pith. sign in

arxiv: 2604.09321 · v1 · submitted 2026-04-10 · 📡 eess.IV · cs.CV

UHD Low-Light Image Enhancement via Real-Time Enhancement Methods with Clifford Information Fusion

Xiaohan Wang , Chen Wu , Dawei Zhao , Guangwei Gao , Dianjie Lu , Guijuan Zhang , Linwei Fan , Xu Lu
show 3 more authors
Shuai Wu Hang Wei Zhuoran Zheng
This is my paper

Pith reviewed 2026-05-10 16:44 UTC · model grok-4.3

classification 📡 eess.IV cs.CV
keywords low-light image enhancementUHD restorationClifford algebrareal-time processingfeature fusionRetinex theorylightweight neural networkfrequency decomposition
0
0 comments X

The pith

Clifford algebra fusion lets a lightweight network enhance 4K and 8K low-light images in milliseconds on ordinary hardware.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a network that restores ultra-high-definition low-light images fast enough for real-time use on consumer devices. It builds a simple four-layer pyramid that splits the input into frequency bands with Gaussian blur and extracts features through a compact U-Net that relies on depthwise separable convolutions. Traditional mixing of those bands often erases structure or adds artifacts, so the method instead converts the feature tensors into multivectors from Clifford algebra in 2D space and combines them according to geometric similarity. This step keeps textures sharp while reducing noise. The network then produces adaptive Gamma and Gain maps that apply Retinex-based brightness correction to produce natural-looking results.

Core claim

By mapping extracted features to a multivector space consisting of scalars, vectors, and bivectors and aggregating them with Clifford similarity, the network overcomes the structural loss and artifacts that arise from conventional high-low frequency fusion, while the lightweight U-Net pyramid and Retinex-constrained output maps together deliver both high restoration quality and millisecond inference on 4K/8K images using only a single consumer-grade device.

What carries the argument

Spatially aware Clifford algebra that maps feature tensors to multivector space (scalars, vectors, bivectors) and applies Clifford similarity for aggregation during frequency fusion.

If this is right

  • Millisecond inference becomes possible for 4K and 8K images on single consumer-grade devices without specialized hardware.
  • The approach outperforms existing state-of-the-art models on standard image restoration metrics.
  • Textures and edges remain sharper than with traditional frequency-band mixing because Clifford similarity suppresses noise while preserving geometry.
  • Adaptive Gamma and Gain maps produce brightness adjustments that obey Retinex light-reflection constraints.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same multivector fusion step could be inserted into other lightweight pipelines for real-time video or mobile photography tasks.
  • Reducing reliance on heavy Transformer or high-dimensional CNN blocks may lower power consumption in edge devices.
  • Testing the method on additional low-light datasets with motion or extreme noise would reveal whether the Clifford step generalizes beyond static images.

Load-bearing premise

That mapping features to Clifford multivectors and combining them by similarity will fix the structural loss and artifacts of ordinary frequency fusion without the lightweight U-Net losing essential high-quality details needed for UHD inputs.

What would settle it

A side-by-side test on standard 4K low-light images that shows either inference time exceeding 100 ms on a typical consumer GPU or restoration scores (PSNR, SSIM) below those of current leading methods.

Figures

Figures reproduced from arXiv: 2604.09321 by Chen Wu, Dawei Zhao, Dianjie Lu, Guangwei Gao, Guijuan Zhang, Hang Wei, Linwei Fan, Shuai Wu, Xiaohan Wang, Xu Lu, Zhuoran Zheng.

Figure 1
Figure 1. Figure 1: Comprehensive performance comparison of low [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed Clifford Pyramid Enhance (CPE) framework. (a) The overall pipeline based on a multi-scale [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Details of the geometric feature fusion module [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Dynamic resolution reconstruction and adaptive [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Visual comparison between our CPE and other mainstream methods in UHD low-light scenarios. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: High-frequency detail reconstruction comparison [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Demonstration of color constancy and global bright [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 6
Figure 6. Figure 6: Evaluation of global illumination recovery in ex [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 9
Figure 9. Figure 9: Application evaluation of different enhancement [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comprehensive performance comparison of low [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Additional visual comparison between our CPE and other mainstream methods at native 4K resolution. Best viewed [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Hardware overhead evaluation of representative [PITH_FULL_IMAGE:figures/full_fig_p012_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Application evaluation of different enhancement [PITH_FULL_IMAGE:figures/full_fig_p013_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Mobile Deployment on Huawei Mate 60 Pro Huawei Mate 60 Pro (with Kirin NPU) and the iPhone 16 Pro (with A18 Pro Neural Engine) [PITH_FULL_IMAGE:figures/full_fig_p014_14.png] view at source ↗
read the original abstract

Considering efficiency, ultra-high-definition (UHD) low-light image restoration is extremely challenging. Existing methods based on Transformer architectures or high-dimensional complex convolutional neural networks often suffer from the "memory wall" bottleneck, failing to achieve millisecond-level inference on edge devices. To address this issue, we propose a novel real-time UHD low-light enhancement network based on geometric feature fusion using Clifford algebra in 2D Euclidean space. First, we construct a four-layer feature pyramid with gradually increasing resolution, which decomposes input images into low-frequency and high-frequency structural components via a Gaussian blur kernel, and adopts a lightweight U-Net based on depthwise separable convolution for dual-branch feature extraction. Second, to resolve structural information loss and artifacts from traditional high-low frequency feature fusion, we introduce spatially aware Clifford algebra, which maps feature tensors to a multivector space (scalars, vectors, bivectors) and uses Clifford similarity to aggregate features while suppressing noise and preserving textures. In the reconstruction stage, the network outputs adaptive Gamma and Gain maps, which perform physically constrained non-linear brightness adjustment via Retinex theory. Integrated with FP16 mixed-precision computation and dynamic operator fusion, our method achieves millisecond-level inference for 4K/8K images on a single consumer-grade device, while outperforming state-of-the-art (SOTA) models on several restoration metrics.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a real-time UHD low-light image enhancement network that constructs a four-layer feature pyramid, employs a lightweight depthwise-separable U-Net for dual-branch low/high-frequency feature extraction, maps features to Clifford multivectors for similarity-based fusion to mitigate structural loss and artifacts, and uses Retinex theory with adaptive Gamma/Gain maps for reconstruction. FP16 mixed-precision and dynamic operator fusion are integrated to enable claimed millisecond inference on 4K/8K images on consumer hardware while outperforming SOTA on restoration metrics.

Significance. If the performance and quality claims hold under rigorous testing, the work could meaningfully advance real-time UHD low-light enhancement on edge devices by addressing memory bottlenecks in Transformer or high-dimensional CNN approaches. The Clifford-algebra fusion step offers a geometrically motivated alternative to standard frequency fusion that may better preserve textures, representing a potentially useful contribution if validated.

major comments (2)
  1. [Abstract] Abstract: the central claims of 'outperforming state-of-the-art (SOTA) models on several restoration metrics' and 'millisecond-level inference for 4K/8K images on a single consumer-grade device' are presented without any quantitative results, tables, figures, ablation studies, error bars, or implementation details (e.g., hardware specs, FPS numbers, or dataset comparisons). This absence makes the performance assertions impossible to evaluate and is load-bearing for the paper's contribution.
  2. [Methods] Methods (Clifford fusion description): the mapping of feature tensors to multivectors (scalars, vectors, bivectors) and the use of Clifford similarity for aggregation are described at a high level but lack explicit equations, definitions of the similarity measure, or proofs showing how this resolves structural loss and artifacts compared to traditional fusion. Without these, the novelty and correctness of the key technical step cannot be assessed.
minor comments (2)
  1. [Abstract] The abstract and title use vague phrasing such as 'Real-Time Enhancement Methods'; consider making the title and opening more specific to the Clifford fusion and Retinex components.
  2. [Methods] Notation for the four-layer pyramid and dual-branch extraction could be clarified with a diagram or explicit variable definitions to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below, indicating the revisions we will incorporate to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claims of 'outperforming state-of-the-art (SOTA) models on several restoration metrics' and 'millisecond-level inference for 4K/8K images on a single consumer-grade device' are presented without any quantitative results, tables, figures, ablation studies, error bars, or implementation details (e.g., hardware specs, FPS numbers, or dataset comparisons). This absence makes the performance assertions impossible to evaluate and is load-bearing for the paper's contribution.

    Authors: We agree that the abstract would be strengthened by including key quantitative results to support the central claims. Although the full manuscript contains detailed tables, ablation studies, error bars, and implementation details (including hardware specifications and FPS measurements) in the Experiments section, we will revise the abstract to incorporate specific highlights such as PSNR/SSIM gains over SOTA methods and millisecond inference times for 4K/8K images on consumer hardware. This change will make the performance assertions immediately evaluable without altering the paper's overall structure. revision: yes

  2. Referee: [Methods] Methods (Clifford fusion description): the mapping of feature tensors to multivectors (scalars, vectors, bivectors) and the use of Clifford similarity for aggregation are described at a high level but lack explicit equations, definitions of the similarity measure, or proofs showing how this resolves structural loss and artifacts compared to traditional fusion. Without these, the novelty and correctness of the key technical step cannot be assessed.

    Authors: The referee correctly identifies that the Clifford fusion step is presented conceptually in the current text. The manuscript provides the overall framework for mapping to multivectors and using similarity-based aggregation, but we acknowledge the need for greater mathematical rigor. We will add explicit equations defining the tensor-to-multivector mapping, the Clifford similarity measure, and a concise comparison or illustrative derivation showing its advantages in mitigating structural loss and artifacts relative to standard frequency fusion. These additions will be placed in the Methods section to allow full assessment of the technical contribution. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper describes a constructive pipeline: a four-layer feature pyramid with Gaussian decomposition, dual-branch lightweight depthwise-separable U-Net extraction, Clifford multivector mapping for fusion, Retinex-based Gamma/Gain reconstruction, and FP16/dynamic-fusion optimizations. No equations, parameter fits, or derivations are shown that reduce the claimed millisecond inference or metric gains to self-definitions, fitted inputs renamed as predictions, or self-citation chains. The central claims rest on independent architectural choices and external benchmarks rather than any load-bearing step that collapses to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities are stated in sufficient detail to populate the ledger.

pith-pipeline@v0.9.0 · 5571 in / 1223 out tokens · 39746 ms · 2026-05-10T16:44:43.029371+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

67 extracted references · 67 canonical work pages

  1. [1]

    S. M. Ayyoubzadeh and X. Wu. 2021. High Frequency Detail Accentuation in CNN Image Restoration.TIP(2021), 1–13

    work page 2021

  2. [2]

    Haowen Bai, Jiangshe Zhang, Zixiang Zhao, Lilun Deng, Yukun Cui, and Shuang Xu. 2025. Retinex-MEF: Retinex-based Glare Effects Aware Unsupervised Multi- Exposure Image Fusion. InICCV

    work page 2025

  3. [3]

    Juan C Benito et al. 2025. FLOL: Fast Baselines for Real-World Low-Light En- hancement.arXiv preprint arXiv:2501.09718(2025)

    work page arXiv 2025

  4. [4]

    Johannes Brandstetter, Rianne van den Berg, Max Welling, and Jayesh K. Gupta

    work page

  5. [5]

    Clifford Neural Layers for PDE Modeling. InICLR

    work page

  6. [6]

    Johann Brehmer, Pim De Haan, Jens Behrmann, and Taco Cohen. 2023. Geometric Algebra Transformer. InNeurIPS

    work page 2023

  7. [7]

    Yuanhao Cai, Hao Bian, Jing Lin, Haoqian Wang, Timofte Radu, and Yulun Zhang

    work page

  8. [8]

    Retinexformer: One-stage Retinex-based Transformer for Low-light Image Enhancement. InICCV

    work page

  9. [9]

    Chen Chen et al. 2018. Learning to See in the Dark. InCVPR

    work page 2018

  10. [10]

    Chen et al

    G. Chen et al. 2024. Bracketing Image Restoration and Enhancement with High- Low Frequency Decomposition. InCVPR

    work page 2024

  11. [11]

    Liangyu Chen, Xiaojie Chu, Xiangyu Zhang, and Jian Sun. 2022. Simple Baselines for Image Restoration. InECCV

    work page 2022

  12. [12]

    S Chen, R Zhou, H Huang, MI Menhas, et al. 2026. DAFNet: Dynamic Adverse- Weather Feature Network for Climate-Resilient Monitoring of Smart Energy Infrastructure.TIA(2026), 1–18

    work page 2026

  13. [13]

    Yunpeng Chen et al . 2019. Drop an octave: Reducing spatial redundancy in convolutional neural networks with octave convolution. InICCV

    work page 2019

  14. [14]

    Jian Cheng et al. 2024. Towards Efficient Image Detail Enhancement on Mobile Devices. InECCV

    work page 2024

  15. [15]

    Y. Cui, W. Ren, X. Cao, and A. Knoll. 2023. Image Restoration via Frequency Selection.TPAMI(2023), 1–14

    work page 2023

  16. [16]

    Ziteng Cui, Guo-Jun Qi, Lin Gu, Shaodi You, Zenghui Zhang, and Tatsuya Harada

    work page

  17. [17]

    Multitask AET with orthogonal tangent regularity for dark object detection. InICCV

    work page

  18. [18]

    Guodong Fan, Zhentao Yao, Guang-Yong Chen, Jian-Nan Su, and Min Gan. 2025. IniRetinex: Rethinking Retinex-type Low-light Image Enhancer via Initialization Perspective. InAAAI

    work page 2025

  19. [19]

    Hansen Feng, Lizhi Wang, Yiqi Huang, Yuzhi Wang, Lin Zhu, and Hua Huang

    work page

  20. [20]

    Learning Physics-Informed Noise Models from Dark Frames for Low-Light Raw Image Denoising.TPAMI(2026), 3952–3969

    work page 2026

  21. [21]

    Michaël Gharbi, Jiawen Chen, Jonathan T Barron, Samuel W Hasinoff, and Frédo Durand. 2017. Deep bilateral learning for real-time image enhancement. In SIGGRAPH

    work page 2017

  22. [22]

    Xiaojie Guo, Yu Li, and Haibin Ling. 2016. LIME: Low-light Image Enhancement via Illumination Map Estimation.TIP(2016), 982–993

    work page 2016

  23. [23]

    Jinhong He, Minglong Xue, Wenhai Wang, and Mingliang Zhou. 2026. Optimizing a 4D Lookup Table for Low-Light Video Enhancement Via Wavelet Priori.TMM (2026), 1–14

    work page 2026

  24. [24]

    Jie Huang, Yajing Liu, Feng Zhao, Keyu Yan, Jinghao Zhang, Yukun Huang, Man Zhou, and Zhiwei Xiong. 2022. Deep Fourier-Based Exposure Correction Network with Spatial-Frequency Interaction. InECCV

    work page 2022

  25. [25]

    Andrey Ignatov, Radu Timofte, et al. 2021. Learned smartphone ISP on mobile NPUs with deep learning, mobile AI 2021 challenge: Report. InCVPRW

    work page 2021

  26. [26]

    Md Tanvir Islam et al. 2024. LoLI-Street: Benchmarking Low-light Image En- hancement and Beyond. InACCV

    work page 2024

  27. [27]

    Zhongping Ji. 2026. CliffordNet: All You Need is Geometric Algebra.arXiv preprint arXiv:2601.06793(2026)

    work page arXiv 2026

  28. [28]

    Hai Jiang, Binhao Guan, Zhen Liu, Xiaohong Liu, Jian Yu, Zheng Liu, Songchen Han, and Shuaicheng Liu. 2025. Learning to See in the Extremely Dark. InICCV

    work page 2025

  29. [29]

    Chongyi Li et al. 2023. Embedding Fourier for Ultra-High-Definition Low-Light Image Enhancement. InICLR

    work page 2023

  30. [30]

    Chongyi Li, Chunle Guo, and Chen Change Loy. 2021. Learning to Enhance Low-Light Image via Zero-Reference Deep Curve Estimation.TPAMI(2021), 4225–4238

    work page 2021

  31. [31]

    Tsung-Yi Lin, Piotr Dollár, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. 2017. Feature Pyramid Networks for Object Detection. InCVPR

    work page 2017

  32. [32]

    Yunlong Lin et al. 2025. AGLLDiff: Guiding Diffusion Models Towards Unsuper- vised Training-Free Real-World Low-Light Image Enhancement. InAAAI

    work page 2025

  33. [33]

    Yidi Liu et al. 2025. UHD-processer: Unified UHD Image Restoration with Pro- gressive Frequency Learning and Degradation-aware Prompts. InCVPR

    work page 2025

  34. [34]

    Yidi Liu, Dong Li, Jie Xiao, Yuanfei Bao, Senyan Xu, and Xueyang Fu. 2025. Drea- mUHD: Frequency Enhanced Variational Autoencoder for Ultra-High-Definition Image Restoration. InAAAI

    work page 2025

  35. [35]

    Ilya Loshchilov and Frank Hutter. 2017. SGDR: Stochastic Gradient Descent with Warm Restarts. InICLR

    work page 2017

  36. [36]

    Ilya Loshchilov and Frank Hutter. 2019. Decoupled Weight Decay Regularization. InICLR

    work page 2019

  37. [37]

    Long Ma, Tengyu Ma, Risheng Liu, Xin Fan, and Zhongxuan Luo. 2022. Toward Fast, Flexible, and Robust Low-Light Image Enhancement. InCVPR. 6823–6841

    work page 2022

  38. [38]

    completely blind

    Anish Mittal, Rajiv Soundararajan, and Alan C Bovik. 2012. Making a “completely blind” image quality analyzer.SPL(2012), 209–212

    work page 2012

  39. [39]

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. 2015. U-Net: Convolutional Networks for Biomedical Image Segmentation. InMICCAI

    work page 2015

  40. [40]

    Gupta, Steven De Keninck, Max Welling, and Johannes Brandstetter

    David Ruhe, Jayesh K. Gupta, Steven De Keninck, Max Welling, and Johannes Brandstetter. 2023. Geometric Clifford Algebra Networks. InICML

    work page 2023

  41. [41]

    Karen Simonyan and Andrew Zisserman. 2015. Very Deep Convolutional Net- works for Large-Scale Image Recognition. InICLR

    work page 2015

  42. [42]

    Shangquan Sun, Wenqi Ren, Jingyang Peng, Fenglong Song, and Xiaochun Cao

    work page

  43. [43]

    DI-Retinex: Digital-Imaging Retinex Model for Low-Light Image Enhance- ment.IJCV(2025), 8293–8314

    work page 2025

  44. [44]

    Ao Wang, Hui Chen, Lihao Liu, Kai Chen, Zijia Lin, Jungong Han, and Guiguang Ding. 2024. YOLOv10: Real-Time End-to-End Object Detection. InNeurIPS

    work page 2024

  45. [45]

    Huake Wang, Xingsong Hou, Jutao Li, Yadi Yan, Wenke Sun, and Xin Zeng

    work page

  46. [46]

    TMM(2025), 5709 – 5721

    Multi-Scale Retinex Unfolding Network for Low-Light Image Enhancement. TMM(2025), 5709 – 5721

    work page 2025

  47. [47]

    Tao Wang et al. 2023. Ultra-High-Definition Low-Light Image Enhancement: A Benchmark and Transformer-Based Method. InAAAI

    work page 2023

  48. [48]

    Xi Wang, Xueyang Fu, Yurui Zhu, and Zheng-Jun Zha. 2025. DDCNet: Advanced Decoupling of Degradation and Content for Adverse Weather Image Restoration. TNNLS(2025), 20288 – 20301. Conference’17, July 2017, Washington, DC, USA Xiaohan Wang, Chen Wu, Dawei Zhao, Guangwei Gao, Dianjie Lu, Guijuan Zhang, Linwei Fan, Xu Lu, Shuai Wu, Hang Wei, and Zhuoran Zheng

    work page 2025

  49. [49]

    Zhou Wang et al. 2004. Image quality assessment: from error visibility to struc- tural similarity.TIP(2004), 600–612

    work page 2004

  50. [50]

    Chen Wei et al. 2018. Deep Retinex Decomposition for Low-light Enhancement. InBMVC

    work page 2018

  51. [51]

    Changguang Wu, Jiangxin Dong, Hao Hou, and Jinhui Tang. 2026. Sparse Curve Estimation for Real-Time Low-Light Ultra-High-Definition Image Enhancement. TCSVT(2026)

    work page 2026

  52. [52]

    C Wu, L Wang, Z Zheng, W Jiang, et al . 2025. Ultra-High-Definition Image Restoration via High-Frequency Enhanced Transformer.TCSVT(2025)

    work page 2025

  53. [53]

    Chien-Sheng Wu et al . 2023. Edge AI: On-demand accelerating deep neural network inference via edge computing.TWC(2023), 45–58

    work page 2023

  54. [54]

    Rui Xu, Yuzhen Niu, Yuezhou Li, Huangbiao Xu, Wenxi Liu, and Yuzhong Chen

    work page

  55. [55]

    URWKV: Unified RWKV Model with Multi-state Perspective for Low-light Image Restoration. InCVPR

    work page

  56. [56]

    Qingsen Yan et al. 2025. HVI: A New Color Space for Low-light Image Enhance- ment. InCVPR

    work page 2025

  57. [57]

    Xunpeng Yi, Han Xu, Hao Zhang, Linfeng Tang, and Jiayi Ma. 2025. Diff- Retinex++: Retinex-Driven Reinforced Diffusion Model for Low-Light Image Enhancement.TPAMI(2025)

    work page 2025

  58. [58]

    Xin Yu, Peng Dai, Wenbo Li, Lan Ma, Jiawei Shen, Jiawei Zhang, and Xiao- juan Qi. 2022. Towards Efficient and Scale-Robust Ultra-High-Definition Image Demoiréing. InECCV

    work page 2022

  59. [59]

    Syed Waqas Zamir et al . 2022. Restormer: Efficient Transformer for High- Resolution Image Restoration. InCVPR

    work page 2022

  60. [60]

    Syed Waqas Zamir, Aditya Arora, Salman Khan, Munawar Hayat, Fahad Shahbaz Khan, Ming-Hsuan Yang, and Ling Shao. 2023. Learning enriched features for fast image restoration and enhancement.TPAMI(2023), 1934–1948

    work page 2023

  61. [61]

    F Zhang, H Deng, Z Li, L Li, B Xu, Q Lu, et al . 2025. High-resolution photo enhancement in real-time: a laplacian pyramid network.TPAMI(2025), 2170 – 2185

    work page 2025

  62. [62]

    Fan Zhang, Yu Li, Shaodi You, and Ying Fu. 2021. Learning Temporal Consistency for Low Light Video Enhancement from Single Images. InCVPR

    work page 2021

  63. [63]

    Richard Zhang et al. 2018. The Unreasonable Effectiveness of Deep Features as a Perceptual Metric. InCVPR

    work page 2018

  64. [64]

    Yonghua Zhang et al. 2021. Beyond Brightening Low-light Images.IJCV(2021), 1153–1184

    work page 2021

  65. [65]

    Chen Zhao et al. 2025. From Zero to Detail: Deconstructing Ultra-high-definition Image Restoration from Progressive Spectral Perspective. InCVPR

    work page 2025

  66. [66]

    P Zheng, H Jiang, F Sun, L Chen, Q Kou, D Cheng, et al . 2026. HMSR: Hypercomplex-Guided Mamba for Fine-Texture Coupling in Single Image Super- Resolution.PR(2026)

    work page 2026

  67. [67]

    Xuanya Zhu, Yi Xu, Hongteng Xu, and Changjian Chen. 2018. Quaternion Convolutional Neural Networks. InECCV. UHD Low-Light Image Enhancement via Real-Time Enhancement Methods with Clifford Information Fusion Conference’17, July 2017, Washington, DC, USA Appendix Overview This appendix aims to provide comprehensive supplementary ma- terial to the main manus...

    work page 2018