The Journal of Korea Robotics Society (로봇학회논문지)

Korea Robotics Society (한국로봇학회)
권호 표지
DOI QR코드

DOI QR Code

The Robust Weight Conversion Learning for Classification of Occlusion Images

폐색 이미지 분류를 위한 강건한 가중치 전환 학습

  • Received : 2022.10.31
  • Accepted : 2022.12.12
  • Published : 2023.02.28
Download PDF
⟨ Previous Next ⟩

Abstract

An unexpected occlusion in a real life, not in a laboratory, can be more fatal to neural networks than expected. In addition, it is virtually impossible to create a network that learns all the environmental changes as well as occlusions. Therefore, we propose an alternative approach in which the architecture and number of parameters remain unchanged while adapting to occlusion circumstances. Learning method with the term Conversion Learning classifies them more robustly by converting the weights from various occlusion situations. The experiments on MNIST dataset showed a 3.07 [%p] performance improvement over the baseline CNN model in a situation where most objects are occluded and unknowing what occlusion will appear in advance. The experimental results suggest that Conversion Learning is an efficient method to respond to environmental changes such as occluded images.

Keywords

Acknowledgement

This research was supported by the MSIT (Ministry of Science, ICT), Korea, under the High-Potential Individuals Global Training Program) (RS-2022-00155054) supervised by the IITP (Institute for Information & Communications Technology Planning & Evaluation). Following are results of a study on the "Leaders in INdustry-university Cooperation 3.0" Project, supported by the Ministry of Education and National Research Foundation of Korea

References

  1. S. O'Gara and K. McGuinness, "Comparing data augmentation strategies for deep image classification," IMVIP 2019: Irish Machine Vision and Image Processing (IMVIP), Dublin, Ireland, 2019, DOI: 10.21427/148b-ar75.
  2. A. Bochkovskiy, C.-Y. Wang, and H.-Y. M. Liao, "Yolov4: Op-timal speed and accuracy of object detection," Computer Vision and Pattern Recognition, 2020, DOI: 10.48550/arXiv.2004.10934.
  3. Z. Zhong, L. Zheng, G. Kang, S. Li, and Y. Yang, "Random erasing data augmentation," AAAI Conference on Artificial Intelligence, vol. 34, no. 7, 2020, DOI: 10.1609/aaai.v34i07.7000.
  4. T. DeVries and G. W. Taylor, "Improved regularization of convolutional neural networks with cutout," Computer Vision and Pattern Recognition, 2017, DOI: 10.48550/arXiv.1708.04552.
  5. K. K. Singh and Y. J. Lee, "Hide-and-seek: Forcing a network to be meticulous for weakly-supervised object and action localization," International Conference on Computer Vision (ICCV), Venice, Italy, 2017, DOI: 10.1109/ICCV.2017.381.
  6. P. Chen, S. Liu, H. Zhao, and J. Jia, "Gridmask data augmentation," Computer Vision and Pattern Recognition, 2020, DOI: 10.48550/arXiv.2001.04086.
  7. D. Walawalkar, Z. Shen, Z. Liu, and M. Savvides, "Attentive cutmix: An enhanced data augmentation approach for deep learning based image classification," Computer Vision and Pattern Recognition, 2020, DOI: 10.48550/arXiv.2003.13048.
  8. D. Zeng, R. Veldhuis, and L. Spreeuwers, "A survey of face recognition techniques under occlusion," IET biometrics, vol. 10, no. 6, pp. 581-606, 2021, DOI: 10.1049/bme2.12029.
  9. L. Song, D. Gong, Z. Li, C. Liu, and W. Liu, "Occlusion robust face recognition based on mask learning with pairwise differential Siamese network," International Conference on Computer Vision (ICCV), 2019, DOI: 10.1109/ICCV.2019.00086.
  10. F. Cen, X. Zhao, W. Li, and G. Wang, "Deep feature augmentation for occluded image classification," Pattern Recognition, vol. 111, 2021, DOI: 10.1016/j.patcog.2020.107737.
  11. Y. LeCun, C. Cortes, and C. J. C. Burges, "The mnist database of handwritten digits," THE MNIST DATABASE, 1998, [Online], http://yann.lecun.com/exdb/mnist/, Accessed: Jan. 04, 2021.
  12. A. Krizhevsky, "Learning multiple layers of features from tiny images," 2009, [Online], http://www.cs.utoronto.ca/~kriz/learningfeatures-2009-TR.pdf.
  13. D. P. Kingma and J. Ba, "Adam: A method for stochastic optimization," 2014, Machine Learning, DOI: 10.48550/arXiv.1412.6980.
  14. T. Tieleman and G. Hinton, "Lecture 6.5-rmsprop: Divide the gradient by a running average of its recent magnitude," COURSERA: Neural networks for machine learning, vol. 4, no. 2, pp. 26-31, 2012, [Online], https://www.coursera.org/lecture/deep-neural-network/rmsprop-BhJlm/, Accessed: Feb. 20, 2021.