pith. sign in

arxiv: 2606.11841 · v1 · pith:JYUKYTY7new · submitted 2026-06-10 · 💻 cs.CV

Scene-Adaptive Nonlinear Tone Curves for Pseudo Ground-Truth Generation in Low-Light 3D Gaussian Splatting

Mingzhe Lyu , Jinqiang Cui , Hong Zhang This is my paper

Pith reviewed 2026-06-27 10:32 UTC · model grok-4.3

classification 💻 cs.CV
keywords low-light novel view synthesis3D Gaussian Splattingpseudo ground-truthtone curvesnonlinear mappingscene-adaptive enhancement
0
0 comments X

The pith

Nonlinear tone curves replace linear pseudo-GT and raise PSNR up to 4.34 dB in low-light 3DGS

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

The paper shows that uniform linear gain for pseudo ground-truth images limits 3D Gaussian Splatting quality in low light because it clips bright regions and under-enhances dark ones. It replaces that step with a scene-adaptive nonlinear tone-curve module that uses percentile-based normalisation to apply curves without per-scene tuning, adds an automatic black-level offset, and supplies two concrete curves: a bounded exponential called Adaptive SoftExp and a cubic polynomial called Adaptive Poly3. The module touches only the supervision targets and leaves the 3DGS optimisation unchanged. On three benchmarks covering 21 scenes the two curves each beat the linear baseline, with the largest gains reaching +4.34 dB on LOM and +3.25 dB on RealX3D, and the similar results from mathematically distinct curves suggest the gain is driven by nonlinearity itself.

Core claim

A scene-adaptive nonlinear tone-curve framework replaces linear pseudo-GT generation in low-light 3DGS by introducing percentile-based normalisation for scene-agnostic application, a scene-adaptive offset for black-level adjustment, and two curves (Adaptive SoftExp, a bounded exponential, and Adaptive Poly3, a data-driven cubic polynomial). Experiments on three benchmarks with 21 scenes show both curves outperform the linear baseline, with PSNR gains up to +4.34 dB on LOM and +3.25 dB on RealX3D, while the 3DGS backbone remains unchanged; similar performance between the curves indicates the benefit is curve-agnostic.

What carries the argument

Scene-adaptive nonlinear tone-curve framework using percentile-based normalisation, scene-adaptive offset, and two curves ASE (bounded exponential) and AP3 (cubic polynomial)

If this is right

  • Both curves deliver consistent PSNR gains across all tested scenes and benchmarks.
  • The improvement appears independent of curve shape because ASE and AP3 yield similar results.
  • Only the pseudo-GT computation changes; the 3DGS reconstruction pipeline stays identical.
  • The method works on 21 scenes drawn from three distinct low-light benchmarks.

Where Pith is reading between the lines

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

  • Nonlinear mappings may preserve structural detail across views better than linear gain, aiding convergence in multi-view optimisation.
  • The same percentile-normalised curve construction could be inserted into other pseudo-GT pipelines that currently rely on linear scaling.
  • Testing whether the gains persist when the curves are combined with existing 2D low-light enhancers would be a direct next experiment.

Load-bearing premise

Percentile-based normalisation lets the same curve shape be applied across scenes while still preserving the multi-view structural consistency needed for stable 3DGS optimisation.

What would settle it

Training 3DGS on a new low-light multi-view dataset with the nonlinear curves and finding equal or lower PSNR than the linear pseudo-GT baseline would falsify the central claim.

read the original abstract

Low-light novel view synthesis is challenging because dark multi-view images contain noise, weak structural detail, and compressed dynamic range. Recent 3D Gaussian Splatting (3DGS) methods address these challenges by generating pseudo ground-truth (pseudo-GT) images as supervision targets when paired normal-light references are unavailable. Existing pseudo-GT methods apply a uniform linear gain to all pixels, which clips bright regions while providing insufficient enhancement in dark regions, limiting reconstruction quality. We observe that nonlinear tone mappings, long established in 2D low-light enhancement, have not been explored for pseudo-GT generation in 3D reconstruction. Accordingly, we propose a scene-adaptive nonlinear tone-curve framework that replaces linear pseudo-GT with nonlinear alternatives. The framework introduces percentile-based normalisation for scene-agnostic curve application, a scene-adaptive offset for automatic black-level adjustment, and two complementary curves: Adaptive SoftExp (ASE), a bounded exponential curve, and Adaptive Poly3 (AP3), a data-driven cubic polynomial. The module changes only the pseudo-GT computation and leaves the 3DGS backbone unchanged. Experiments on three benchmarks covering 21 scenes show that both curves consistently outperform the linear baseline with PSNR improvements up to +4.34 dB on LOM and +3.25 dB on RealX3D. Both curves achieve similar performance despite their different mathematical forms, suggesting the improvement is curve-agnostic. Code is available at https://github.com/lvmingzhe/adaptiveToneCurve

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 / 1 minor

Summary. The paper claims that replacing linear pseudo-GT generation with a scene-adaptive nonlinear tone-curve framework (using percentile-based normalisation, a scene-adaptive offset, and either Adaptive SoftExp or Adaptive Poly3 curves) improves low-light 3D Gaussian Splatting quality while leaving the 3DGS backbone unchanged. Experiments across three benchmarks (21 scenes) report consistent PSNR gains over the linear baseline, up to +4.34 dB on LOM and +3.25 dB on RealX3D, with code released.

Significance. If the multi-view consistency assumption holds, the work offers a lightweight, backbone-agnostic improvement to low-light novel view synthesis by leveraging established 2D nonlinear tone mappings for pseudo-GT. The code release and the observation that two mathematically distinct curves yield similar gains are strengths that support reproducibility and suggest the benefit is curve-agnostic rather than form-specific.

major comments (2)
  1. [Abstract] Abstract (framework components paragraph): The percentile-based normalisation is presented as enabling 'scene-agnostic curve application,' but the text does not state whether the percentile thresholds are derived from the joint distribution across all views of a scene or computed independently per image. Per-image computation would produce view-dependent mappings, directly threatening the multi-view structural consistency required for stable 3DGS optimization and contradicting the claim that 'the module changes only the pseudo-GT computation.'
  2. [Abstract] Abstract (experiments paragraph): The reported PSNR gains (+4.34 dB on LOM, +3.25 dB on RealX3D) are given without reference to error bars, number of random seeds, or explicit baseline implementations, and no view-consistency metrics (e.g., cross-view intensity variance after tone mapping) are mentioned. These omissions make it impossible to verify that the gains are robust and that the central consistency assumption is satisfied.
minor comments (1)
  1. [Abstract] The abstract refers to 'RealX3D' without a clarifying citation or expansion; a standard reference or dataset name should be supplied for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract (framework components paragraph): The percentile-based normalisation is presented as enabling 'scene-agnostic curve application,' but the text does not state whether the percentile thresholds are derived from the joint distribution across all views of a scene or computed independently per image. Per-image computation would produce view-dependent mappings, directly threatening the multi-view structural consistency required for stable 3DGS optimization and contradicting the claim that 'the module changes only the pseudo-GT computation.'

    Authors: The percentile thresholds are derived from the joint distribution across all views of each scene (not per-image) to ensure a single consistent mapping per scene. This preserves multi-view structural consistency while enabling scene-adaptive application. We will explicitly state this computation detail in the revised abstract and methods section. revision: yes

  2. Referee: [Abstract] Abstract (experiments paragraph): The reported PSNR gains (+4.34 dB on LOM, +3.25 dB on RealX3D) are given without reference to error bars, number of random seeds, or explicit baseline implementations, and no view-consistency metrics (e.g., cross-view intensity variance after tone mapping) are mentioned. These omissions make it impossible to verify that the gains are robust and that the central consistency assumption is satisfied.

    Authors: We will revise the abstract to reference standard deviations across the 21 scenes and clarify the linear baseline implementation. In the experiments section we will add explicit details on random seeds (where applicable) and include a new view-consistency analysis using cross-view intensity variance after tone mapping to directly support the consistency assumption. revision: partial

Circularity Check

0 steps flagged

No circularity; empirical gains from nonlinear pseudo-GT curves are independent of inputs

full rationale

The paper introduces percentile-based normalisation, scene-adaptive offset, ASE and AP3 curves for pseudo-GT generation in low-light 3DGS, claiming only the pseudo-GT step changes while the backbone remains fixed. Reported PSNR gains (+4.34 dB on LOM, +3.25 dB on RealX3D) are presented as experimental outcomes on 21 scenes against a linear baseline, with code released. No equations, fitted parameters renamed as predictions, or self-citation chains appear in the provided text that would make the central claim reduce to its own inputs by construction. The framework is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 0 axioms · 0 invented entities

Abstract-only review; the framework introduces adaptation parameters whose selection and potential fitting are not detailed, plus the two curve forms themselves.

free parameters (2)
  • percentile thresholds for normalisation
    Used to achieve scene-agnostic curve application; values chosen from image statistics but not specified as fixed or fitted.
  • scene-adaptive offset
    Automatic black-level adjustment parameter per scene.

pith-pipeline@v0.9.1-grok · 5817 in / 1329 out tokens · 37002 ms · 2026-06-27T10:32:45.029137+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

53 extracted references · 28 canonical work pages

  1. [1]

    CRC Press, https://doi.org/10

    Banterle F, Artusi A, Debattista K, et al (2017) Advanced High Dynamic Range Imag- ing, 2nd edn. CRC Press, https://doi.org/10. 1201/9781315119526

    2017

  2. [3]

    In: Pro- ceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp 12504–12513, https://doi.org/10.1109/ ICCV51070.2023.01149

    Cai Y, Bian H, Lin J, et al (2023) Retinex- former: One-stage retinex-based transformer for low-light image enhancement. In: Pro- ceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp 12504–12513, https://doi.org/10.1109/ ICCV51070.2023.01149

    arXiv 2023

  3. [4]

    The Visual Computer 41:5815–5833

    Cao Z, Wang Z, Zhang Y, et al (2025) Enhancing 3d gaussian splatting for low- quality images: Semantically guided training and unsupervised quality assessment. The Visual Computer 41:5815–5833. https://doi. org/10.1007/s00371-024-03754-z

    work page doi:10.1007/s00371-024-03754-z 2025

  4. [5]

    In: Proceedings of the AAAI Conference on Artificial Intel- ligence, pp 2367–2375, https://doi.org/10

    Chen Z, Lu Z, Ma D, et al (2025) EvHDR- GS: Event-guided HDR video reconstruction with 3D gaussian splatting. In: Proceedings of the AAAI Conference on Artificial Intel- ligence, pp 2367–2375, https://doi.org/10. 1609/aaai.v39i3.32237

    2025

  5. [6]

    In: Proceedings of the British Machine Vision Conference (BMVC), p 238

    Cui Z, Li K, Gu L, et al (2022) You only need 90k parameters to adapt light: A light weight transformer for image enhancement and exposure correction. In: Proceedings of the British Machine Vision Conference (BMVC), p 238

    2022

  6. [9]

    The Visual Computer 41:9243–9255

    Dong H, Cheng S, Fang T, et al (2025) DED-Net: Multi-scale fusion and illumination-guided enhance- ment for low-light image restoration. The Visual Computer 41:9243–9255. https://doi.org/10.1007/s00371-025-03924-7

    work page doi:10.1007/s00371-025-03924-7 2025

  7. [10]

    In: Computer Graphics Forum, pp 419–426, https://doi

    Drago F, Myszkowski K, Annen T, et al (2003) Adaptive logarithmic mapping for dis- playing high contrast scenes. In: Computer Graphics Forum, pp 419–426, https://doi. org/10.1111/1467-8659.00689

    work page doi:10.1111/1467-8659.00689 2003

  8. [11]

    In: ACM SIGGRAPH 2002 Papers, pp 257–266, https://doi.org/10.1145/566570

    Durand F, Dorsey J (2002) Fast bilateral fil- tering for the display of high-dynamic-range images. In: ACM SIGGRAPH 2002 Papers, pp 257–266, https://doi.org/10.1145/566570. 566574

    work page doi:10.1145/566570 2002

  9. [12]

    HaoChen, J

    Guo C, Li C, Guo J, et al (2020) Zero- reference deep curve estimation for low-light image enhancement. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 1780– 1789, https://doi.org/10.1109/CVPR42600. 2020.00185

    work page doi:10.1109/cvpr42600 2020

  10. [13]

    Game Developers Conference (GDC)

    Hable J (2010) Uncharted 2: HDR lighting. Game Developers Conference (GDC)

    2010

  11. [14]

    2d gaussian splatting for geometrically accurate radiance fields,

    Huang B, Yu Z, Chen A, et al (2024) 2D gaussian splatting for geometrically accurate radiance fields. In: ACM SIGGRAPH 2024 Conference Papers, pp 1–11, https://doi.org/ 10.1145/3641519.3657428

    work page doi:10.1145/3641519.3657428 2024

  12. [16]

    In: Computer Vision – ECCV 2024

    Jiang H, Luo A, Liu X, et al (2024) Ligh- tenDiffusion: Unsupervised low-light image enhancement with latent-retinex diffusion 13 models. In: Computer Vision – ECCV 2024. Springer, pp 161–179, https://doi.org/10. 1007/978-3-031-73195-2_10

    2024

  13. [17]

    IEEE Transactions on Image Processing 30:2340–2349

    Jiang Y, Gong X, Liu D, et al (2021) Enlight- enGAN: Deep light enhancement without paired supervision. IEEE Transactions on Image Processing 30:2340–2349. https://doi. org/10.1109/TIP.2021.3051462

    work page doi:10.1109/tip.2021.3051462 2021

  14. [18]

    3d gaussian splatting for real-time radiance field rendering,

    Kerbl B, Kopanas G, Leimkühler T, et al (2023) 3d gaussian splatting for real-time radiance field rendering. ACM Transac- tions on Graphics (TOG) 42(4):139:1–139:14. https://doi.org/10.1145/3592433

    work page doi:10.1145/3592433 2023

  15. [19]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 19096–19106

    Kumar A, Ambasamduram RN (2026) l2dgs: Low-light dynamic gaussian splatting. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 19096–19106

    2026

  16. [20]

    IEEE Transactions on Pattern Analysis and Machine Intelli- gence 44(8):4225–4238

    Li C, Guo C, Loy CC (2022) Learning to enhance low-light image via zero-reference deep curve estimation. IEEE Transactions on Pattern Analysis and Machine Intelli- gence 44(8):4225–4238. https://doi.org/10. 1109/TPAMI.2021.3063604

    arXiv 2022

  17. [21]

    arXiv preprint arXiv:251223437 https://doi

    Liu S, Bao C, Cui Z, et al (2025) RealX3D: A physically-degraded 3d benchmark for multi- view visual restoration and reconstruction. arXiv preprint arXiv:251223437 https://doi. org/10.48550/arXiv.2512.23437

    work page doi:10.48550/arxiv.2512.23437 2025

  18. [22]

    In: Advances in Neural Information Processing Systems (NeurIPS)

    Liu S, Gu L, Cui Z, et al (2025) I2- NeRF: Learning neural radiance fields under physically-grounded media interactions. In: Advances in Neural Information Processing Systems (NeurIPS)

    2025

  19. [23]

    The Visual Computer 42(3)

    Liu S, Xu X, Liu Y, et al (2026) Components- balanced guided diffusion for retinex-based low-light image enhancement. The Visual Computer 42(3). https://doi.org/10.1007/ s00371-026-04366-5

    2026

  20. [24]

    Com- puters & Graphics 128:104209

    Liu Z, Meng Q, Lu D, et al (2025) GloNeRF: Boosting NeRF capabilities and multi-view consistency in low-light environments. Com- puters & Graphics 128:104209. https://doi. org/10.1016/j.cag.2025.104209

    work page doi:10.1016/j.cag.2025.104209 2025

  21. [25]

    URL https://proceedings.mlr

    Lu T, Yu M, Xu L, et al (2024) Scaffold-GS: Structured 3D gaussians for view-adaptive rendering. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 20654–20664, https: //doi.org/10.1109/CVPR52733.2024.01952

    work page doi:10.1109/cvpr52733.2024.01952 2024

  22. [26]

    Price, I., Sanchez-Gonzalez, A., Alet, F., Ewalds, T., El- Kadi, A., Stott, J., Mohamed, S., Battaglia, P

    Ma L, Ma T, Liu R, et al (2022) Toward fast, flexible, and robust low-light image enhance- ment. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 5637–5646, https: //doi.org/10.1109/CVPR52688.2022.00555

    work page doi:10.1109/cvpr52688.2022.00555 2022

  23. [27]

    In: ACM SIG- GRAPH 2008 Papers, pp 1–10, https://doi

    Mantiuk R, Daly S, Kerofsky L (2008) Dis- play adaptive tone mapping. In: ACM SIG- GRAPH 2008 Papers, pp 1–10, https://doi. org/10.1145/1399504.1360667

    work page doi:10.1145/1399504.1360667 2008

  24. [28]

    NeRF: Representing scenes as neural radiance fields for view synthesis,

    Mildenhall B, Srinivasan PP, Tancik M, et al (2020) NeRF: Representing scenes as neu- ral radiance fields for view synthesis. In: European Conference on Computer Vision (ECCV). Springer, pp 405–421, https://doi. org/10.1007/978-3-030-58452-8_24

    work page doi:10.1007/978-3-030-58452-8_24 2020

  25. [29]

    Price, I., Sanchez-Gonzalez, A., Alet, F., Ewalds, T., El- Kadi, A., Stott, J., Mohamed, S., Battaglia, P

    Mildenhall B, Hedman P, Martin-Brualla R, et al (2022) Nerf in the dark: High dynamic range view synthesis from noisy raw images. In: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recog- nition (CVPR), pp 16169–16178, https://doi. org/10.1109/CVPR52688.2022.01571

    work page doi:10.1109/cvpr52688.2022.01571 2022

  26. [30]

    The Visual Computer 40:6779–6795

    Pan H, Gao B, Wang X, et al (2024) DICNet: Achieve low-light image enhancement with image decomposition, illumination enhance- ment, and color restoration. The Visual Computer 40:6779–6795. https://doi.org/10. 1007/s00371-024-03262-0

    2024

  27. [31]

    In: Advances in Neural Information Processing Systems (NeurIPS), https://doi.org/10.52202/079017-3486

    Qu Z, Xu K, Hancke GP, et al (2024) LuSh- NeRF: Lighting up and sharpening NeRFs for low-light scenes. In: Advances in Neural Information Processing Systems (NeurIPS), https://doi.org/10.52202/079017-3486

    work page doi:10.52202/079017-3486 2024

  28. [32]

    ACM Transactions on Graph- ics (TOG) 21(3):267–276

    Reinhard E, Stark M, Shirley P, et al (2002) Photographic tone reproduction for digital images. ACM Transactions on Graph- ics (TOG) 21(3):267–276. https://doi.org/ 10.1145/566654.566575

    work page doi:10.1145/566654.566575 2002

  29. [33]

    The Visual Computer 40:8481–8499

    Shi Y, Liu D, Zhang L, et al (2024) MaCo: Efficient unsupervised low-light image enhancement via illumination-based magnitude control. The Visual Computer 40:8481–8499. https://doi.org/10.1007/ s00371-023-03249-3

    2024

  30. [34]

    The Visual Com- puter 40:121–136

    Singh K, Parihar AS (2024) Illumina- tion estimation for nature preserving low- light image enhancement. The Visual Com- puter 40:121–136. https://doi.org/10.1007/ s00371-023-02770-9 14

    2024

  31. [35]

    Zhai, X., Mustafa, B., Kolesnikov, A., and Beyer, L

    Wang H, Xu X, Xu K, et al (2023) Light- ing up NeRF via unsupervised decompo- sition and enhancement. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp 12632– 12641, https://doi.org/10.1109/ICCV51070. 2023.01161

    work page doi:10.1109/iccv51070 2023

  32. [36]

    In: 2025 IEEE International Conference on Robotics and Automation (ICRA), https://doi.org/10

    Wang H, Huang J, Yang L, et al (2025) LLGS: Unsupervised gaussian splatting for image enhancement and reconstruction in pure dark environment. In: 2025 IEEE International Conference on Robotics and Automation (ICRA), https://doi.org/10. 1109/ICRA55743.2025.11127924

    arXiv 2025

  33. [37]

    Anomalygpt: Detecting industrial anomalies using large vision-language models

    Wang M, Huang X, Zhou G, et al (2025) Bright-NeRF: Brightening neural radiance field with color restoration from low- light RA W images. In: Proceedings of the AAAI Conference on Artificial Intelligence, pp 7817–7825, https://doi.org/10.1609/aaai. v39i8.32842

    work page doi:10.1609/aaai 2025

  34. [38]

    The Visual Computer 41:3867–3882

    Wang S, Wu G, Wang J, et al (2025) SBC- Net: Semantic-guided brightness curve esti- mation network for low-light image enhance- ment. The Visual Computer 41:3867–3882. https://doi.org/10.1007/s00371-024-03636-4

    work page doi:10.1007/s00371-024-03636-4 2025

  35. [39]

    In: Proceedings of the AAAI Confer- ence on Artificial Intelligence, pp 2654–2662, https://doi.org/10.1609/aaai.v37i3.25364

    Wang T, Zhang K, Shen T, et al (2023) Ultra- high-definition low-light image enhance- ment: A benchmark and transformer-based method. In: Proceedings of the AAAI Confer- ence on Artificial Intelligence, pp 2654–2662, https://doi.org/10.1609/aaai.v37i3.25364

    work page doi:10.1609/aaai.v37i3.25364 2023

  36. [40]

    IEEE Transactions on Image Processing 13(4), 600–612 (Apr 2004)

    Wang Z, Bovik AC, Sheikh HR, et al (2004) Image quality assessment: From error visi- bility to structural similarity. IEEE Trans- actions on Image Processing 13(4):600–612. https://doi.org/10.1109/TIP.2003.819861

    work page doi:10.1109/tip.2003.819861 2004

  37. [41]

    In: British Machine Vision Conference (BMVC)

    Wei C, Wang W, Yang W, et al (2018) Deep retinex decomposition for low-light enhance- ment. In: British Machine Vision Conference (BMVC)

    2018

  38. [42]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 5891–5900, https://doi.org/10

    Wu W, Weng J, Zhang P, et al (2022) URetinex-Net: Retinex-based deep unfolding network for low-light image enhancement. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 5891–5900, https://doi.org/10. 1109/CVPR52688.2022.00581

    arXiv 2022

  39. [43]

    Price, I., Sanchez-Gonzalez, A., Alet, F., Ewalds, T., El- Kadi, A., Stott, J., Mohamed, S., Battaglia, P

    Xu X, Wang R, Fu CW, et al (2022) SNR- aware low-light image enhancement. In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 17693–17703, https://doi.org/ 10.1109/CVPR52688.2022.01719

    work page doi:10.1109/cvpr52688.2022.01719 2022

  40. [44]

    URL https://proceedings.mlr

    Yang Z, Gao X, Zhou W, et al (2024) Deformable 3D gaussians for high-fidelity monocular dynamic scene reconstruction. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 20331–20341, https://doi.org/ 10.1109/CVPR52733.2024.01922

    work page doi:10.1109/cvpr52733.2024.01922 2024

  41. [45]

    Computer Graph- ics Forum 43(7):e15213

    Ye S, Dong ZH, Hu Y, et al (2024) Gaussian in the dark: Real-time view synthesis from inconsistent dark images using gaussian splatting. Computer Graph- ics Forum 43(7):e15213. https://doi.org/10. 1111/cgf.15213

    2024

  42. [46]

    IEEE Transactions on Pattern Analysis and Machine Intelligence 47(8):6823–6841

    Yi X, Xu H, Zhang H, et al (2025) Diff- Retinex++: Retinex-driven reinforced diffu- sion model for low-light image enhancement. IEEE Transactions on Pattern Analysis and Machine Intelligence 47(8):6823–6841. https: //doi.org/10.1109/TPAMI.2025.3563612

    work page doi:10.1109/tpami.2025.3563612 2025

  43. [47]

    arXiv preprint arXiv:260424053

    Yin Y, Wang Z, Zhang Y, et al (2026) Light ’em up: Enabling few-shot low- light 3D gaussian splatting with multi- scale explicit retinex illumination decoupling. arXiv preprint arXiv:260424053

    2026

  44. [48]

    The Visual Computer 39:4165–4175

    Yu X, Li H, Yang H (2023) Two-stage image decomposition and color regulator for low-light image enhancement. The Visual Computer 39:4165–4175. https://doi.org/10. 1007/s00371-022-02582-3

    2023

  45. [49]

    URL https://proceedings.mlr

    Yu Z, Chen A, Huang B, et al (2024) Mip- Splatting: Alias-free 3d gaussian splatting. In: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recog- nition (CVPR), pp 19447–19456, https://doi. org/10.1109/CVPR52733.2024.01839

    work page doi:10.1109/cvpr52733.2024.01839 2024

  46. [50]

    In: European Conference on Computer Vision (ECCV), pp 492–511, https://doi.org/10

    Zamir SW, Arora A, Khan SH, et al (2020) Learning enriched features for real image restoration and enhancement. In: European Conference on Computer Vision (ECCV), pp 492–511, https://doi.org/10. 1007/978-3-030-58595-2_30

    2020

  47. [51]

    The Visual Com- puter 40:109–120

    Zhang J, Wang H, Wu X, et al (2024) Invertible network for unpaired low-light 15 image enhancement. The Visual Com- puter 40:109–120. https://doi.org/10.1007/ s00371-023-02769-2

    2024

  48. [52]

    In: 2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018

    Zhang R, Isola P, Efros AA, et al (2018) The unreasonable effectiveness of deep features as a perceptual metric. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp 586–595, https://doi.org/10.1109/CVPR.2018.00068

    work page doi:10.1109/cvpr.2018.00068 2018

  49. [53]

    In: 2024 IEEE/RSJ Inter- national Conference on Intelligent Robots and Systems (IROS)

    Zhang T, Huang K, Zhi W, et al (2024) DarkGS: Learning neural illumination and 3D gaussians relighting for robotic explo- ration in the dark. In: 2024 IEEE/RSJ Inter- national Conference on Intelligent Robots and Systems (IROS). IEEE, pp 12864– 12871, https://doi.org/10.1109/IROS58592. 2024.10802684

    work page doi:10.1109/iros58592 2024

  50. [54]

    The Visual Computer 42(3)

    Zhang X, Zhang L (2026) Enhancing low- light image quality: A multi-granularity memory approach with RetinexMemory. The Visual Computer 42(3). https://doi.org/10. 1007/s00371-026-04375-4

    2026

  51. [55]

    In: Proceedings of the 27th ACM International Conference on Multime- dia, pp 1632–1640, https://doi.org/10.1145/ 3343031.3350926

    Zhang Y, Zhang J, Guo X (2019) Kin- dling the darkness: A practical low-light image enhancer. In: Proceedings of the 27th ACM International Conference on Multime- dia, pp 1632–1640, https://doi.org/10.1145/ 3343031.3350926

    arXiv 2019

  52. [56]

    International Journal of Computer Vision 129(4):1013–1037

    Zhang Y, Guo X, Ma J, et al (2021) Beyond brightening low-light images. International Journal of Computer Vision 129(4):1013–1037. https://doi.org/10.1007/ s11263-020-01407-x

    2021

  53. [57]

    com- press highlights, lift midtones

    Zhou H, Dong W, Chen J (2025) LITA- GS: Illumination-agnostic novel view syn- thesis via reference-free 3d gaussian splat- ting and physical priors. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp 21580–21589, https://doi.org/10.1109/ CVPR52734.2025.02010 A Full Curve Family Search This appendix reports the...

    arXiv 2025