Yixu Feng

CVPR 2025

HVI: A New Color Space for
Low-light Image Enhancement

Qingsen Yan*Yixu Feng*Cheng ZhangGuansong PangKangbiao ShiPeng WuWei DongJinqiu SunYanning Zhang

* Equal contribution · Yixu Feng is the student first author

1 Northwestern Polytechnical University2 Singapore Management University3 Xi'an University of Architecture and Technology

A color space designed for low-light vision—separating chromaticity from intensity to suppress red discontinuity and black-plane noise.

Paper arXiv Code Live Demo

01 · Motivation

Rethinking the color space

Standard sRGB couples brightness and color, while HSV introduces red discontinuity and black-plane noise. HVI reshapes the hue-saturation plane and learns an intensity collapse function for stable enhancement.

Comparison of noise in sRGB and HSV with the proposed HVI color space
HVI reduces the distance between perceptually similar red coordinates and compresses unstable low-intensity regions.

02 · Overview

Color and Intensity Decoupling Network

CIDNet processes the HV chromatic plane and the intensity axis with two dedicated branches, then exchanges complementary information through Lighten Cross-Attention blocks.

HVI-CIDNet pipeline
End-to-end pipeline: sRGB → HVI transformation → dual-branch CIDNet → perceptual inverse HVI transformation.

LIGHTEN CROSS-ATTENTION

Separate, exchange, reconstruct.

The intensity branch focuses on illumination recovery; the HV branch removes chromatic noise. Cross-attention lets each branch borrow only the information it needs.

Lighten Cross-Attention block structure
LCA block structure.

03 · Results

Consistent enhancement across benchmarks

HVI-CIDNet surpasses prior low-light enhancement methods across ten datasets while remaining lightweight and robust to severe noise, color shifts, and non-uniform illumination.

Visual comparisons on LOL-v1, LOL-v2-real and LOL-v2-synthetic
Visual comparison on paired LOL benchmarks. HVI-CIDNet restores natural brightness and color while preserving fine details.
10datasets evaluated
28.39 dBLOLv2-Real PSNR with GT mean
29.57 dBLOLv2-Synthetic PSNR with GT mean
3.13average NIQE on five unpaired sets

04 · Abstract

Designed for low-light vision

Low-Light Image Enhancement (LLIE) aims to recover detailed visual information from corrupted low-light images. Existing sRGB-based methods often produce color bias and brightness artifacts because brightness and chromaticity remain coupled, while HSV-based alternatives introduce red and black noise artifacts.

We propose Horizontal/Vertical-Intensity (HVI), a new color space defined by polarized HS maps and learnable intensity. Polarization closes the red discontinuity, and the learnable intensity function compresses unstable low-light regions. A Color and Intensity Decoupling Network (CIDNet) then learns accurate photometric mappings in HVI space. Comprehensive benchmark and ablation studies show that HVI with CIDNet outperforms state-of-the-art methods across ten datasets.

05 · Citation

BibTeX

@InProceedings{Yan_2025_CVPR,
  author    = {Yan, Qingsen and Feng, Yixu and Zhang, Cheng and
               Pang, Guansong and Shi, Kangbiao and Wu, Peng and
               Dong, Wei and Sun, Jinqiu and Zhang, Yanning},
  title     = {HVI: A New Color Space for Low-light Image Enhancement},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages     = {5678--5687},
  year      = {2025}
}