Pith. sign in

REVIEW 3 major objections 7 minor 42 references

U2NeRF: Unsupervised Underwater Image Restoration and Neural Radiance Fields

T0 review · 3 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A transformer-based radiance field, trained without clean reference images, renders and restores underwater views by decomposing each predicted patch into scene radiance, transmission maps, and background light.

desk verdict A plausible engineering combo—transformer NeRF plus UPIFM-style physics losses—with a useful new dataset, but the restoration claim rests on an unverified decomposition and metrics that mirror the training losses. read the letter →

arxiv 2411.16172 v1 pith:MXGG5QAX submitted 2024-11-25 cs.CV

Vinayak Gupta , Manoj S , Mukund Varma T , Kaushik Mitra This is my paper · ORCID
classification cs.CV
keywords underwaterimagerestorationneuralradiancefieldsnovelviewsynthesisself-supervisedlearningformationmodelmulti-viewgeometrytransformer
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper asks whether a neural radiance field can perform underwater image restoration at the same time as novel view synthesis, using no clean ground-truth images at all. The authors adapt a generalizable transformer-based NeRF to predict image patches rather than single pixel colors, and decompose each patch through a physics-based underwater image formation model into scene radiance, two transmission maps, and global background light; the decomposition is trained only by reconstructing the original degraded views plus heuristic regularizers. On their new 12-scene Underwater View Synthesis benchmark, the resulting U2NeRF outperforms both rendering-plus-restoration baselines and restoration-only methods on perceptual and color-quality metrics, by up to 11% in LPIPS and 5% in UIQM on average. The broader claim is that multi-view geometry and spatial context can substitute for supervision, baking physical decomposition into the network implicitly.

What carries the argument

The load-bearing object is the disentangled image formation model of Eq. 5, in which the observed underwater color $I$ at a pixel is written as $J \cdot T_D + (1 - T_B) \cdot A$, with $J$ scene radiance, $T_D$ direct transmission, $T_B$ backscatter transmission, and $A$ global background light. Around this, the method builds on a generalizable NeRF transformer architecture from the paper's reference [30]: a view transformer aggregates multi-view image features along epipolar lines, a ray transformer composes point features along a ray, and the final ray feature is upsampled by convolutions to emit a $p \times p$ patch for each component through separate output heads, while a small VAE estimates $A$. The reconstruction loss on recombining the components is the only direct supervision; the other five losses act as regularizers that guide the factorization toward physically meaningful values.

What would settle it

On a scene with available clean reference views, compare the network's predicted scene radiance patch against the true clean image of the same view; if the predicted radiance is not substantially closer to the clean image than the degraded input is, while still satisfying the self-supervised losses, the restoration claim is not supported.

Watch

Extended reading notes

Core claim

The central discovery is that unsupervised restoration and rendering can be achieved in one network if the radiance field predicts a local image patch instead of a point color, and if that patch is forced through the underwater image formation model $I(i) = J(i)T_D(i) + (1 - T_B(i))A$. With the patch providing spatial context and the formation model providing structure, the self-supervised reconstruction loss plus lightweight regularizers (contrast, gray-world color constancy, backscatter consistency, smooth global light) are sufficient to separate scene radiance $J$ from attenuation and backscatter. Visualizations show that the estimated transmission maps track scene depth and the background light converges to the brightest region, indicating physical grounding. The method generalizes across scenes after pretraining and, with a short finetune, matches or beats a per-scene vanilla NeRF.

Load-bearing premise

The method assumes that the reconstruction loss plus the heuristic regularizers push the network to a physically correct factorization of the degraded image into scene radiance, transmission, and background light, even though infinitely many decompositions satisfy the same image-formation equation.

Editorial extensions

If this is right

  • Joint rendering and restoration requires no clean data: a single underwater scene's posed views are enough to train U2NeRF end-to-end.
  • Multi-view geometry helps restoration: U2NeRF beats restoration-only baselines that assume direct access to the target view, because the geometry gives consistent scene information across views.
  • Spatial context is what makes unsupervised restoration work in a radiance field: pixel-level NeRF output cannot restore, while predicting patches of size 4 strikes the best balance.
  • The same pretrained network generalizes to unseen underwater scenes, and fine-tuning for only 50k steps suffices to match or exceed a per-scene vanilla NeRF.
  • Even without explicit denoising training, patch prediction gives some robustness to Gaussian noise in the input views.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension is to verify on synthetic scenes with known clean references whether the predicted scene radiance $J$ itself matches the true clean image, not just the final recombined output, which would confirm that the factorization is physically faithful.
  • If the decomposition is truly physical, the same patch-based self-supervision recipe should transfer to other scattering media such as haze, fog, or turbid water by swapping the formation model, and to video by enforcing temporal consistency.
  • The released benchmark with easy, medium, and hard splits gives the community a way to compare joint rendering-restoration methods, and the easy split's known clean references allow future work to test physical correctness directly.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. The paper proposes U2NeRF, an extension of the generalizable NeRF transformer (GNT) that simultaneously renders novel views and restores underwater images. Instead of predicting single pixel colors, U2NeRF predicts image patches and decomposes each patch into scene radiance J, direct transmission TD, backscatter transmission TB, and global background light A, combining them via the image formation model I(i) = J(i)TD(i) + (1 − TB(i))A (Eq. 5). Training is fully self-supervised using a photometric reconstruction loss plus five regularizers (contrast, color constancy, KL, transmission consistency, global smoothness) largely borrowed from the UPIFM baseline [4]. The authors introduce the UVS dataset with 12 scenes split into easy (synthetic), medium (real-world high quality), and hard (real-world low quality) splits, and report single-scene and cross-scene experiments against rendering-only and restoration-only baselines. The central claims are simultaneous rendering and restoration with improvements in LPIPS, UIQM, and UCIQE.

Significance. If the decomposition in Eq. 5 were identifiable and the evaluations sound, U2NeRF would be a meaningful step toward combining multi-view geometry with unsupervised underwater restoration, and the UVS dataset would be a useful benchmark. The paper has potentially valuable components: a novel dataset with SfM-calibrated real underwater scenes, an architecture that predicts patches rather than pixels to enable restoration, and an attempt at a physics-informed self-supervised NeRF. However, as presented, the central restoration claim is not well supported by the reported tables, and the identifiability of the decomposition from the self-supervised losses is not established. The significance of the contribution therefore depends on additional controlled experiments that are not currently in the paper.

major comments (3)
  1. [§4.4, Tables 1a–1c, Tables 5–6] The claims of superior restoration quality are also not supported on the only split with ground truth, which is a load-bearing issue for the paper's central thesis.
  2. [§3.2, Eq. (5), Appendix A, §4.7] This is a load-bearing issue because the paper's stated contribution is simultaneous rendering and restoration; if the factorization is arbitrary, the 'restored' output J is just a color-adjusted reconstruction, not the true scene radiance.
  3. [§4.1 (Metrics), §4.4, Appendix A (Eqs. 8–9)] This concern is specific to the real-scene evaluation, which is the only setting where the paper claims both rendering and restoration improvements.
minor comments (7)
  1. [§4.1 vs Appendix A] The total-loss weights are inconsistent: Section 4.1 sets λ6 = 0.1, while Appendix A sets λ6 = 1. Please harmonize these values and state which were used for the reported results.
  2. [Appendix B] The text refers to 'Fig. 8 and Fig. ??' in Appendix B; the unresolved reference 'Fig. ??' should be fixed.
  3. [§4.2] The phrase '12 scenes, equally split into easy (synthetic), medium (real-world high quality), hard (real-world low quality)' is awkward because the splits are 4+4+4, not 'equal' in a ternary sense; consider rewording to 'split into three groups of four scenes each'.
  4. [Eq. (6) and Eq. (7)] The reconstruction loss is denoted Lrec in Eq. (6) but LRec in Eq. (7); please use a single notation consistently throughout.
  5. [§4.1, Table 1b] The description of LPIPS (gray) is unclear: 'LPIPS scores between the gray scale image of rendered and restored target views using [7] (to remove color differences)' — please specify whether the images are converted to grayscale before computing LPIPS or processed by UIESS first, and what exactly [7] is used for.
  6. [§4.6, Fig. 6] The denoising experiment (Section 4.6, Fig. 6) is qualitative only; reporting PSNR or other metrics on noisy inputs would strengthen the claim.
  7. [§3.2] The explanation of the VAE for estimating A is brief; it is not clear how the global background light estimated from the nearest source image is made consistent across different target views, especially since Lglob only enforces spatial smoothness within a single prediction.

Circularity Check

2 steps flagged · score 6.0 of 10

Restoration gains on real scenes are the training losses restated as no-reference metrics; the Eq. 5 decomposition is unidentifiable, so the 'restored' J is not shown to be physical scene radiance.

  1. fitted input called prediction [Sec. 3.2 (loss definitions and Eq. 6); Sec. 4.4 (real-scene results); Sec. 4.1 (Metrics)]
    "we (1) minimize the difference between encoded feature z and latent code sampled from Gaussian ˆz in the vae ( Lkl), (2) minimize the difference between the saturation and brightness of the predicted scene radiance to reduce haze ( Lcon), (3) minimize the potential color deviations in the scene radiance( Lcol), (4) ensure constant back-scatter coefficients ( Ltrans) across channels, (5) enforce constant global background light by minimizing variance within each local neighbourhood (Lglob) as proposed in the original paper [4]."

    The paper's real-scene restoration claim is measured with UIQM and UCIQE, no-reference metrics that reward contrast, saturation, and colorfulness. Those are exactly the qualities that Lcon and Lcol explicitly enforce: Lcon drives the saturation-brightness difference of J to zero (a haze/contrast proxy) and Lcol drives each channel mean to 0.5 (a color-balance/colorfulness proxy). Thus the reported 'UIQM ↑5%, UCIQE ↑4%' improvement is, by construction, a restatement of the training objective under an evaluation name, not independent evidence that J equals true scene radiance. This is corroborated on the easy split with ground truth, where U2NeRF (LPIPS 0.230) is worse than the no-rendering restorer UIESS (0.174).

  2. self definitional [Sec. 3.2, Eq. 5; Appendix A.1]
    "I(i) = J(i)TD(i) + (1 − TB(i))A (5) ... This enables our network to be trained in a fully self-supervised manner in the absence of ground truth image. ... Together, the network is trained to optimize: L = λ1Lrec, +λ2Lcon + λ3Lcol + λ4Lkl +λ5Ltrans + λ6Lglob"

    For fixed observed I and a chosen A, Eq. 5 is one equation in unknown per-pixel fields J, TD, and TB; infinitely many triples satisfy the same reconstruction. The regularizers are heuristic priors inherited from the same [4] paper used as a baseline, and the paper gives no identifiability proof and no constraint tying either transmission map to scene depth. Therefore 'scene radiance' J is not recovered from physical measurement; it is defined as the network output that minimizes reconstruction plus these heuristics. The assertion that U2NeRF 'implicitly learns to generate physically meaningful image components' (Sec. 4.7) is a self-description of the factorization, not an externally verified restoration.

full rationale

The rendering half of U2NeRF is largely independent: the GNT-style multi-view transformer is benchmarked with rendering baselines, and the architectural self-citation to [30] is not load-bearing evidence. The UVS dataset contribution is also non-circular. However, the restoration half of the central claim is not established independently. Eq. 5 defines J, TD, and TB through a self-supervised reconstruction that is unidentifiable without additional physically grounded constraints, and the regularizers that select a factorization are imported from the same UPIFM paper [4] that serves as a baseline. On real scenes, the reported UIQM/UCIQE gains reward color and contrast properties that the training losses explicitly optimize, so the evaluation reduces, at least partially, to the fitted objective. The synthetic easy split, where ground truth exists, shows U2NeRF trailing UIESS (0.230 vs 0.174 LPIPS), further weakening the restoration claim. Overall score 6: one or more 'predictions' reduce by construction, while the rendering and dataset contributions retain independent content.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The method borrows its image formation model and all regularization losses from UPIFM [4] and its rendering architecture from GNT [30], with one overlapping author on GNT. The central restoration claim rests on an unproven identifiability assumption, hyperparameter choices, and heuristic losses. No new physical entities are introduced.

free parameters (5)
  • Loss weights lambda1..lambda6 = 1, 0.1, 1, 1, 0.1, 0.1 in Sec. 4.1; 1, 0.1, 1, 1, 0.1, 1 in Appendix A.6
    Hand-set via unspecified search; the inconsistency between main text and appendix means the exact configuration is ambiguous.
  • Patch size p = 4
    Chosen from an ablation on the starfish scene only; controls the spatial context available for restoration.
  • Number of transformer blocks = 4 for single scene, 8 for generalization
    Architectural choice with no ablation on block count; affects model capacity.
  • Number of source views N = Sampled from 8-12 during training; tested at 3, 5, 8, 10 in ablation
    Performance varies slightly with N; no principled selection criterion is provided.
  • Coarse points per ray = 192
    Set to compare with GNT; no ablation reported for this value.
assumptions (4)
  • standard math Volume rendering integral (Eq. 1) and GNT attention-based rendering (Eqs. 3-4) correctly model light transport in the scene.
    Background adopted from NeRF and GNT without proof; standard in the field.
  • domain assumption Underwater image formation model I = J*TD + (1-TB)*A (Eq. 5) with global background light A constant and backscatter coefficient constant across channels.
    Adopted from UPIFM [4]; the physical model is an approximation and underdetermines the decomposition.
  • ad hoc to paper The heuristic losses Lcon, Lcol, Ltrans, and Lglob from [4] are sufficient to separate scene radiance from degradation in the absence of ground truth.
    No proof is given that minimizing Eq. 6 recovers physical scene radiance; this is the key untested premise.
  • domain assumption Scenes are static except for acknowledged motion in plants; all views share one geometry.
    NeRF and SfM assume a static scene; the paper's own limitation section notes blur around moving plants.

how reviews work

0 comments
Cite this review

Pith. "Pith review of U2NeRF: Unsupervised Underwater Image Restoration and Neural Radiance Fields." pith.science (2026). https://pith.science/paper/MXGG5QAX

@misc{pith2026241116172,
  author       = {Pith},
  title        = {Pith review of: U2NeRF: Unsupervised Underwater Image Restoration and Neural Radiance Fields},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MXGG5QAX}},
  note         = {Machine review of arXiv:2411.16172}
}
read the original abstract

Underwater images suffer from colour shifts, low contrast, and haziness due to light absorption, refraction, scattering and restoring these images has warranted much attention. In this work, we present Unsupervised Underwater Neural Radiance Field U2NeRF, a transformer-based architecture that learns to render and restore novel views conditioned on multi-view geometry simultaneously. Due to the absence of supervision, we attempt to implicitly bake restoring capabilities onto the NeRF pipeline and disentangle the predicted color into several components - scene radiance, direct transmission map, backscatter transmission map, and global background light, and when combined reconstruct the underwater image in a self-supervised manner. In addition, we release an Underwater View Synthesis UVS dataset consisting of 12 underwater scenes, containing both synthetically-generated and real-world data. Our experiments demonstrate that when optimized on a single scene, U2NeRF outperforms several baselines by as much LPIPS 11%, UIQM 5%, UCIQE 4% (on average) and showcases improved rendering and restoration capabilities. Code will be made available upon acceptance.

Figures

Figures reproduced from arXiv: 2411.16172 by the authors.

Figure 1
Figure 1. Unlike standalone methods like Radiance Fields [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of U2NeRF: 1) Identify source views for a given target view, 2) Extract features for epipolar points using a trainable [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Qualitative results for single-scene rendering. In the Debris scene (row-1), U2NeRF is able to successfully recover and restore [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Qualitative results for cross-scene rendering. We visualize the underwater scene (row-1), novel views rendered using the pre [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Illustrative examples of scenes from the UVS Dataset, [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Denoising results of U2NeRF trained model on the starfish scene but with fewer source views. From Table. 4, we can see that as the number of source views increase, the model performs better. How￾ever, there is almost no significant drop in performance even when only 3 …
Figure 7
Figure 7. Figure 7: Visualisations of the predicted image components (scene radiance, transmission maps, global light). [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Qualitative results on single-scene rendering for Medium and Hard Scenes. The top 4 rows depict the scenes from the hard [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 24 canonical work pages

  1. [4]

    Unsupervised and untrained underwater im- age restoration based on physical image formation model

    Shu Chai, Zhenqi Fu, Yue Huang, Xiaotong Tu, and Xing- hao Ding. Unsupervised and untrained underwater im- age restoration based on physical image formation model. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages 2774–2778. IEEE, 2022. 1, 2, 3, 4, 5, 7, 10, 11

    work page 2022

  2. [1]

    Mip-nerf: A multiscale representation for anti-aliasing neu- ral radiance fields

    Jonathan T Barron, Ben Mildenhall, Matthew Tancik, Peter Hedman, Ricardo Martin-Brualla, and Pratul P Srinivasan. Mip-nerf: A multiscale representation for anti-aliasing neu- ral radiance fields. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 5855–5864,

  3. [2]

    Mip-nerf 360: Unbounded anti-aliased neural radiance fields

    Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P Srinivasan, and Peter Hedman. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5470–5479, 2022. 2

    2022

  4. [3]

    A spatial processor model for ob- ject colour perception

    Gershon Buchsbaum. A spatial processor model for ob- ject colour perception. Journal of the Franklin institute , 310(1):1–26, 1980. 10

    work page 1980

  5. [5]

    Nerv: Neural representations for videos

    Hao Chen, Bo He, Hanyu Wang, Yixuan Ren, Ser Nam Lim, and Abhinav Shrivastava. Nerv: Neural representations for videos. Advances in Neural Information Processing Systems, 34, 2021. 2

    work page 2021

  6. [6]

    Aug-nerf: Training stronger neural radiance fields with triple-level physically-grounded augmentations

    Tianlong Chen, Peihao Wang, Zhiwen Fan, and Zhangyang Wang. Aug-nerf: Training stronger neural radiance fields with triple-level physically-grounded augmentations. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15191–15202, 2022. 1

    work page 2022

  7. [7]

    Domain adaptation for underwater image enhancement via content and style sepa- ration

    Yu-Wei Chen and Soo-Chang Pei. Domain adaptation for underwater image enhancement via content and style sepa- ration. IEEE Access, 10:90523–90534, 2022. 1, 4, 5, 10, 11

    work page 2022

  8. [8]

    Depth-supervised nerf: Fewer views and faster train- ing for free

    Kangle Deng, Andrew Liu, Jun-Yan Zhu, and Deva Ra- manan. Depth-supervised nerf: Fewer views and faster train- ing for free. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12882– 12891, 2022. 2

    2022

Show all 42 references
  1. [9]

    Aquagan: Restoration of underwater images

    Chaitra Desai, Badduri Sai Sudheer Reddy, Ramesh Ashok Tabib, Ujwala Patil, and Uma Mudenagudi. Aquagan: Restoration of underwater images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 296–304, 2022. 1, 4, 10

    2022

  2. [10]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

    2021

  3. [11]

    Uncertainty inspired underwater image enhance- ment

    Zhenqi Fu, Wu Wang, Yue Huang, Xinghao Ding, and Kai- Kuang Ma. Uncertainty inspired underwater image enhance- ment. In European Conference on Computer Vision , pages 465–482. Springer, 2022. 1

    2022

  4. [12]

    Uncertainty inspired underwater image enhance- ment

    Zhenqi Fu, Wu Wang, Yue Huang, Xinghao Ding, and Kai- Kuang Ma. Uncertainty inspired underwater image enhance- ment. In European Conference on Computer Vision (ECCV), pages 465–482, 2022. 1

    2022

  5. [13]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 1

    2016

  6. [14]

    Ray tracing volume densities

    J KAJITA. Ray tracing volume densities. In SIGGRAPH’84, volume 18, pages 165–174, 1984. 2

    1984

  7. [15]

    Zero-shot single image restoration through controlled perturbation of koschmieder’s model

    Aupendu Kar, Sobhan Kanti Dhara, Debashis Sen, and Prabir Kumar Biswas. Zero-shot single image restoration through controlled perturbation of koschmieder’s model. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 16205–16215, 2021. 1

    2021

  8. [16]

    Vision transformer for nerf-based view synthesis from a single input image

    Kai-En Lin, Lin Yen-Chen, Wei-Sheng Lai, Tsung-Yi Lin, Yi-Chang Shih, and Ravi Ramamoorthi. Vision transformer for nerf-based view synthesis from a single input image. In WACV, 2023. 6

    2023

  9. [17]

    Optical models for direct volume rendering

    Nelson Max. Optical models for direct volume rendering. IEEE Transactions on Visualization and Computer Graphics, 1(2):99–108, 1995. 2

    1995

  10. [18]

    Local light field fusion: Practical view syn- thesis with prescriptive sampling guidelines

    Ben Mildenhall, Pratul P Srinivasan, Rodrigo Ortiz-Cayon, Nima Khademi Kalantari, Ravi Ramamoorthi, Ren Ng, and Abhishek Kar. Local light field fusion: Practical view syn- thesis with prescriptive sampling guidelines. ACM Transac- tions on Graphics (TOG), 38(4):1–14, 2019. 4

    2019

  11. [19]

    Srinivasan, Rodrigo Ortiz-Cayon, Nima Khademi Kalantari, Ravi Ramamoorthi, Ren Ng, and Abhishek Kar

    Ben Mildenhall, Pratul P. Srinivasan, Rodrigo Ortiz-Cayon, Nima Khademi Kalantari, Ravi Ramamoorthi, Ren Ng, and Abhishek Kar. Local light field fusion: Practical view syn- thesis with prescriptive sampling guidelines. ACM Transac- tions on Graphics (TOG), 2019. 10

    2019

  12. [20]

    Nerf: Representing scenes as neural radiance fields for view syn- thesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceed- ing...

    2020

  13. [21]

    Reg- nerf: Regularizing neural radiance fields for view synthesis from sparse inputs

    Michael Niemeyer, Jonathan T Barron, Ben Mildenhall, Mehdi SM Sajjadi, Andreas Geiger, and Noha Radwan. Reg- nerf: Regularizing neural radiance fields for view synthesis from sparse inputs. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition,...

    2022

  14. [22]

    Unisurf: Unifying neural implicit surfaces and radiance fields for multi-view reconstruction

    Michael Oechsle, Songyou Peng, and Andreas Geiger. Unisurf: Unifying neural implicit surfaces and radiance fields for multi-view reconstruction. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 5589–5599, 2021. 2

    2021

  15. [23]

    Human-visual- system-inspired underwater image quality measures

    Karen Panetta, Chen Gao, and Sos Agaian. Human-visual- system-inspired underwater image quality measures. IEEE Journal of Oceanic Engineering, 41(3):541–551, 2015. 4

    2015

  16. [24]

    Barron, Sofien Bouaziz, Dan B Goldman, Steven M

    Keunhong Park, Utkarsh Sinha, Jonathan T. Barron, Sofien Bouaziz, Dan B Goldman, Steven M. Seitz, and Ricardo Martin-Brualla. Nerfies: Deformable neural radiance fields. ICCV, 2021. 2

    2021

  17. [25]

    Hypernerf: A higher- dimensional representation for topologically varying neu- ral radiance fields

    Keunhong Park, Utkarsh Sinha, Peter Hedman, Jonathan T Barron, Sofien Bouaziz, Dan B Goldman, Ricardo Martin- Brualla, and Steven M Seitz. Hypernerf: A higher- dimensional representation for topologically varying neu- ral radiance fields. In ACM Conference and Exhibition on Co...

    2021

  18. [26]

    Nan: Noise- aware nerfs for burst-denoising

    Naama Pearl, Tali Treibitz, and Simon Korman. Nan: Noise- aware nerfs for burst-denoising. In CVPR, 2022. 2

    2022

  19. [27]

    D-nerf: Neural radiance fields for dynamic scenes

    Albert Pumarola, Enric Corona, Gerard Pons-Moll, and Francesc Moreno-Noguer. D-nerf: Neural radiance fields for dynamic scenes. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 10318–10327, 2021. 2

    2021

  20. [28]

    Structure-from-motion revisited

    Johannes Lutz Sch ¨onberger and Jan-Michael Frahm. Structure-from-motion revisited. In Conference on Com- puter Vision and Pattern Recognition (CVPR), 2016. 5

    2016

  21. [29]

    Pixelwise view selection for un- structured multi-view stereo

    Johannes Lutz Sch ¨onberger, Enliang Zheng, Marc Pollefeys, and Jan-Michael Frahm. Pixelwise view selection for un- structured multi-view stereo. In European Conference on Computer Vision (ECCV), 2016. 5

    2016

  22. [30]

    Is attention all nerf needs? arXiv preprint arXiv:2207.13298, 2022

    Mukund Varma T, Peihao Wang, Xuxi Chen, Tianlong Chen, Subhashini Venugopalan, and Zhangyang Wang. Is attention all nerf needs? arXiv preprint arXiv:2207.13298, 2022. 2, 4

    2022 arXiv

  23. [31]

    Ref-nerf: Struc- tured view-dependent appearance for neural radiance fields

    Dor Verbin, Peter Hedman, Ben Mildenhall, Todd Zickler, Jonathan T Barron, and Pratul P Srinivasan. Ref-nerf: Struc- tured view-dependent appearance for neural radiance fields. arXiv preprint arXiv:2112.03907, 2021. 2

    2021 arXiv

  24. [32]

    Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction

    Peng Wang, Lingjie Liu, Yuan Liu, Christian Theobalt, Taku Komura, and Wenping Wang. Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction. arXiv preprint arXiv:2106.10689, 2021. 2

    2021 arXiv

  25. [33]

    Ibr- net: Learning multi-view image-based rendering

    Qianqian Wang, Zhicheng Wang, Kyle Genova, Pratul P Srinivasan, Howard Zhou, Jonathan T Barron, Ricardo Martin-Brualla, Noah Snavely, and Thomas Funkhouser. Ibr- net: Learning multi-view image-based rendering. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and ...

    2021

  26. [34]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600–612, 2004. 4

    2004

  27. [35]

    Nex: Real-time view synthesis with neural basis expansion

    Suttisak Wizadwongsa, Pakkapon Phongthawee, Jiraphon Yenphraphai, and Supasorn Suwajanakorn. Nex: Real-time view synthesis with neural basis expansion. InIEEE Confer- ence on Computer Vision and Pattern Recognition (CVPR) ,

  28. [36]

    Sinnerf: Training neural radiance fields on complex scenes from a single image.arXiv preprint arXiv:2204.00928, 2022

    Dejia Xu, Yifan Jiang, Peihao Wang, Zhiwen Fan, Humphrey Shi, and Zhangyang Wang. Sinnerf: Training neural radiance fields on complex scenes from a single image.arXiv preprint arXiv:2204.00928, 2022. 2

    2022 arXiv

  29. [37]

    An underwater color im- age quality evaluation metric

    Miao Yang and Arcot Sowmya. An underwater color im- age quality evaluation metric. IEEE Transactions on Image Processing, 24(12):6062–6071, 2015. 4

    2015

  30. [38]

    V ol- ume rendering of neural implicit surfaces

    Lior Yariv, Jiatao Gu, Yoni Kasten, and Yaron Lipman. V ol- ume rendering of neural implicit surfaces. Advances in Neu- ral Information Processing Systems, 34, 2021. 2

    2021

  31. [39]

    Fmsnet: Un- derwater image restoration by learning from a synthesized dataset

    Xiangyu Yin, Xiaohong Liu, and Huan Liu. Fmsnet: Un- derwater image restoration by learning from a synthesized dataset. In International Conference on Artificial Neural Net- works, pages 421–432. Springer, 2021. 1

    2021

  32. [40]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 586–595, 2018. 4

    2018

  33. [41]

    A fast single image haze removal algorithm using color attenuation prior

    Qingsong Zhu, Jiaming Mai, and Ling Shao. A fast single image haze removal algorithm using color attenuation prior. IEEE transactions on image processing, 24(11):3522–3533,

  34. [2015]

    Loss Functions To achieve the rendering and restoration in an unsuper- vised manner, it is important to regularise the model with appropriate losses

    10 A. Loss Functions To achieve the rendering and restoration in an unsuper- vised manner, it is important to regularise the model with appropriate losses. We propose 6 distinct loss functions which act upon the different output maps similar to [4] A.1. Reconstruction Loss We ...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.