Image Denoising Meets Quantization: Exploring the Effects of Post-Training Quantization

초록

The field of deep learning-based image denoising has seen significant progress, with various advanced architectures achieving state-of-the-art performance. However, improved performance is often accompanied by an increased model complexity and higher computational costs. These constraints present significant challenges for the deployment in resource-constrained environments such as edge devices. To address this issue, we explore the effects of post-training quantization (PTQ) on image denoising models. By applying existing PTQ techniques, model complexity can be significantly reduced while maintaining competitive denoising performance. We conduct a comparative analysis of different PTQ methods to evaluate their impact on denoising quality. In particular, we implement these techniques on the NAFNet architecture and assess their efficacy on SIDD, thereby analyzing their impact on denoising performance. © 2025 IEEE.

키워드