Pith. sign in

REVIEW 3 major objections 5 minor 39 references

ProbeSDF: Light Field Probes for Neural Surface Reconstruction

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper claims that co-located spatial-and-directional appearance decoders are the bottleneck in SDF-based surface reconstruction, and that replacing them with separate fine spatial and coarse angular probe grids yields faster training…

desk verdict Genuinely new decoupled appearance model with real speed and credible geometry, but the 'supersedes all baselines' claim is constrained by per-camera biases and resolution asymmetries in the evaluation. read the letter →

arxiv 2412.10084 v2 pith:UZE5HSGW submitted 2024-12-13 cs.CV

Briac Toussaint , Diego Thomas , Jean-Sébastien Franco This is my paper · ORCID
classification cs.CV
keywords neuralsurfacereconstructionsigneddistancefunctionlightfieldprobessphericalharmonicsvolumerenderingmulti-view3Dreal-timeradiancedecoupling
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

SDF-based neural surface reconstruction normally stores color and view-direction information together in one high-capacity decoder. This paper claims that the two components have very different spatial bandwidth: surface texture varies fast, while lighting, and therefore angular radiance, varies slowly because light sources are usually far away. It therefore separates them, encoding spatial appearance in a fine voxel grid and angular appearance in a coarse grid of spherical-harmonic 'light field probes' at one sixteenth of the voxel resolution, with a tiny two-layer MLP and a polynomial Fresnel term as the only decoder. On human and object benchmarks (MVMannequins, ActorsHQ, DTU, BlendedMVS) the paper reports that this minimal refactor beats re-trained baselines on chamfer distance and PSNR while training in 1–4 minutes and rendering at 200–400 Hz. A sympathetic reader would take away that the co-located decoder is not necessary and that separating spatial and angular bandwidth is a cheap, general improvement.

What carries the argument

Light-field probes: a coarse grid (1/16 of voxel resolution) whose nodes store spherical-harmonic coefficient vectors $b_{ij}$; at a shading point the probe features are trilinearly interpolated (Eqs. 6–7) and the SH basis is evaluated at the reflected vector $r$, giving $F_a(x,r)$. Spatial features use a planar factorization $F_s = F_x\otimes F_y\otimes F_z$ per $16^3$ tile. A two-hidden-layer 32-neuron MLP decodes $F_s$, $F_a$, and the powers $(1-n\cdot v)^k$ (k=0..5), approximating the Fresnel term, into color. Everything—probe interpolation, feature lookup, and the MLP call—is fused into one single CUDA kernel, and the probes add only 1/6 of the spatial-feature memory (8×16 coefficients vs 3×16×16 per tile at $l=4$).

What would settle it

Train ProbeSDF on a synthetic glossy object illuminated by a point light placed close to the surface (distance comparable to object size) and compare against a second, distant-light condition while keeping the SDF grid resolution fixed; if chamfer error grows substantially and PSNR plateaus in the near-light case relative to a co-located decoder, the low-frequency probe assumption is falsified. A more direct check is to compare the interpolated probe-radiance field $F_a(x,r)$ at a surface point against a dense ground-truth angular radiance sampled from the renderer; the paper's assumption predicts that the residual rises as light distance shrinks.

Watch

Extended reading notes

Core claim

The central discovery is Eq. 4: $C = \mathrm{MLP}(F_s(x), F_a(x,r), (1-n\cdot v)^0,\dots,(1-n\cdot v)^5)$, where the spatial features $F_s(x)$ come from a high-resolution sparse voxel grid and the angular features $F_a(x,r)$ are computed by trilinearly interpolating spherical-harmonic coefficients stored in a coarse probe grid and evaluating them at the reflected direction $r$ (Eqs. 5–7). This decoupling rests on the claim that angular radiance is nearly parallax-invariant because lighting is mostly distant, so a low spatial frequency probe grid suffices. The paper's tables show this parametrization consistently improves chamfer distance and PSNR over MMH, Voxurf, NeuS2, and 2DGS on MVMannequins (1.04 mm), DTU (0.68 mm), BlendedMVS (2.21 mm), and ActorsHQ (37.48 dB), while cutting training time by roughly 3–15x and reaching 200–400 Hz rendering.

Load-bearing premise

The load-bearing premise is that angular radiance varies slowly enough in space to be represented by probes stored at one sixteenth of the voxel resolution with at most fourth-order spherical harmonics, which the paper states directly as the assumption that 'the lights are far enough, so that probes at a low spatial frequency can explain the parallax.' If a scene contains close light sources, sharp self-shadows, or strong local inter-reflections, the probe grid cannot explain the images and the shape-radiance optimization will push geometry off.

Editorial extensions

If this is right

  • The probe-based appearance model can serve as a drop-in replacement for the co-located decoders in SDF pipelines, since the SDF opacity and regularization structure (NeuS equations, eikonal, smoothness losses) is unchanged.
  • Training times drop to 1–4 minutes on a single RTX A6000 for both object and human scenes, putting millimetric reconstruction within interactive capture workflows.
  • Rendering at 200–400 Hz makes real-time inspection and rendering of reconstructed surfaces practical with a roughly 30–232 MB model.
  • Raising the SH order $l$ from 1 to 4 or increasing $n_s,n_a$ systematically improves PSNR and chamfer, so specularity is controllable with a single parameter knob, as shown in the ablations.
  • Per-camera bias vectors absorb inconsistent exposure and shadows on DTU and BlendedMVS, extending the method to casually lit multi-view captures.

Reading between the lines

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

  • The paper's own bandwidth argument suggests a test: for scenes with near-field lighting, the probe grid at 1/16 resolution should become the limiting factor, and one could verify this by training on synthetic scenes with a point light at decreasing distances and measuring chamfer error against probe spacing.
  • Because probes and geometry are optimized jointly but stored separately, the angular grid is a natural handle for relighting or lighting transfer: after training, editing probe coefficients (or fine-tuning them to a new environment) could update appearance without retraining geometry, which the paper lists as future work.
  • The same decoupling could extend to temporal reconstruction: since lighting is smooth in space, probe coefficients might be shared or tracked across frames, a direction the paper mentions but does not explore.
  • The ablation showing that $l=1$ (constant SH) degrades chamfer more than PSNR suggests geometry itself is regularized by the angular shading model, implying that probe resolution and SH order are geometric hyperparameters, not just appearance hyperparameters.
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 / 5 minor

Summary. ProbeSDF proposes a reparameterization of the appearance model for SDF-based multi-view reconstruction: spatial features are stored in a high-resolution sparse voxel grid, while angular features are stored in a coarser grid of light-field probes using low-order spherical harmonics, decoded by a tiny MLP that also receives powers of (1 - n·v) as a learned Fresnel proxy. The method is evaluated on MVMannequins, ActorsHQ, DTU, and BlendedMVS against Voxurf, NeuS2, MMH, and 2DGS, with per-scene tables, ablations over (ns, na, l), probe smoothing, the Fresnel term, and qualitative comparisons. The central claim is that this decoupling yields superior geometry and image metrics on all four benchmarks while providing large training and rendering speedups.

Significance. If the reported comparisons are accepted at face value, the contribution is significant and practical: the probe-based angular parametrization is conceptually clean, the MLP is dramatically smaller, training is minutes rather than tens of minutes, and rendering reaches hundreds of Hz. The paper has several concrete strengths: the baselines are re-trained under a stated common protocol, per-scene numbers are provided for all datasets, the ablations cover the main knobs of the parametrization, and source code is promised. The geometric results against external ground truth (chamfer on MVMannequins, DTU, BlendedMVS) appear substantially better than several baselines, which is the most trustworthy part of the evidence. However, the headline claim of superiority rests on evaluation asymmetries that are not yet resolved, so the contribution is promising but not yet established at the level claimed.

major comments (3)
  1. [Sec. 5, Tables 3 and 4; Sec. 5 text] The DTU and BlendedMVS comparisons are confounded by per-camera bias vectors. The text states that 'some datasets may contain inconsistent shadows or large exposure changes (DTU and BMVS) that are hard to model by our representation so we optionally train per-camera bias vectors in the MLP,' and Tables 3 and 4 mark the ProbeSDF rows with a checkmark for this option. Voxurf, NeuS2, and 2DGS are not given an equivalent per-camera or per-image adjustment. These bias vectors can absorb per-camera exposure/color differences, which directly inflates the image metric and also changes the photometric gradients that drive SDF optimization, so both the PSNR gaps (37.03/37.74 versus 37.08/36.00/36.03 on DTU; 35.19/35.89 versus 35.11/33.62 on BlendedMVS) and the chamfer comparisons are not apples-to-apples. A controlled comparison that either trains the same bias mechanism for the baselines or removes it from ProbeSDF is required before the claimed superiority on these two benchmarks can be accepted.
  2. [Sec. 5, Table 2 and ActorsHQ paragraph] The ActorsHQ comparison is asymmetric in input resolution. Voxurf and NeuS2 are trained at half resolution 'since we ran into issues with full resolution images,' while ProbeSDF is trained at full resolution. The headline PSNR of 37.48 dB is at r/1; at the matched resolution r/2, ProbeSDF obtains 36.62 dB versus 36.56 dB for Voxurf, a difference of only 0.06 dB. The statement that ProbeSDF 'also outperform[s] both when training at the same resolution' is technically true but not representative of the claimed margin, and the full-resolution comparison for the baselines is missing. This is a load-bearing issue because the resolution advantage is the main source of the reported 0.92 dB gap. A fair conclusion requires either full-resolution baseline runs or a presentation in which the matched-resolution numbers are the primary head-to-head comparison.
  3. [Sec. 5, Protocol paragraph] The image-metric claim is computed on the training images, not on held-out or novel views. The protocol says 'all available images were given as input and used for evaluation,' and PSNR is computed within the silhouette region. Consequently, the reported PSNR measures training-set reconstruction fidelity rather than generalization, and a model with extra per-camera parameters can increase this number without any improvement in geometry or in the rendering of unseen viewpoints. The geometric chamfer results against external scanners are less affected, but the paper should explicitly qualify the image-metric superiority claim as a training-fit measure, and ideally add a held-out-view PSNR evaluation or at least an analysis showing that the ranking is stable under held-out views.
minor comments (5)
  1. [Sec. 3.3] The phrase 'convolved by a53 gaussian kernel' appears to be a missing-formatting error; it should read 'a 5×3 Gaussian kernel' or 'a 5-tap Gaussian kernel' as appropriate.
  2. [Sec. 5, MVMannequins] The resolution '20482' should be typed as '2048×2048' or '2048²' to avoid a formatting artifact.
  3. [References] Reference [28] lists the last author as 'Franco Jean-Sébastien'; this should be normalized to 'Jean-Sébastien Franco' to match the author list of the present paper and standard citation style.
  4. [Abstract and Introduction] The phrase 'superseding both 3D and image metric performances with four popular benchmarks' is ambiguous and slightly overreaching: on ActorsHQ no independent geometric metric is reported, so the claim of both 3D and image superiority does not apply to all four benchmarks individually.
  5. [Sec. 5, ablations] The ablations cover the feature dimensionalities, the SH order, probe smoothing, and the Fresnel input, but not the probe grid resolution (the fixed 1/16th ratio). Given the central modeling assumption, an ablation varying this ratio would strengthen the paper.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the core contribution is an empirical architecture change evaluated against external benchmarks and external ground-truth geometry.

full rationale

ProbeSDF's central claim is not a derivation but an empirical architecture proposal. The appearance model in Eq. 4 is a proposed parametrization, not a quantity fitted to the benchmarks and then reported as a prediction; its components (spherical-harmonic probe interpolation, Fresnel power terms) are drawn from external sources [5, 14, 17, 18, 25, 26] and are explicitly adapted rather than imported as validated constraints. Geometry is optimized jointly with appearance through the photometric loss (Eq. 16) and is evaluated against external ground-truth meshes (DTU scans, BlendedMVS meshes, MVMannequins official metrics), so the chamfer results do not reduce to a fitted input. The use of the authors' own MMH code [28] as the sparse-voxel backbone is engineering reuse; the MMH comparison is a baseline, not the evidence for the probe assumption, and no load-bearing argument depends on a self-citation. The stated Limitation section concedes the failure mode under weak camera overlap and local lighting effects, which is an empirical risk rather than a circular step. Evaluation asymmetries noted by reviewers (per-camera bias vectors on DTU/BMVS, reduced-resolution baselines on ActorsHQ, PSNR computed on training images) concern whether the superiority claim is fairly supported, not whether the result is circular. No equation in the paper defines the output in terms of the claim, and no fitted parameter is renamed as a prediction. Therefore no circular step is present.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central claims rest on the domain hypothesis that view-dependent lighting is smoother than surface texture, plus a stack of standard components (NeuS opacity, TensorRF planar factorization, SH basis, NeuS volume rendering) taken from cited prior work. No new physical entities are postulated: the neural probes are a learned parameter cache borrowed from real-time rendering, not a new force, particle, or conserved quantity. The hand-set items are the feature dimensionalities (ns, na), SH order l, the 1/16 probe-grid ratio, per-camera bias vectors, and dataset-specific loss schedules (Tables 7 to 10); the ablations show the method is sensitive to several of these (Table 6: chamfer 2.58 to 2.22 across configurations). The 'Fresnel' term is an input-feature choice whose weights are learned, so the physics content is limited to feeding (1 - n.v)^k terms and conditioning on the reflected vector.

free parameters (5)
  • Feature dimensionalities (ns, na) and SH order l = (4,4,4), (8,8,4), (12,12,4), (4,4,1), (4,4,2), (4,4,3)
    Chosen by hand per configuration; Table 6 ablates them, showing the appearance model's capacity is tuned, not derived.
  • Probe grid resolution ratio = 1/16th of the main voxel grid
    Manual architectural choice in Sec. 3.2 and Fig. 2 that encodes the smooth-lighting hypothesis.
  • Per-camera bias vectors = trained per camera
    Optional extra parameters on DTU and BlendedMVS that absorb per-camera exposure and shadow inconsistencies; baselines do not use them. Ablation (Table 6) shows chamfer 2.47 to 2.37 and PSNR 34.76 to 35.19 on BlendedMVS.
  • Loss weights and per-LOD schedules = See Tables 7 to 10 (lambda_Eik from 1.0 to 0.025, learning rates 0.025 to 0.001)
    Hand-set per dataset and level of detail; different schedules for DTU, BlendedMVS, MVMannequins, and ActorsHQ.
  • Empirical gradient weighting factors = (max(c,cgt)+eps)^-1 and (1+|s|*5)^-1
    Introduced in Sec. 3.3 to reweight photometric and SDF losses; the factors are empirical and are not derived from any physical principle.
assumptions (6)
  • domain assumption Angular radiance is spatially smooth at the probe scale: a 1/16-resolution grid with SH order up to 4 can explain view-dependent appearance.
    Core decoupling hypothesis introduced in Sec. 1 ('angular components... can be encoded at a lower spatial resolution') and made concrete in Sec. 3.2 ('We also assume that the lights are far enough'). Fails for close lights, sharp self-shadows, and under-observed directions; the authors acknowledge the ActorsHQ latitude artifacts in Limitations.
  • domain assumption Reflectance is isotropic and dominated by a lobe around the reflected vector r.
    Sec. 3.1: 'Most materials reflect the light in a cone around the reflected vector'; 'eq. 4 assumes isotropic materials'. Retro-reflective and anisotropic materials are excluded from the model by construction.
  • standard math The NeuS opacity mapping (Eqs. 8 to 10) turns optimized SDF values into valid volume densities.
    Adopted from NeuS [31] without modification; standard in the SDF-rendering literature and cited as such.
  • standard math The planar factorization Fs = Fx ⊗ Fy ⊗ Fz captures high-frequency spatial texture.
    Taken from TensorRF [2]; a low-rank assumption on spatial features whose failure would force texture into the angular branch.
  • domain assumption Learned per-point radiance plus volume rendering approximates image formation for the tested captures.
    Standard NeRF-style assumption (Eq. 8); no explicit light transport is simulated, and the authors note in Limitations that shape-radiance ambiguity can cause local minima.
  • standard math Spherical harmonics form a valid basis and trilinear probe interpolation is continuous.
    Standard math used in Eqs. 5 to 7; interpolation continuity is exploited by the Lprobes smoothness loss.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ProbeSDF: Light Field Probes for Neural Surface Reconstruction." pith.science (2026). https://pith.science/paper/UZE5HSGW

@misc{pith2026241210084,
  author       = {Pith},
  title        = {Pith review of: ProbeSDF: Light Field Probes for Neural Surface Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UZE5HSGW}},
  note         = {Machine review of arXiv:2412.10084}
}
read the original abstract

SDF-based differential rendering frameworks have achieved state-of-the-art multiview 3D shape reconstruction. In this work, we re-examine this family of approaches by minimally reformulating its core appearance model in a way that simultaneously yields faster computation and increased performance. To this goal, we exhibit a physically-inspired minimal radiance parametrization decoupling angular and spatial contributions, by encoding them with a small number of features stored in two respective volumetric grids of different resolutions. Requiring as little as four parameters per voxel, and a tiny MLP call inside a single fully fused kernel, our approach allows to enhance performance with both surface and image (PSNR) metrics, while providing a significant training speedup and real-time rendering. We show this performance to be consistently achieved on real data over two widely different and popular application fields, generic object and human subject shape reconstruction, using four representative and challenging datasets.

Figures

Figures reproduced from arXiv: 2412.10084 by the authors.

Figure 1
Figure 1. We design a new appearance model for neural surface approaches, which combines high resolution spatial features and lower [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of the color prediction for a single voxel inside a [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The reflectivity cannot be encoded in Fa alone because different incident vectors can map to the same reflected vector (in red). In that case, similar angular features will be obtained for the two viewpoints due to the spatial and angular proximity of the lookups. The angle of incidence disambiguates the two situations. The reflectivity, also called the Fresnel term, is com￾monly approximated as follows for renderin… view at source ↗
Figures from the paper (19 more)
Figure 4
Figure 4. Figure 4: Left and center: The lighting direction changes at a lower [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: High-resolution reconstructions for ActorsHQ. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Reconstructions on challenging examples of DTU. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 9
Figure 9. Figure 9: Effect of the learned Fresnel term on three examples [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 8
Figure 8. Figure 8: Our model can be intuitively explained. We disable some [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 10
Figure 10. Figure 10: We train a scene with l = 4, that we then visualize with l varying from 1 to 4, left to right. The corresponding angular features are on the top row, and the predicted color is on the bottom row. We observe that the specularities can be removed simply by disabling the…
Figure 12
Figure 12. Figure 12: Reconstruction results on ActorsHQ. Left to right: RealityCapture, Ours, Voxurf, NeuS2. [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]
Figure 13
Figure 13. Figure 13: Reconstruction results on ActorsHQ. Left to right: RealityCapture, Ours, Voxurf, NeuS2. [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]
Figure 14
Figure 14. Figure 14: Reconstruction results on ActorsHQ. Left to right: RealityCapture, Ours, Voxurf, NeuS2. [PITH_FULL_IMAGE:figures/full_fig_p015_14.png]
Figure 15
Figure 15. Figure 15: Reconstruction results on ActorsHQ. Left to right: RealityCapture, Ours, Voxurf, NeuS2. [PITH_FULL_IMAGE:figures/full_fig_p016_15.png]
Figure 16
Figure 16. Figure 16: Qualitative comparison on ActorsHQ. Left to right: ground truth, Ours, Voxurf, NeuS2. Our method is able to handle the full [PITH_FULL_IMAGE:figures/full_fig_p017_16.png]
Figure 17
Figure 17. Figure 17: Qualitative comparison on ActorsHQ. Left to right: ground truth, Ours, Voxurf, NeuS2. [PITH_FULL_IMAGE:figures/full_fig_p018_17.png]
Figure 18
Figure 18. Figure 18: Qualitative comparison on ActorsHQ. Left to right: ground truth, Ours, Voxurf, NeuS2. [PITH_FULL_IMAGE:figures/full_fig_p019_18.png]
Figure 19
Figure 19. Figure 19: Reconstruction results on DTU. Left to right: Ours, Voxurf, NeuS2, 2DGS. We find that 2DGS excels at reconstructing flat [PITH_FULL_IMAGE:figures/full_fig_p020_19.png]
Figure 20
Figure 20. Figure 20: Qualitative comparison on DTU. Top: scan 24, middle: scan 37, bottom: scan 63. Left to right: ground truth, Ours, Voxurf, [PITH_FULL_IMAGE:figures/full_fig_p021_20.png]
Figure 21
Figure 21. Figure 21: Qualitative comparison on DTU. Top: scan 69, middle: scan 97, bottom: scan 110. Left to right: ground truth, Ours, Voxurf, [PITH_FULL_IMAGE:figures/full_fig_p022_21.png]
Figure 22
Figure 22. Figure 22: Reconstruction results on BlendedMVS. Left to right: Ground Truth, Ours, Voxurf, NeuS2. [PITH_FULL_IMAGE:figures/full_fig_p023_22.png]
Figure 23
Figure 23. Figure 23: Accuracy heatmaps on BlendedMVS. The pink color denotes points too far away from the ground truth, which are ignored in [PITH_FULL_IMAGE:figures/full_fig_p024_23.png]
Figure 24
Figure 24. Figure 24: Qualitative comparison on BlendedMVS. Left to right: ground truth, Ours, Voxurf, NeuS2. [PITH_FULL_IMAGE:figures/full_fig_p025_24.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 28 canonical work pages

  1. [1]

    https://www.capturingreality

    Realitycapture. https://www.capturingreality. com/. Accessed: 2024-11-21. 1

    work page 2024

  2. [2]

    Tensorf: Tensorial radiance fields

    Anpei Chen, Zexiang Xu, Andreas Geiger, Jingyi Yu, and Hao Su. Tensorf: Tensorial radiance fields. In European Conference on Computer Vision (ECCV), 2022. 4

    work page 2022

  3. [3]

    Pgsr: Planar-based gaussian splatting for efficient and high-fidelity surface reconstruction

    Danpeng Chen, Hai Li, Weicai Ye, Yifan Wang, Weijian Xie, Shangjin Zhai, Nan Wang, Haomin Liu, Hujun Bao, and Guofeng Zhang. Pgsr: Planar-based gaussian splatting for efficient and high-fidelity surface reconstruction. arXiv preprint arXiv:2406.06521, 2024. 2

    arXiv 2024

  4. [4]

    3d re- construction with fast dipole sums

    Hanyu Chen, Bailey Miller, and Ioannis Gkioulekas. 3d re- construction with fast dipole sums. ACM Transactions on Graphics (TOG), 43(6), 2024. 2

    work page 2024

  5. [5]

    Plenoxels: Radiance fields without neural networks

    Sara Fridovich-Keil, Alex Yu, Matthew Tancik, Qinhong Chen, Benjamin Recht, and Angjoo Kanazawa. Plenoxels: Radiance fields without neural networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 1, 2, 4

    work page 2022

  6. [6]

    Nerfren: Neural radiance fields with reflec- tions

    Yuan-Chen Guo, Di Kang, Linchao Bao, Yu He, and Song- Hai Zhang. Nerfren: Neural radiance fields with reflec- tions. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 18409– 18418, 2022. 3

    2022

  7. [7]

    2d gaussian splatting for geometrically accu- rate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically accu- rate radiance fields. In SIGGRAPH 2024 Conference Papers. Association for Computing Machinery, 2024. 2, 6, 7

    work page 2024

  8. [8]

    Sur2f: A hybrid representation for high-quality and efficient surface reconstruction from multi-view images

    Zhangjin Huang, Zhihao Liang, and Kui Jia. Sur2f: A hybrid representation for high-quality and efficient surface reconstruction from multi-view images. arXiv preprint arXiv:2401.03704, 2024. 2

    arXiv 2024

Show all 39 references
  1. [9]

    Humanrf: High-fidelity neural radiance fields for humans in motion

    Mustafa Is ¸ık, Martin R ¨unz, Markos Georgopoulos, Taras Khakhulin, Jonathan Starck, Lourdes Agapito, and Matthias Nießner. Humanrf: High-fidelity neural radiance fields for humans in motion. ACM Transactions on Graphics (TOG), 42(4):1–12, 2023. 2, 6, 1

    2023

  2. [10]

    Large scale multi-view stereopsis eval- uation

    Rasmus Jensen, Anders Dahl, George V ogiatzis, Engil Tola, and Henrik Aanæs. Large scale multi-view stereopsis eval- uation. In 2014 IEEE Conference on Computer Vision and Pattern Recognition, pages 406–413. IEEE, 2014. 2, 6

    2014

  3. [11]

    Gaussianshader: 3d gaussian splatting with shading functions for reflective surfaces

    Yingwenqi Jiang, Jiadong Tu, Yuan Liu, Xifeng Gao, Xiaox- iao Long, Wenping Wang, and Yuexin Ma. Gaussianshader: 3d gaussian splatting with shading functions for reflective surfaces. arXiv preprint arXiv:2311.17977, 2023. 3

    2023 arXiv

  4. [12]

    Tensoir: Tensorial inverse rendering

    Haian Jin, Isabella Liu, Peijia Xu, Xiaoshuai Zhang, Song- fang Han, Sai Bi, Xiaowei Zhou, Zexiang Xu, and Hao Su. Tensoir: Tensorial inverse rendering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 3

    2023

  5. [13]

    Real shading in unreal engine

    Brian Karis and Epic Games. Real shading in unreal engine

  6. [14]

    Physically Based Shading Theory Practice, 4(3):1,

    Proc. Physically Based Shading Theory Practice, 4(3):1,

  7. [15]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics (TOG), 42(4), 2023. 1, 2, 4, 6

    2023

  8. [16]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. CoRR, abs/1412.6980, 2014. 5

    2014 arXiv

  9. [17]

    Neuralangelo: High-fidelity neural surface reconstruction

    Zhaoshuo Li, Thomas M ¨uller, Alex Evans, Russell H Tay- lor, Mathias Unberath, Ming-Yu Liu, and Chen-Hsuan Lin. Neuralangelo: High-fidelity neural surface reconstruction. In 2023 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR). IEEE, 2023. 2

    2023

  10. [18]

    Scaling probe-based real-time dynamic global il- lumination for production

    Zander Majercik, Adam Marrs, Josef Spjut, and Morgan McGuire. Scaling probe-based real-time dynamic global il- lumination for production. arXiv preprint arXiv:2009.10796,

    2009 arXiv

  11. [19]

    Real-time global illumination using precom- puted light field probes

    Morgan McGuire, Mike Mara, Derek Nowrouzezahrai, and David Luebke. Real-time global illumination using precom- puted light field probes. InProceedings of the 21st ACM SIG- GRAPH symposium on interactive 3D graphics and games , pages 1–11, 2017. 3

    2017

  12. [20]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. In European Conference on Computer Vision (ECCV), 2020. 1, 2

    2020

  13. [21]

    Srinivasan, and Jonathan T

    Ben Mildenhall, Peter Hedman, Ricardo Martin-Brualla, Pratul P. Srinivasan, and Jonathan T. Barron. Nerf in the dark: High dynamic range view synthesis from noisy raw im- ages. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 16169–16178, 2021. 5

    2022

  14. [22]

    Objects as volumes: A stochastic geometry view of opaque solids

    Bailey Miller, Hanyu Chen, Alice Lai, and Ioannis Gkioulekas. Objects as volumes: A stochastic geometry view of opaque solids. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 87–97, 2024. 3

    2024

  15. [23]

    Instant neural graphics primitives with a mul- tiresolution hash encoding

    Thomas M ¨uller, Alex Evans, Christoph Schied, and Alexan- der Keller. Instant neural graphics primitives with a mul- tiresolution hash encoding. ACM Transactions on Graphics (TOG), 41(4):102:1–102:15, 2022. 6

    2022

  16. [24]

    Extracting Triangular 3D Models, Materials, and Light- ing From Images

    Jacob Munkberg, Jon Hasselgren, Tianchang Shen, Jun Gao, Wenzheng Chen, Alex Evans, Thomas M¨uller, and Sanja Fi- dler. Extracting Triangular 3D Models, Materials, and Light- ing From Images. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognitio...

    2022

  17. [25]

    Differentiable volumetric rendering: Learn- ing implicit 3d representations without 3d supervision

    Michael Niemeyer, Lars Mescheder, Michael Oechsle, and Andreas Geiger. Differentiable volumetric rendering: Learn- ing implicit 3d representations without 3d supervision. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 3504–3515, 2020. 2

    2020

  18. [26]

    Siggraph 2015 course: Physically based shading in theory and practice - advanced lighting r&d at ready at dawn studios

    David Neubelt & Matt Pettineo. Siggraph 2015 course: Physically based shading in theory and practice - advanced lighting r&d at ready at dawn studios. https : / / blog.selfshadow.com/publications/s2015- shading-course/. Accessed: 2024-11-07. 1, 3

    2015

  19. [27]

    Christopher M. Schlick. An inexpensive brdf model for physically-based rendering. Computer Graphics Forum, 13,

  20. [28]

    Structure-from-motion revisited

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

    2016

  21. [29]

    Millimetric Human Surface Capture in Minutes

    Briac Toussaint, Laurence Boissieux, Diego Thomas, Ed- mond Boyer, and Franco Jean-S´ebastien. Millimetric Human Surface Capture in Minutes. In SIGGRAPH Asia 2024 - 17th ACM SIGGRAPH Conference and Exhibition on Computer Graphics and Interactive Techniques in Asia , pages 1–12...

    2024

  22. [30]

    Barron, and Pratul P

    Dor Verbin, Peter Hedman, Ben Mildenhall, Todd Zickler, Jonathan T. Barron, and Pratul P. Srinivasan. Ref-NeRF: Structured view-dependent appearance for neural radiance fields. Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR) , 2022. 2, 3, 4

    2022

  23. [31]

    Nerf-casting: Improved view-dependent ap- pearance with consistent reflections

    Dor Verbin, Pratul P Srinivasan, Peter Hedman, Ben Milden- hall, Benjamin Attal, Richard Szeliski, and Jonathan T Barron. Nerf-casting: Improved view-dependent ap- pearance with consistent reflections. arXiv preprint arXiv:2405.14871, 2024. 3

    2024 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. NeurIPS, 2021. 1, 2, 3, 5, 7

    2021

  25. [33]

    Neus2: Fast learning of neural implicit surfaces for multi-view recon- struction

    Yiming Wang, Qin Han, Marc Habermann, Kostas Dani- ilidis, Christian Theobalt, and Lingjie Liu. Neus2: Fast learning of neural implicit surfaces for multi-view recon- struction. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023. 1, 2, 6, 7

    2023

  26. [34]

    Neural directional encoding for efficient and accurate view-dependent appearance modeling

    Liwen Wu, Sai Bi, Zexiang Xu, Fujun Luan, Kai Zhang, Iliyan Georgiev, Kalyan Sunkavalli, and Ravi Ramamoor- thi. Neural directional encoding for efficient and accurate view-dependent appearance modeling. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern ...

    2024

  27. [35]

    V oxurf: V oxel-based efficient and accurate neural surface reconstruction

    Tong Wu, Jiaqi Wang, Xingang Pan, Xudong Xu, Christian Theobalt, Ziwei Liu, and Dahua Lin. V oxurf: V oxel-based efficient and accurate neural surface reconstruction. In In- ternational Conference on Learning Representations (ICLR),

  28. [36]

    Blendedmvs: A large-scale dataset for generalized multi-view stereo net- works

    Yao Yao, Zixin Luo, Shiwei Li, Jingyang Zhang, Yufan Ren, Lei Zhou, Tian Fang, and Long Quan. Blendedmvs: A large-scale dataset for generalized multi-view stereo net- works. Computer Vision and Pattern Recognition (CVPR) ,

  29. [37]

    Multiview neu- ral surface reconstruction by disentangling geometry and ap- pearance

    Lior Yariv, Yoni Kasten, Dror Moran, Meirav Galun, Matan Atzmon, Basri Ronen, and Yaron Lipman. Multiview neu- ral surface reconstruction by disentangling geometry and ap- pearance. Advances in Neural Information Processing Sys- tems, 33, 2020. 2, 6

    2020

  30. [38]

    V olume rendering of neural implicit surfaces

    Lior Yariv, Jiatao Gu, Yoni Kasten, and Yaron Lipman. V olume rendering of neural implicit surfaces. In Thirty- Fifth Conference on Neural Information Processing Systems,

  31. [39]

    Srinivasan, Boyang Deng, Paul Debevec, William T

    Xiuming Zhang, Pratul P. Srinivasan, Boyang Deng, Paul Debevec, William T. Freeman, and Jonathan T. Barron. Ner- factor: neural factorization of shape and reflectance under an unknown illumination. ACM Transactions on Graphics (TOG), 40(6), 2021. 3 ProbeSDF: Light Field Probes...

    2021

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.