TY  - JOUR
T1  - Mask-Based Selective Downsampling in Convolutional Neural Networks
AU  - Kwak, Chulyoung 
AU  - Bahk, Saewoong 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2025
DA  - 2025/1/1
DO  - 10.7840/kics.2025.50.5.722
KW  - Down sampling
KW  - Convolutional Neural network
KW  - Signal Processing
AB  - This letter presents a study on selective down-sampling techniques designed to enhance the efficiency of convolutional neural networks (CNNs).
By selectively adjusting the resolution of feature layers based on the down-sampling mask, this method aims to improve efficiency. We develop a network that is 80% lighter than the baseline scheme, and introduce several methods to mitigate the performance degradation of light weight networks. The performance of the proposed methods is evaluated through extensive experiments.