., et al.: The open images dataset V4: unified image classification, object detection, and visual relationship detection at scale. Int. J. Comput. Vis. 128(7), 1956–1981 (2020)
Kylberg, G.: Kylberg texture dataset v. 1.0. Centre for Image Analysis, Swedish University of Agricultural Sciences and Uppsala University (2011)
Lazebnik, S., et al.: A sparse texture representation using local affine regions. IEEE Trans. Pattern Anal. Mach. Intell. 27(8), 1265–78 (2005)
Nesterov, Y.: A method of solving a convex programming problem with convergence rate O (1/k 2) O (1/k2). In: Soviet Mathematics. Doklady, vol. 27, no. 2
Parkhi, O.M., et al.: Cats and dogs. In: CVPR, pp. 3498–3505. IEEE (2012)
Rahaman, N., et al.: On the spectral bias of neural networks. In: ICML, pp. 5301–10. PMLR (2019)
Rolnick, D., et al.: Deep learning is robust to massive label noise. arXiv preprint arXiv:1705.10694 (2017)
Rudin, L.I., et al.: Nonlinear total variation based noise removal algorithms. Phys. D: Nonlinear Phenom. 60(1–4), 259–268 (1992)
Shafahi, A., et al.: Are adversarial examples inevitable? In: ICLR (2019)
Sharif, M., et al.: On the suitability of Lp-norms for creating and preventing adversarial examples. In: CVPRW, pp. 1605–13 (2018)
Summers, C., Dinneen, M.J.: Improved mixed-example data augmentation. In: WACV, pp. 1262–70. IEEE (2019)
Summers, C., Dinneen, M.J.: Nondeterminism and instability in neural network optimization. In: ICML, pp. 9913–22. PMLR (2021)
Takahashi, R., et al.: Data augmentation using random image cropping and patching for deep CNNs. IEEE Trans. Circuits Syst. Video Technol. 30(9), 2917–31 (2019)
Tomasi, C., Manduchi, R.: Bilateral filtering for gray and color images. In: ICCV, pp. 839–846. IEEE (1998)
Zagoruyko, S., Komodakis, N.: Wide residual networks. In: BMVC. BMVA (2016)
Zhang, C., et al.: Understanding deep learning (still) requires rethinking generalization. Commun. ACM 64(3), 107–15 (2021)
Zhang, H., et al.: MixUp: beyond empirical risk minimization. arXiv preprint arXiv:1710.09412 (2017)
Zhong, Z., et al.: Random erasing data augmentation. In: AAAI, vol. 34, pp. 13001–13008 (2020)
Zoph, B., et al.: Rethinking pre-training and self-training. In: NeurIPS, pp. 3833–3845. ACM (2020)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2023 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Dash, A., Albu, A.B. (2023). Texture-Based Data Augmentation for Small Datasets. In: Blanc-Talon, J., Delmas, P., Philips, W., Scheunders, P. (eds) Advanced Concepts for Intelligent Vision Systems. ACIVS 2023. Lecture Notes in Computer Science, vol 14124. Springer, Cham. https://doi.org/10.1007/978-3-031-45382-3_29
Download citation
DOI: https://doi.org/10.1007/978-3-031-45382-3_29
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-45381-6
Online ISBN: 978-3-031-45382-3
eBook Packages: Computer ScienceComputer Science (R0)