Design of Depth Adjustable Neural Networks for Vision Tasks 


Vol. 14,  No. 8, pp. 627-632, Aug.  2025
https://doi.org/10.3745/TKIPS.2025.14.8.627


PDF
  Abstract

In this paper, we propose the architecture and training method of a depth adjustable deep neural network that enables on-the-fly selection of different inference accuracy-efficiency trade-offs using a single trained model. The proposed depth adjustable network splits each residual stage into two parts: one part learns hierarchical representations, while the other refines the learned features using a self-distillation technique. This training strategy allows the model to jointly learn multiple embedded subnetworks of varying depths with at most twice the training time of a single network. The proposed depth adjustable model is applied to image classification and object detection tasks, demonstrating the ability to adjust the trade-off between accuracy and efficiency without sacrificing maximum accuracy

  Statistics
  Cite this article

[IEEE Style]

K. Woochul, "Design of Depth Adjustable Neural Networks for Vision Tasks," The Transactions of the Korea Information Processing Society, vol. 14, no. 8, pp. 627-632, 2025. DOI: https://doi.org/10.3745/TKIPS.2025.14.8.627.

[ACM Style]

Kang Woochul. 2025. Design of Depth Adjustable Neural Networks for Vision Tasks. The Transactions of the Korea Information Processing Society, 14, 8, (2025), 627-632. DOI: https://doi.org/10.3745/TKIPS.2025.14.8.627.

Forms

Search
(IN TITLE, AUTHOR, ABSTRACT,KEYWORDS)

Advanced Search ▼

POPULAR KEYWORDS
(TOP 10 KEYWORDS)

Recent Publications
(LAST 3 YEARS)

Old Journals

Indexing

All publications of TKIPS is indexed in DOI, EBSCO, Google Scholar, Crossref, and CrossCheck.

TKIPS is also selected as the Journal for Accreditation by NRF (National Research Foundation of Korea).

Related Journals