Lightweighting of Super-Resolution Model Using Depth-Wise Separable Convolution 


Vol. 46,  No. 4, pp. 591-597, Apr.  2021
10.7840/kics.2021.46.4.591


PDF Full-Text
  Abstract

Super-resolution is the process of upscaling low-resolution images into high-resolution images and is relatively focused on research that improves system performance by utilizing heavy computation. We recognized that deep convolutional neural network based super-resolution models need to be lightweighted when they ared used with other technologies or in mobile environments. In this paper, we propose a separable convolution-based multi-scale residual network(SMSRN) structure using depth-wise separable convolution, which is a lightweight version of the latest super-resolution model, that is, the multi-scale residual network (MSRN). Compare to the MSRN, the number of parameters in SMSRN is 14.64% of that in MSRN. On the other hand, quantitative experiments on a variety of benchmark datasets resulted in the remaining performance of 98.53%, and qualitative experiments showed that performance degradation was difficult to identify. Since the SMSRN is a structure that uses a variety of convolution filter sizes, it is expected that it can also be applied to various architectures of super-resolution models to make them lighter.

  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]

D. Kim, J. Kang, J. Lee, "Lightweighting of Super-Resolution Model Using Depth-Wise Separable Convolution," The Journal of Korean Institute of Communications and Information Sciences, vol. 46, no. 4, pp. 591-597, 2021. DOI: 10.7840/kics.2021.46.4.591.

[ACM Style]

Daehee Kim, Juhee Kang, and Jaekoo Lee. 2021. Lightweighting of Super-Resolution Model Using Depth-Wise Separable Convolution. The Journal of Korean Institute of Communications and Information Sciences, 46, 4, (2021), 591-597. DOI: 10.7840/kics.2021.46.4.591.

[KICS Style]

Daehee Kim, Juhee Kang, Jaekoo Lee, "Lightweighting of Super-Resolution Model Using Depth-Wise Separable Convolution," The Journal of Korean Institute of Communications and Information Sciences, vol. 46, no. 4, pp. 591-597, 4. 2021. (https://doi.org/10.7840/kics.2021.46.4.591)

Search
(IN TITLE, AUTHOR, ABSTRACT,KEYWORDS)

Advanced Search ▼

POPULAR KEYWORDS
(TOP 10 KEYWORDS)

Recent Publications
(LAST 3 YEARS)






pISSN : 1226 - 4717
eISSN : 2287 - 3880