| Title |
Lightweight Dental Image Segmentation with Combined Importance and Redundancy-based Pruning |
| Authors |
현민주(Minju Hyun) ; 송병철(Byung Cheol Song) |
| DOI |
https://doi.org/10.5573/ieie.2025.62.12.89 |
| Keywords |
Image segmentation; Model compression; Filter pruning |
| Abstract |
Deep learning-based dental segmentation models achieve higher accuracy as their architectures become deeper and more complex; however, their application in real-world industrial settings is hindered by high computational and memory demands. To address this issue, pruning techniques have been employed, but existing methods focus solely on filter importance and fail to account for redundancy. In this paper, we propose DDGWD, a method that integrates Manhattan distance-based redundancy evaluation with importance to consider both factors simultaneously. Experiments on a U-Net-based model show that the proposed method reduces computational cost by approximately 9.7% and GPU memory usage by about 4.8% compared to existing importance-based methods, while limiting performance degradation to only 0.04. |