ETRI Journal

Electronics and Telecommunications Research Institute (한국전자통신연구원)
권호 표지
DOI QR코드

DOI QR Code

A novel MobileNet with selective depth multiplier to compromise complexity and accuracy

  • Chan Yung Kim (Department of Electronic and Electrical Engineering, Hongik University) ;
  • Kwi Seob Um (Department of Electronic and Electrical Engineering, Hongik University) ;
  • Seo Weon Heo (Department of Electronic and Electrical Engineering, Hongik University)
  • Received : 2022.03.16
  • Accepted : 2022.08.07
  • Published : 2023.08.10
Download PDF
⟨ Previous Next ⟩

Abstract

In the last few years, convolutional neural networks (CNNs) have demonstrated good performance while solving various computer vision problems. However, since CNNs exhibit high computational complexity, signal processing is performed on the server side. To reduce the computational complexity of CNNs for edge computing, a lightweight algorithm, such as a MobileNet, is proposed. Although MobileNet is lighter than other CNN models, it commonly achieves lower classification accuracy. Hence, to find a balance between complexity and accuracy, additional hyperparameters for adjusting the size of the model have recently been proposed. However, significantly increasing the number of parameters makes models dense and unsuitable for devices with limited computational resources. In this study, we propose a novel MobileNet architecture, in which the number of parameters is adaptively increased according to the importance of feature maps. We show that our proposed network achieves better classification accuracy with fewer parameters than the conventional MobileNet.

Keywords

Acknowledgement

This work was supported by the 2021 Hongik University Research Fund.

References

  1. A. Krizhevsky, I. Sutskever, and G. E. Hinton, Imagenet classification with deep convolutional neural networks, Adv. Neural Inf. Proc. Syst. 25 (2012), 1097-1105.
  2. K. Simonyan and A. Zisserman, Very deep convolutional networks for large-scale image recognition, arXiv preprint, 2014. https://doi.org/10.48550/arXiv.1409.1556
  3. K. He, X. Zhang, S. Ren, and J. Sun, Deep residual learning for image recognition, arXiv preprint, 2015. https://doi.org/10.48550/arXiv.1512.03385
  4. S. Han, J. Pool, J. Tran, and W. J. Dally, Learning Both Weights, and Connections for Efficient Neural Networks, In Advances in Neural Information Processing Systems, NIPS, 2015, 1135-1143.
  5. S. Han, J. Pool, S. Narang, H. Mao, E. Gong, S. Tang, E. Elsen, P. Vajda, M. Paluri, J. Tran, and B. Catanzaro, DSD: Densesparse-dense training for deep neural networks, arXiv preprint, 2017. https://doi.org/10.48550/arXiv.1607.04381
  6. G. Li and G. Xu, Providing clear pruning threshold: A novel CNN pruning method via L0 regularisation, IET Image Process. 15 (2020), no. 2.
  7. C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich, Going deeper with convolutions, arXiv preprint, 2014. https://doi.org/10.48550/arXiv.1409.4842
  8. X. Zhang, X. Zhou, M. Lin, and J. Sun, ShuffleNet: An extremely efficient convolutional neural network for Mobile devices, (Conference on Computer Vision, and Pattern Recognition, Salt Lake City, UT, USA), 2018, pp. 6848-6856.
  9. M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, MobileNetV2: Inverted residuals, and linear bottlenecks, arXiv preprint, 2018. https://doi.org/10.48550/arXiv.1801.04381
  10. A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam, MobileNets: Efficient convolutional neural networks for Mobile vision applications, arXiv preprint, 2017. https://doi.org/10.48550/arXiv.1704.04861
  11. J. T. Springenberg, A. Dosovitskiy, T. Brox, and M. Riedmiller, Striving for simplicity: The all convolutional net, arXiv preprint, 2015. https://doi.org/10.48550/arXiv.1412.6806
  12. F. Chollet, Xception: Deep learning with depthwise separable convolutions, arXiv preprint, 2016. https://doi.org/10.48550/arXiv.1610.02357
  13. Y. Zhou, Y. Liu, G. Han, and Y. Fu, Face recognition based on the improved MobileNet, (IEEE Symposium Series on Computational Intelligence, Xiamen, China), 2019, pp. 2776-2781. https://doi.org/10.1109/SSCI44817.2019.9003100
  14. C. Bi, J. Wang, Y. Duan, B. Fu, J. R. Kang, and Y. Shi, MobileNet based apple leaf diseases identification, Mobile Netw. Appl. 27 (2022), 172-180. https://doi.org/10.1007/s11036-020-01640-1
  15. J. Chen, D. Zhang, and M. Suzauddola, Identification of plant disease images via a squeeze-and-excitation MobileNet model and twice transfer learning, IET Image Process. 15 (2021), no. 1, 23-26.
  16. A. Wibowo, C. A. Hartanto, and P. W. Wirawan, Android skin cancer detection and classification based on MobileNet v2 model, Int. J. Adv. Intell. Inf. 6 (2020), no. 2, 135-148.
  17. V. S. K. Tangudu, J. Kakarla, and I. B. Venkateswarlu, COVID19 detection from chest x-ray using MobileNet and residual separable convolution block, Soft. Comput. 26 (2022), 2197-2208. https://doi.org/10.1007/s00500-021-06579-3
  18. M. H. Firmansyah, S.-J. Koh, and W. K. Dewanto, Light-weight MobileNet for fast detection of COVID-19, Jurnal Teknologi Informasi Dan Terapan, J-TIT 8 (2021), no. 1, 2580-2291.
  19. A. Howard, M. Sandler, G. Chu, L.-C. Chen, B. Chen, M. Tan, W. Wang, Y. Zhu, R. Pang, V. Vasudevan, and Q. V. Le, Searching for MobileNetV3, arXiv preprint, 2019. https://doi.org/10.48550/arXiv.1905.02244
  20. J. Hu, L. Shen, S. Albanie, G. Sun, and E. Wu, Squeeze-andexcitation networks, arXiv preprint, 2017. https://doi.org/10.48550/arXiv.1709.01507
  21. S. Woo, J. Park, J. Lee, and I.S. Kweon, CBAM: Convolutional block attention module, (Proceedings of the European Conference on Computer Vision, Munich, Germany), 2018, pp. 3-19.
  22. C.-H. Tu, J.-H. Lee, Y.-M. Chan, and C.-S. Chen, Pruning depthwise separable convolutions for MobileNet compression, (Proc. International Joint Conference on Neural Netw, Glasgow, UK), 2020, pp. 1-8.
  23. M. Ayi and M. El-Sharkawy, RMNv2: Reduced Mobilenet V2 for CIFAR10, (10th Annual Computing and Communication Workshop and Conference, Las Vegas, NV, USA), 2020. https://doi.org/10.1109/CCWC47524.2020.9031131
  24. P. Singh, V. K. Verma, P. Rai, and V. P. Namboodiri, HetConv: Heterogeneous kernel-based convolutions for deep CNNs, arXiv preprint, 2019. https://doi.org/10.48550/arXiv.1903.04120
  25. V.-T. Hoang and K.-H. Jo, PydMobileNet: Pyramid Depthwise separable convolution networks for image classification, (IEEE 28th International Symposium on Industrial Electronics, Vancouver, Canada), 2019, pp. 1430-1434. https://doi.org/10.1109/ISIE.2019.8781130
  26. N. A. Mohamed, M. A. Zulkifley, and S. R. Abdani, Spatial pyramid pooling with Atrous convolutional for MobileNet, (IEEE student Conference on Research and Development, Batu Pahat, Malaysia), 2020, pp. 333-336. https://doi.org/10.1109/SCOReD50371.2020.9250928
  27. P. S. P. Kavyashree and M. El-Sharkawy, Compressed MobileNet V3:A light weight variant for resource-constrained platforms, (IEEE 11th Annual Computing and Communication Workshop and Conference, NV, USA), 2021. https://doi.org/10.1109/CCWC51732.2021.9376113
  28. S. Bouguezzi, H. Faiedh, and C. Souani, Slim MobileNet: An enhanced deep convolutional neural network, (18th International Multi-Conference on Systems, Signals & Devices, Monastir, Tunisia), 2021. https://doi.org/10.1109/SSD52085.2021.9429519
  29. H.-Y. Chen and C.-Y. Su, An enhanced hybrid MobileNet, (International Conference on Awareness Science, and Technology, Fukuoka, Japan), 2018, pp. 308-312.
  30. D. Sinha and M. El-Sharkawy, Ultra-thin MobileNet, (IEEE Aannual Computing, and Communication Workshop, and Conference, Las Vegas, NV, USA), 2020, pp. 234-240.