Low-Light Image Enhancement aims to restore images captured under poor il- lumination conditions. Previous methods that apply uniform correction across the entire image fail to adjust local illumination variations, leading to noise amplification in dark re- gions and oversaturation in bright regions. To address these limitations, we propose Spatial Modulation Module (SMM) based on the Horizontal/Vertical-Intensity (HVI) color space. SMM captures contextual information and dynamically generates saturation and brightness correction maps in a pixel-wise manner. We introduce Intensity Mean Loss designed to adap- tively balance brightness alignment and detail restoration based on distribution divergence. This loss guides the network to focus on structural detail restoration. We also employ Inter- mediate Supervision Loss to resolve the scale ambiguity problem in SMM, ensuring training stability. Experimental results on diverse datasets demonstrated that HVI-SMM achieved superior performance improvements, validating the effectiveness of pixel-wise optimization for sophisticated local restoration. The code is publicly available.
LEE et al. (Thu,) studied this question.