Light-CAM: A Lightweight Model for Weakly Supervised Object Localization of Embedded Devices 


Vol. 47,  No. 8, pp. 1144-1152, Aug.  2022
10.7840/kics.2022.47.8.1144


PDF
  Abstract

With the continuous development of hardware performance, many applications that utilize deep learning in mobile and embedded devices have emerged, even if they are not high-performance PCs. However, despite advances in hardware performance, there are limitations in using heavy models with a large number of parameters on mobile and embedded devices. In this paper, we design a network to be used for embedded devices when performing weak supervised object localization using a class activation map. The proposed model, Light-CAM, designs the layer of the Class Activation Map network shallowly, reducing the number of parameters of the model and minimizing the reduction in localization performance. Experiments show that the Light-CAM+BR-AvgCAM combination showed the third-highest performance when comparing localization accuracy using multiple models and multiple CAM methods on Bird and Dog datasets. Compared to the VGG16+BR-AvgCAM combination with the highest performance, the localization accuracy is 5.9% lower, but the number of parameters is reduced by 9.34 times. It can be seen that the proposed Light-CAM model is a suitable model for small embedded devices with minimal computing.

  Statistics
Cumulative Counts from November, 2022
Multiple requests among the same browser session are counted as one view. If you mouse over a chart, the values of data points will be shown.


  Cite this article

[IEEE Style]

Y. Kim, J. Kim, H. Park, "Light-CAM: A Lightweight Model for Weakly Supervised Object Localization of Embedded Devices," The Journal of Korean Institute of Communications and Information Sciences, vol. 47, no. 8, pp. 1144-1152, 2022. DOI: 10.7840/kics.2022.47.8.1144.

[ACM Style]

Yongho Kim, Jiha Kim, and Hyunhee Park. 2022. Light-CAM: A Lightweight Model for Weakly Supervised Object Localization of Embedded Devices. The Journal of Korean Institute of Communications and Information Sciences, 47, 8, (2022), 1144-1152. DOI: 10.7840/kics.2022.47.8.1144.

[KICS Style]

Yongho Kim, Jiha Kim, Hyunhee Park, "Light-CAM: A Lightweight Model for Weakly Supervised Object Localization of Embedded Devices," The Journal of Korean Institute of Communications and Information Sciences, vol. 47, no. 8, pp. 1144-1152, 8. 2022. (https://doi.org/10.7840/kics.2022.47.8.1144)

Search
(IN TITLE, AUTHOR, ABSTRACT,KEYWORDS)

Advanced Search ▼

POPULAR KEYWORDS
(TOP 10 KEYWORDS)

Recent Publications
(LAST 3 YEARS)






pISSN : 1226 - 4717
eISSN : 2287 - 3880