| Title |
Dental Image Segmentation Network Quantization via Block-utility Analysis |
| Authors |
김성민(Seong Min Kim) ; 송병철(Byung Cheol Song) |
| DOI |
https://doi.org/10.5573/ieie.2026.63.1.45 |
| Keywords |
Medical image segmentation; Quantization; Deep neural network compression |
| Abstract |
Deep learning?based dental image segmentation algorithms contribute to improved diagnostic efficiency and accuracy. However, hybrid networks that combine CNNs and Transformers incur high computational costs, which constrain their applicability in real-world scenarios. Among lightweight approaches to overcome this limitation, quantization has drawn considerable attention. Nevertheless, insufficient consideration of block utility may lead to unnecessary quantization complexity or limitations in performance preservation. In this paper, we analyze block utility based on performance contribution and quantization sensitivity, and through a quantization strategy that incorporates these factors, we quantize hybrid dental image segmentation networks while minimizing performance degradation. |