pith. sign in

arxiv: 2604.24053 · v1 · submitted 2026-04-27 · 💻 cs.CV

Light 'em Up: Enabling Few-Shot Low-Light 3D Gaussian Splatting with Multi-Scale Explicit Retinex Illumination Decoupling

YuHao Yin , Zongji Wang , Yuanben Zhang , Biqing Li , Jiesong Bai , Junyi Liu This is my paper

Pith reviewed 2026-05-08 04:41 UTC · model grok-4.3

classification 💻 cs.CV
keywords low-light novel view synthesis3D Gaussian SplattingRetinex illumination decouplingfew-shot 3D reconstruction360 degree scenesmulti-scale frequency gatingcross-scene generalization
0
0 comments X

The pith

MERID-GS decouples illumination from reflectance at multiple scales to stabilize few-shot low-light 360° 3D Gaussian Splatting.

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

The paper seeks to enable consistent novel-view synthesis in poorly lit 360° scenes when only a handful of input images are available. Existing 3D Gaussian Splatting pipelines break down under low light because noise amplifies and lighting varies sharply with viewpoint, while pure 2D enhancement networks fail to generalize across scenes. By explicitly factoring each image into an illumination map and a reflectance map according to Retinex theory, then feeding the reflectance into a 3D Gaussian pipeline, the method suppresses noise in dark regions and removes view-dependent color shifts without retraining. A sympathetic reader would care because this turns ordinary low-light phone captures into reliable 3D models for robotics, AR, or virtual tourism.

Core claim

MERID-GS achieves state-of-the-art low-light 360° novel-view synthesis by explicitly separating illumination and reflectance via multi-scale Retinex processing, applying a learnable gain and Illumination-State-Guided Frequency Gating to suppress noise while preserving structure, and combining the resulting reflectance with a lightweight Reflection Head inside 3D Gaussian Splatting; the resulting model adapts to new scenes from sparse views and maintains geometric and photometric consistency across large viewpoint changes.

What carries the argument

Multi-scale explicit Retinex illumination decoupling, which factors each view into separate illumination and reflectance components before they enter the 3D Gaussian Splatting optimization.

If this is right

  • New scenes can be modeled from only a few low-light photographs without per-scene retraining.
  • Noise is suppressed and dark structures are enhanced while view-to-view photometric consistency is preserved.
  • Full 360° synthesis becomes feasible under illumination conditions that defeat prior unsupervised and supervised baselines.

Where Pith is reading between the lines

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

  • The same decoupling might reduce artifacts in other challenging capture conditions such as motion blur or mixed indoor-outdoor lighting.
  • The released low-light multi-view dataset offers a concrete benchmark that future 3D reconstruction methods can be measured against for generalization.
  • Integrating the illumination-reflectance split into real-time 3D Gaussian pipelines could support low-light AR or robotic navigation.

Load-bearing premise

That factoring illumination away from reflectance at multiple scales will not create new geometric or color inconsistencies when the reflectance is later reconstructed in 3D from sparse low-light inputs.

What would settle it

A controlled test on the paper's own low-light 360° multi-view dataset showing large color drift or broken geometry in novel views synthesized from only a few input frames would falsify the central claim.

Figures

Figures reproduced from arXiv: 2604.24053 by Biqing Li, Jiesong Bai, Junyi Liu, Yuanben Zhang, YuHao Yin, Zongji Wang.

Figure 1
Figure 1. Figure 1 view at source ↗
Figure 2
Figure 2. Figure 2: Overview of MERID-GS. The first stage (a)(b) corresponds to Multi-scale Ex￾plicit Retinex Illumination Decoupling (MERID), which explicitly separates illumina￾tion and reflectance based on Retinex theory, and incorporates an Illumination-State￾guided Frequency Gated Attention (IS-FGA, see (d)) within a U-Net-like architecture. The second stage (c) is the Reflection Head, which performs lightweight fine-tun… view at source ↗
Figure 3
Figure 3. Figure 3: Visual comparison on the bike scene in LLD, including our result, ground truth , the input, and the model without the Reflection Head, along with the corresponding brightness curves. Let the reflectance feature predicted by MERID be R0 ∈ R H×W×3 , the final reflectance output is expressed as: \mathbf {R}=\mathbf {R}_0+\psi (\mathbf {R}_0), (9) where ψ(·) denotes a lightweight channel mapping network. Due t… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparisons on representative scenes (bike, sofa, kitchen) from the NeRF360, LOM_full, and LLD datasets, against representative baseline methods view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparisons on representative scenes (bike, chair, and counter) from the NeRF360, LOM_full, and LLD datasets, against representative baseline methods view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparisons on representative scenes (bike, chair, and bonsai) from the NeRF360, LOM_full, and LLD datasets, along with comparative analysis against four ablation settings. MERID-GS outperforms existing methods in synthesis quality and cross-scene generalization, especially in few-shot settings. The LLD dataset provides a new benchmark for low-light novel view synthesis. Future work will explor… view at source ↗
read the original abstract

Full 360$^\circ$ novel view synthesis under low-light conditions remains challenging. Insufficient illumination, noise amplification, and view-dependent photometric inconsistencies prevent existing methods from jointly preserving geometric consistency and photorealism. Unsupervised approaches often exhibit color drift under large viewpoint variations, while supervised low-light enhancement models, though effective for 2D tasks, struggle to generalize to new scenes and typically require retraining. To address this issue, we propose MERID-GS, a Multi-Scale Explicit Retinex Illumination-Decoupled Gaussian framework for low-light 360$^\circ$ synthesis. Based on Retinex theory, the method explicitly separates illumination and reflectance, and suppresses noise propagation while enhancing dark-region structures via a learnable gain and Illumination-State-Guided Frequency Gating. Combined with lightweight Reflection Head and 3D Gaussian Splatting, MERID-GS adapts to new scenes with only a few shots and enables stable low-light novel view synthesis from sparse-view observations. In addition, we construct a low-light multi-view dataset covering full 360$^\circ$ scenes for joint evaluation. Thorough experiments across multiple datasets in this area demonstrate that MERID-GS achieves SOTA performance, exhibiting superior cross-scene generalization and view consistency. The source code and pre-trained models are available at https://github.com/YhuoyuH/MERID-GS..

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 MERID-GS, a Multi-Scale Explicit Retinex Illumination-Decoupled Gaussian framework for few-shot low-light 360° novel view synthesis. It explicitly separates illumination and reflectance per Retinex theory, applies a learnable gain and Illumination-State-Guided Frequency Gating to handle noise and dark regions, and feeds a lightweight Reflection Head into 3D Gaussian Splatting for adaptation to new scenes from sparse views. The authors also introduce a low-light multi-view 360° dataset and claim SOTA results with superior cross-scene generalization and view consistency.

Significance. If the performance claims hold with proper validation, the work could advance practical few-shot 3D reconstruction in low-light settings by reducing reliance on per-scene supervised retraining and addressing photometric inconsistencies, with potential applications in robotics and immersive environments.

major comments (2)
  1. [Abstract] Abstract: The central claims of SOTA performance, superior cross-scene generalization, and view consistency are asserted without any quantitative tables, ablation results, error analysis, or specific metrics (e.g., PSNR, SSIM, or consistency measures), making it impossible to verify that the method outperforms baselines rather than merely describing it.
  2. [Methods] Methods (Retinex decoupling and Reflection Head): The per-image operations (learnable gain, Illumination-State-Guided Frequency Gating) are described without an explicit multi-view reflectance consistency loss, 3D-aware regularization term, or cross-view constraint in the optimization. This leaves the view-consistency claim vulnerable, as independent per-view decoupling on sparse noisy inputs can introduce photometric inconsistencies that 3DGS optimization may bake in rather than suppress.
minor comments (2)
  1. [Abstract] The acronym MERID-GS is introduced without immediate expansion in the abstract, which could be clarified for readability.
  2. [Experiments] The dataset construction is mentioned but lacks details on capture conditions, number of scenes/views, or lighting variation ranges, which would strengthen reproducibility claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below with clarifications and indicate revisions to the manuscript where appropriate.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claims of SOTA performance, superior cross-scene generalization, and view consistency are asserted without any quantitative tables, ablation results, error analysis, or specific metrics (e.g., PSNR, SSIM, or consistency measures), making it impossible to verify that the method outperforms baselines rather than merely describing it.

    Authors: The abstract is intended as a concise summary of contributions and claims. The full manuscript contains quantitative tables in the Experiments section reporting PSNR, SSIM, LPIPS, and other metrics, along with ablations, error analysis, and baseline comparisons that support the SOTA claims, cross-scene generalization, and view consistency. We agree that key numerical results would strengthen the abstract. We will revise the abstract to include specific metrics (e.g., average PSNR improvement) drawn from the experimental results. revision: yes

  2. Referee: [Methods] Methods (Retinex decoupling and Reflection Head): The per-image operations (learnable gain, Illumination-State-Guided Frequency Gating) are described without an explicit multi-view reflectance consistency loss, 3D-aware regularization term, or cross-view constraint in the optimization. This leaves the view-consistency claim vulnerable, as independent per-view decoupling on sparse noisy inputs can introduce photometric inconsistencies that 3DGS optimization may bake in rather than suppress.

    Authors: The referee is correct that no explicit multi-view reflectance consistency loss or cross-view regularization term is defined for the per-image decoupling stage. Our design relies on the 3D Gaussian Splatting optimization and the lightweight Reflection Head to enforce consistency via the shared 3D representation and rendering losses across views. This implicit mechanism is intended to suppress inconsistencies from sparse low-light inputs. To strengthen the presentation, we will add a discussion in the Methods section explaining the role of 3DGS in view consistency and include a quantitative analysis of photometric consistency in the experiments. revision: partial

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper introduces MERID-GS as a novel architectural combination of multi-scale explicit Retinex decoupling (learnable gain, frequency gating) with a lightweight Reflection Head and 3D Gaussian Splatting. No equations or derivations are presented that reduce the claimed SOTA cross-scene generalization or view consistency to quantities defined by the method's own fitted parameters or prior self-citations. Retinex theory is invoked as an external standard reference rather than a self-derived result, and performance assertions rest on empirical evaluation across datasets rather than tautological reduction. The construction is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on Retinex theory as a domain assumption and introduces several new algorithmic components whose independent validation is not provided in the abstract.

free parameters (1)
  • learnable gain
    A trainable scalar or map used to enhance dark-region structures; its value is learned during optimization rather than derived from first principles.
axioms (1)
  • domain assumption Retinex theory applies directly to multi-view low-light images and permits clean separation of illumination and reflectance without residual view-dependent effects
    The method is built on the premise that illumination and reflectance can be explicitly decoupled at multiple scales while preserving geometric consistency in 3D splatting.
invented entities (1)
  • Illumination-State-Guided Frequency Gating no independent evidence
    purpose: To suppress noise propagation and enhance dark-region structures during illumination decoupling
    A new gating mechanism conditioned on illumination state that is introduced as part of the framework.

pith-pipeline@v0.9.0 · 5571 in / 1470 out tokens · 34015 ms · 2026-05-08T04:41:56.566681+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

78 extracted references · 78 canonical work pages

  1. [1]

    IEEE TCE53(2), 593–600 (2007) 4

    Abdullah-Al-Wadud, M., Kabir, M.H., Dewan, M.A.A., Chae, O.: A dynamic his- togram equalization for image contrast enhancement. IEEE TCE53(2), 593–600 (2007) 4

    work page 2007

  2. [2]

    In: CVPR

    Barron, J.T., Mildenhall, B., Verbin, D., Srinivasan, P.P., Hedman, P.: Mip-nerf 360: Unbounded anti-aliased neural radiance fields. In: CVPR. pp. 5470–5479 (2022) 2, 3, 4

    work page 2022

  3. [3]

    In: ICCV

    Barron, J.T., Mildenhall, B., Verbin, D., Srinivasan, P.P., Hedman, P.: Zip-nerf: Anti-aliased grid-based neural radiance fields. In: ICCV. pp. 19697–19705 (2023) 2, 4

    work page 2023

  4. [4]

    In: ICCV

    Cai, Y., Bian, H., Lin, J., Wang, H., Timofte, R., Zhang, Y.: Retinexformer: One- stage retinex-based transformer for low-light image enhancement. In: ICCV. pp. 12504–12513 (2023) 3, 5, 10, 11, 12

    work page 2023

  5. [5]

    In: NeurIPS

    Cai, Y., Lin, J., Wang, H., Yuan, X., Ding, H., Zhang, Y., Timofte, R., Gool, L.V.:Degradation-awareunfoldinghalf-shuffletransformerforspectralcompressive imaging. In: NeurIPS. pp. 37749–37761 (2022) 5

    work page 2022

  6. [6]

    In: NeurIPS

    Cai,Y.,Xiao,Z.,Liang,Y.,Qin,M.,Zhang,Y.,Yang,X.,Liu,Y.,Yuille,A.L.:Hdr- gs: Efficient high dynamic range novel view synthesis at 1000x speed via gaussian splatting. In: NeurIPS. pp. 68453–68471 (2024) 2

    work page 2024

  7. [7]

    IEEE TIP20(12), 3431–3441 (2011) 4

    Celik, T., Tjahjadi, T.: Contextual and variational contrast enhancement. IEEE TIP20(12), 3431–3441 (2011) 4

    work page 2011

  8. [8]

    Charatan,D.,Li,S.L.,Tagliasacchi,A.,Sitzmann,V.:pixelsplat:3dgaussiansplats fromimagepairsforscalablegeneralizable3dreconstruction.In:CVPR.pp.19457– 19467 (2024) 2, 4

    work page 2024

  9. [9]

    In: ICCV

    Chen, C., Chen, Q., Do, M.N., Koltun, V.: Seeing motion in the dark. In: ICCV. pp. 3185–3194 (2019) 3

    work page 2019

  10. [10]

    In: CVPR

    Chen, H., Wang, Y., Guo, T., Xu, C., Deng, Y., Liu, Z., Ma, S., Xu, C., Xu, C., Gao, W.: Pre-trained image processing transformer. In: CVPR. pp. 12299–12310 (2021) 5 16 Y. Yin et al

    work page 2021

  11. [11]

    In: CVPR

    Chen, X., Zhang, Q., Li, X., Chen, Y., Feng, Y., Wang, X., Wang, J.: Hallucinated neural radiance fields in the wild. In: CVPR. pp. 12943–12952 (2022) 2

    work page 2022

  12. [12]

    Digital Signal Processing14(2), 158–170 (2004) 4

    Cheng, H.D., Shi, X.: A simple and effective histogram equalization approach to image enhancement. Digital Signal Processing14(2), 158–170 (2004) 4

    work page 2004

  13. [13]

    In: SIGGRAPH Asia

    Chugunov, I., Joshi, A., Murthy, K., Bleibel, F., Heide, F.: Neural light spheres for implicit image stitching and view synthesis. In: SIGGRAPH Asia. pp. 1–11 (2024) 4

    work page 2024

  14. [14]

    In: CVPR

    Cui, Z., Chu, X., Harada, T.: Luminance-gs: Adapting 3d gaussian splatting to challenging lighting conditions with view-adaptive curve adjustment. In: CVPR. pp. 26472–26482 (2025) 2, 5, 10, 11

    work page 2025

  15. [15]

    In: AAAI

    Cui, Z., Gu, L., Sun, X., Ma, X., Qiao, Y., Harada, T.: Aleth-nerf: Illumination adaptive nerf with concealing field assumption. In: AAAI. pp. 1435–1444 (2024) 2, 3, 5

    work page 2024

  16. [16]

    In: CVPR

    Deng, K., Liu, A., Zhu, J.Y., Ramanan, D.: Depth-supervised nerf: Fewer views and faster training for free. In: CVPR. pp. 12882–12891 (2022) 2, 4

    work page 2022

  17. [17]

    In: NeurIPS

    Fan, Z., Wang, K., Wen, K., Zhu, Z., Xu, D., Wang, Z.: Lightgaussian: Un- bounded 3d gaussian compression with 15x reduction and 200+ fps. In: NeurIPS. pp. 140138–140158 (2024) 2

    work page 2024

  18. [18]

    In: CVPR

    Fu, X., Zeng, D., Huang, Y., Zhang, X.P., Ding, X.: A weighted variational model for simultaneous reflectance and illumination estimation. In: CVPR. pp. 2782–2790 (2016) 4

    work page 2016

  19. [19]

    Knowledge-Based Systems240, 108010 (2022) 4

    Fu, Y., Hong, Y., Chen, L., You, S.: Le-gan: Unsupervised low-light image enhance- ment network using attention module and identity invariant loss. Knowledge-Based Systems240, 108010 (2022) 4

    work page 2022

  20. [20]

    In: CVPR

    Guo, C., Li, C., Guo, J., Loy, C.C., Hou, J., Kwong, S., Cong, R.: Zero-reference deep curve estimation for low-light image enhancement. In: CVPR. pp. 1780–1789 (2020) 3, 4, 10, 11

    work page 2020

  21. [21]

    In: ACM SIGGRAPH

    Huang, B., Yu, Z., Chen, A., Geiger, A., Gao, S.: 2d gaussian splatting for geo- metrically accurate radiance fields. In: ACM SIGGRAPH. pp. 1–11 (2024) 4

    work page 2024

  22. [22]

    IEEE TIP22(3), 1032–1041 (2012) 4

    Huang, S.C., Cheng, F.C., Chiu, Y.S.: Efficient contrast enhancement using adap- tive gamma correction with weighting distribution. IEEE TIP22(3), 1032–1041 (2012) 4

    work page 2012

  23. [23]

    IEEE TIP30, 2340–2349 (2021) 4

    Jiang, Y., Gong, X., Liu, D., Cheng, Y., Fang, C., Shen, X., Yang, J., Zhou, P., Wang, Z.: Enlightengan: Deep light enhancement without paired supervision. IEEE TIP30, 2340–2349 (2021) 4

    work page 2021

  24. [24]

    IEEE TIP6(7), 965–976 (1997) 4

    Jobson, D.J., Rahman, Z.u., Woodell, G.A.: A multiscale retinex for bridging the gap between color images and the human observation of scenes. IEEE TIP6(7), 965–976 (1997) 4

    work page 1997

  25. [25]

    IEEE TIP6(3), 451–462 (1997) 4

    Jobson, D.J., Rahman, Z.u., Woodell, G.A., et al.: Properties and performance of a center/surround retinex. IEEE TIP6(3), 451–462 (1997) 4

    work page 1997

  26. [26]

    ACM TOG42(4), 1–14 (2023) 2, 4, 11

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM TOG42(4), 1–14 (2023) 2, 4, 11

    work page 2023

  27. [27]

    In: NeurIPS

    Kulhanek, J., Peng, S., Kukelova, Z., Pollefeys, M., Sattler, T.: Wildgaussians: 3d gaussian splatting in the wild. In: NeurIPS. pp. 21271–21288 (2024) 2

    work page 2024

  28. [28]

    Scientific American237(6), 108–129 (1977) 3

    Land, E.H.: The retinex theory of color vision. Scientific American237(6), 108–129 (1977) 3

    work page 1977

  29. [29]

    IEEE TIP27(6), 2828–2841 (2018) 4

    Li, M., Liu, J., Yang, W., Sun, X., Guo, Z.: Structure-revealing low-light image enhancement via robust retinex model. IEEE TIP27(6), 2828–2841 (2018) 4

    work page 2018

  30. [30]

    In: ECCV

    Li, R., Fidler, S., Kanazawa, A., Williams, F.: Nerf-xl: Scaling nerfs with multiple gpus. In: ECCV. pp. 92–107. Springer (2024) 2 Title Suppressed Due to Excessive Length 17

    work page 2024

  31. [31]

    In: ICCV

    Li, R., Gao, H., Tancik, M., Kanazawa, A.: Nerfacc: Efficient sampling accelerates nerfs. In: ICCV. pp. 18537–18546 (2023) 2, 4

    work page 2023

  32. [32]

    In: ECCV

    Liu, T., Wang, G., Hu, S., Shen, L., Ye, X., Zang, Y., Cao, Z., Li, W., Liu, Z.: Mvsgaussian: Fast generalizable gaussian splatting reconstruction from multi-view stereo. In: ECCV. pp. 37–53. Springer (2024) 2, 4

    work page 2024

  33. [33]

    PR61, 650–662 (2017) 5

    Lore,K.G.,Akintayo,A.,Sarkar,S.:Llnet:Adeepautoencoderapproachtonatural low-light image enhancement. PR61, 650–662 (2017) 5

    work page 2017

  34. [34]

    In: CVPR

    Lu, T., Yu, M., Xu, L., Xiangli, Y., Wang, L., Lin, D., Dai, B.: Scaffold-gs: Struc- tured 3d gaussians for view-adaptive rendering. In: CVPR. pp. 20654–20664 (2024) 4

    work page 2024

  35. [35]

    In: BMVC

    Lv, F., Lu, F., Wu, J., Lim, C.: Mbllen: Low-light image/video enhancement using cnns. In: BMVC. p. 4. Northumbria University (2018) 5

    work page 2018

  36. [36]

    In: CVPR

    Ma, L., Ma, T., Liu, R., Fan, X., Luo, Z.: Toward fast, flexible, and robust low-light image enhancement. In: CVPR. pp. 5637–5646 (2022) 3, 4, 10, 11

    work page 2022

  37. [37]

    In: CVPR

    Martin-Brualla,R.,Radwan,N.,Sajjadi,M.S.,Barron,J.T.,Dosovitskiy,A.,Duck- worth, D.: Nerf in the wild: Neural radiance fields for unconstrained photo collec- tions. In: CVPR. pp. 7210–7219 (2021) 2

    work page 2021

  38. [38]

    In: CVPR

    Mildenhall, B., Hedman, P., Martin-Brualla, R., Srinivasan, P.P., Barron, J.T.: Nerf in the dark: High dynamic range view synthesis from noisy raw images. In: CVPR. pp. 16190–16199 (2022) 2

    work page 2022

  39. [39]

    In: ECCV (2020) 2, 4

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. In: ECCV (2020) 2, 4

    work page 2020

  40. [40]

    In: CVPR

    Moran, S., Marza, P., McDonagh, S., Parisot, S., Slabaugh, G.: Deeplpf: Deep local parametric filters for image enhancement. In: CVPR. pp. 12826–12835 (2020) 5

    work page 2020

  41. [41]

    ACM TOG41(4), 1–15 (2022) 4

    Müller, T., Evans, A., Schied, C., Keller, A.: Instant neural graphics primitives with a multiresolution hash encoding. ACM TOG41(4), 1–15 (2022) 4

    work page 2022

  42. [42]

    In: CVPR

    Niedermayr, S., Stumpfegger, J., Westermann, R.: Compressed 3d gaussian splat- ting for accelerated novel view synthesis. In: CVPR. pp. 10349–10358 (2024) 2, 4

    work page 2024

  43. [43]

    In: NeurIPS (2019) 9

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al.: Pytorch: An imperative style, high- performance deep learning library. In: NeurIPS (2019) 9

    work page 2019

  44. [44]

    Journal of Digital Imaging11(4), 193 (1998) 4

    Pisano, E.D., Zong, S., Hemminger, B.M., DeLuca, M., Johnston, R.E., Muller, K., Braeuning, M.P., Pizer, S.M.: Contrast limited adaptive histogram equaliza- tion image processing to improve the detection of simulated spiculations in dense mammograms. Journal of Digital Imaging11(4), 193 (1998) 4

    work page 1998

  45. [45]

    Computer Vision, Graphics, and Image Processing 39(3), 355–368 (1987) 4

    Pizer, S.M., Amburn, E.P., Austin, J.D., Cromartie, R., Geselowitz, A., Greer, T., ter Haar Romeny, B., Zimmerman, J.B., Zuiderveld, K.: Adaptive histogram equalization and its variations. Computer Vision, Graphics, and Image Processing 39(3), 355–368 (1987) 4

    work page 1987

  46. [46]

    In: SIGGRAPH Asia

    Pu, G., Zhao, Y., Lian, Z.: Pano2room: Novel view synthesis from a single indoor panorama. In: SIGGRAPH Asia. pp. 1–11 (2024) 4

    work page 2024

  47. [47]

    EURASIP Journal on Image and Video Processing2016(1), 35 (2016) 4

    Rahman, S., Rahman, M.M., Abdullah-Al-Wadud, M., Al-Quaderi, G.D., Shoyaib, M.: An adaptive gamma correction for image enhancement. EURASIP Journal on Image and Video Processing2016(1), 35 (2016) 4

    work page 2016

  48. [48]

    In: CVPR

    Sharma, A., Tan, R.T.: Nighttime visibility enhancement by increasing the dy- namic range and suppression of light effects. In: CVPR. pp. 11977–11986 (2021) 5 18 Y. Yin et al

    work page 2021

  49. [49]

    Sun, H., Yu, F., Xu, H., Zhang, T., Zou, C.: Ll-gaussian: Low-light scene recon- structionandenhancementviagaussiansplattingfornovelviewsynthesis.In:ACM MM. pp. 4261–4270 (2025) 2, 4, 5

    work page 2025

  50. [50]

    In: CVPR

    Tancik, M., Casser, V., Yan, X., Pradhan, S., Mildenhall, B., Srinivasan, P.P., Bar- ron, J.T., Kretzschmar, H.: Block-nerf: Scalable large scene neural view synthesis. In: CVPR. pp. 8248–8258 (2022) 4

    work page 2022

  51. [51]

    In: ICCV

    Wang, H., Xu, X., Xu, K., Lau, R.W.: Lighting up nerf via unsupervised decom- position and enhancement. In: ICCV. pp. 12632–12641 (2023) 2

    work page 2023

  52. [52]

    In: ICCV

    Wang, R., Xu, X., Fu, C.W., Lu, J., Yu, B., Jia, J.: Seeing dynamic scene in the dark: A high-quality video dataset with mechatronic alignment. In: ICCV. pp. 9700–9709 (2021) 3

    work page 2021

  53. [53]

    In: CVPR

    Wang, R., Zhang, Q., Fu, C.W., Shen, X., Zheng, W.S., Jia, J.: Underexposed photo enhancement using deep illumination estimation. In: CVPR. pp. 6849–6857 (2019) 5

    work page 2019

  54. [54]

    IEEE TIP22(9), 3538–3548 (2013) 4

    Wang, S., Zheng, J., Hu, H.M., Li, B.: Naturalness preserved enhancement algo- rithm for non-uniform illumination images. IEEE TIP22(9), 3538–3548 (2013) 4

    work page 2013

  55. [55]

    In: CVPR

    Wang, W., Yang, H., Fu, J., Liu, J.: Zero-reference low-light enhancement via physical quadruple priors. In: CVPR. pp. 26057–26066 (2024) 3, 4, 10, 11

    work page 2024

  56. [56]

    ACM TOG43(4), 1–13 (2024) 2

    Wang, Y., Wang, C., Gong, B., Xue, T.: Bilateral guided radiance field processing. ACM TOG43(4), 1–13 (2024) 2

    work page 2024

  57. [57]

    Displays30(3), 133–139 (2009) 4

    Wang, Z.G., Liang, Z.H., Liu, C.L.: A real-time image processor with combin- ing dynamic contrast ratio enhancement and inverse gamma correction for pdp. Displays30(3), 133–139 (2009) 4

    work page 2009

  58. [58]

    In: BMVC (2018) 3

    Wei, C., Wang, W., Yang, W., Liu, J.: Deep retinex decomposition for low-light enhancement. In: BMVC (2018) 3

    work page 2018

  59. [59]

    In: NeurIPS

    Weng, J., Yan, Z., Tai, Y., Qian, J., Yang, J., Li, J.: Mamballie: Implicit retinex- aware low light enhancement with global-then-local state space. In: NeurIPS. pp. 27440–27462 (2024) 3, 5, 10, 11, 12

    work page 2024

  60. [60]

    In: CVPR

    Wu, G., Yi, T., Fang, J., Xie, L., Zhang, X., Wei, W., Liu, W., Tian, Q., Wang, X.: 4d gaussian splatting for real-time dynamic scene rendering. In: CVPR. pp. 20310–20320 (2024) 4

    work page 2024

  61. [61]

    In: CVPR

    Xu, X., Wang, R., Fu, C.W., Jia, J.: Snr-aware low-light image enhancement. In: CVPR. pp. 17714–17724 (2022) 3, 5, 10, 11, 12

    work page 2022

  62. [62]

    In: ICCV

    Xu, X., Wu, J., Yan, Q., Cui, J., Hong, R., Yu, B.: Learnable feature patches and vectors for boosting low-light image enhancement without external knowledge. In: ICCV. pp. 7761–7770 (2025) 10, 11, 12

    work page 2025

  63. [63]

    In: ICCV

    Yang, S., Ding, M., Wu, Y., Li, Z., Zhang, J.: Implicit neural representation for cooperative low-light image enhancement. In: ICCV. pp. 12918–12927 (2023) 3, 4, 10, 11

    work page 2023

  64. [64]

    IEEE TIP30, 2072– 2086 (2021) 3

    Yang, W., Wang, W., Huang, H., Wang, S., Liu, J.: Sparse gradient regularized deep retinex network for robust low-light image enhancement. IEEE TIP30, 2072– 2086 (2021) 3

    work page 2072

  65. [65]

    In: ICCV

    Yang, Y., Zhang, S., Huang, Z., Zhang, Y., Tan, M.: Cross-ray neural radiance fields for novel-view synthesis from unconstrained image collections. In: ICCV. pp. 15901–15911 (2023) 2

    work page 2023

  66. [66]

    JMLR 26(34), 1–17 (2025) 10 Title Suppressed Due to Excessive Length 19

    Ye, V., Li, R., Kerr, J., Turkulainen, M., Yi, B., Pan, Z., Seiskari, O., Ye, J., Hu, J., Tancik, M., et al.: gsplat: An open-source library for gaussian splatting. JMLR 26(34), 1–17 (2025) 10 Title Suppressed Due to Excessive Length 19

    work page 2025

  67. [67]

    In: CVPR

    Zamir, S.W., Arora, A., Khan, S., Hayat, M., Khan, F.S., Yang, M.H.: Restormer: Efficient transformer for high-resolution image restoration. In: CVPR. pp. 5728– 5739 (2022) 3, 5, 10, 11, 12

    work page 2022

  68. [68]

    In: ECCV

    Zamir, S.W., Arora, A., Khan, S., Hayat, M., Khan, F.S., Yang, M.H., Shao, L.: Learning enriched features for real image restoration and enhancement. In: ECCV. pp. 492–511. Springer (2020) 3, 5, 10, 11, 12

    work page 2020

  69. [69]

    In: ECCV

    Zhang, D., Wang, C., Wang, W., Li, P., Qin, M., Wang, H.: Gaussian in the wild: 3d gaussian splatting for unconstrained image collections. In: ECCV. pp. 341–359. Springer (2024) 2

    work page 2024

  70. [70]

    In: CVPR

    Zhang, F., Li, Y., You, S., Fu, Y.: Learning temporal consistency for low light video enhancement from single images. In: CVPR. pp. 4967–4976 (2021) 3, 4, 10, 11

    work page 2021

  71. [71]

    Nerf++: Analyzing and improving neural radiance fields

    Zhang, K., Riegler, G., Snavely, N., Koltun, V.: Nerf++: Analyzing and improving neural radiance fields. arXiv preprint arXiv:2010.07492 (2020) 2

    work page arXiv 2010

  72. [72]

    IJCV129(4), 1013–1037 (2021) 3

    Zhang, Y., Guo, X., Ma, J., Liu, W., Zhang, J.: Beyond brightening low-light images. IJCV129(4), 1013–1037 (2021) 3

    work page 2021

  73. [73]

    In: ACM MM

    Zhang, Y., Zhang, J., Guo, X.: Kindling the darkness: A practical low-light image enhancer. In: ACM MM. pp. 1632–1640 (2019) 3, 5

    work page 2019

  74. [74]

    In: WACV

    Zheng, S., Gupta, G.: Semantic-guided zero-shot learning for low-light image/video enhancement. In: WACV. pp. 581–590 (2022) 3, 4, 10, 11

    work page 2022

  75. [75]

    In: ECCV

    Zhou, S., Li, C., Change Loy, C.: Lednet: Joint low-light enhancement and deblur- ring in the dark. In: ECCV. pp. 573–589. Springer (2022) 5

    work page 2022

  76. [76]

    In: NeurIPS

    Zhu, C., Wan, R., Shi, B.: Neural transmitted radiance fields. In: NeurIPS. pp. 38994–39006 (2022) 4

    work page 2022

  77. [77]

    In: ACM MM

    Zou, W., Gao, H., Yang, W., Liu, T.: Wave-mamba: Wavelet state space model for ultra-high-definition low-light image enhancement. In: ACM MM. pp. 1534–1543 (2024) 3, 5, 10, 11, 12

    work page 2024

  78. [78]

    In: AAAI

    Zou, Y., Li, X., Jiang, Z., Liu, J.: Enhancing neural radiance fields with adaptive multi-exposure fusion: A bilevel optimization approach for novel view synthesis. In: AAAI. pp. 7882–7890 (2024) 2

    work page 2024