QuantDemoire: Quantization with Outlier Aware for Image Demoiréing

TL;DR

This work tackles efficient on-device image demoiréing by introducing QuantDemoire, a post-training quantization framework tailored for demoiréing models. It combines an outlier-aware quantizer, which uses sampling-based activation range estimation and FP16 preservation of a small fraction of weight outliers, with a frequency-aware calibration strategy that concentrates on mid- and low-frequency components to mitigate banding. Across UHDM, FHDMi, and LCDMoiré, QuantDemoire delivers substantial parameter and FLOP reductions (over 86% at 4-bit) with only about a 4.7% PSNR drop, and it exceeds state-of-the-art quantization approaches by more than 4 dB on UHDM. The approach enables practical, edge-device demoiréing while maintaining high restoration quality and provides a generalizable technique for frequency-aware PTQ in low-bit vision models.

Abstract

Demoiréing aims to remove moiré artifacts that often occur in images. While recent deep learning-based methods have achieved promising results, they typically require substantial computational resources, limiting their deployment on edge devices. Model quantization offers a compelling solution. However, directly applying existing quantization methods to demoiréing models introduces severe performance degradation. The main reasons are distribution outliers and weakened representations in smooth regions. To address these issues, we propose QuantDemoire, a post-training quantization framework tailored to demoiréing. It contains two key components. **First}, we introduce an outlier-aware quantizer to reduce errors from outliers. It uses sampling-based range estimation to reduce activation outliers, and keeps a few extreme weights in FP16 with negligible cost. **Second**, we design a frequency-aware calibration strategy. It emphasizes low- and mid-frequency components during fine-tuning, which mitigates banding artifacts caused by low-bit quantization. Extensive experiments validate that our QuantDemoire achieves large reductions in parameters and computation while maintaining quality. Meanwhile, it outperforms existing quantization methods by over **4 dB** on W4A4. Code is released at: https://github.com/zhengchen1999/QuantDemoire.

Figures (6)

• Figure 1: Comparison with recent quantization methods on UHDM yu2022towards. The full-precision backbone is ESDNet yu2022towards. Left: PSNR performance at different bit-widths. Right: visual comparison. Our QuantDemoire demonstrates superior efficiency and performance.
• Figure 2: The overview of our QuantDemoire. (a) Outlier-Aware Quantizer: activations discard outliers through sample-based estimation, while weights preserve outliers in FP16 precision. (b) Frequency-Aware Calibration Strategy: quantizer parameters are optimized on mid- and low-frequency components, which are extracted through the frequency extraction process.
• Figure 3: Activation distributions (Original, Smooth (Raw), Smooth (Sample)). (a) Original: ranges differ across channels; (b) Smooth (Raw): align maxima only; (c) Smooth (Sample): align the main body (excluding outliers), improving suitability for quantization.
• Figure 4: Visualization of activation and weight distributions from randomly selected layers in ESDNet yu2022towards. The distributions are approximately Gaussian or exponential.
• Figure 5: Visualization of results under different calibration strategies. Original: training in the spatial domain suffers from banding artifacts under low-bit quantization. Mid&Low-Freq: our frequency-aware calibration strategy leverages mid- and low-frequency to mitigate banding.