Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Acquisition of Spatially-Varying Reflectance and Surface Normals via Polarized Reflectance Fields

T0 review · 4 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 A single polarized reflectance-field capture yields 0.1mm/px normals and full SVBRDF maps for diffuse-to-glossy objects.

desk verdict A serious and novel polarized OLAT capture system, but the headline normal-accuracy claim is not supported by any independent measurement; deserves peer review with major revision expected. read the letter →

arxiv 2412.09772 v1 pith:R6ERLUNC submitted 2024-12-13 cs.CV cs.GR

Jing Yang , Pratusha Bhuvana Prasad , Qing Zhang , Yajie Zhao This is my paper · ORCID
classification cs.CVcs.GR
keywords polarizedreflectancefieldSVBRDFacquisitionsurfacenormalestimationsphericalgradientilluminationdiffuse-specularseparationWardBRDFlightstagecaptureOLAT
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

The paper proposes a complete capture pipeline in which a light stage records the object under 346 one-light-at-a-time directions through cross- and parallel-polarized light, producing a polarized reflectance field. Its central claim is that statistical analysis of this sequence removes overexposure, interreflection, and lens flare artifacts, and that an optimization over the whole image collection refines per-pixel surface normals to within 0.1mm/px while fitting albedo, specular, roughness, and anisotropy. This matters because current polarization-based methods either separate diffuse and specular reflectance well or handle complex non-convex objects, but not both, and glossy objects with concave geometry remain a failure mode for studio capture. A working version of this pipeline would give realistic-rendering and inverse-rendering research a practical source of measured materials and a dataset of polarized reflectance fields for training.

What carries the argument

The load-bearing identity is the polarization separation $I_d=2I_\perp$ and $I_s=2I_\parallel-2I_\perp$, which converts one hardware pass into two reflectance sequences. The optimization that carries the accuracy claim is normalized cross-correlation between the observed radiance vector and the predicted Lambertian irradiance vector $\nu_k\, \mathbf n\cdot \boldsymbol\omega_i^k$ for the diffuse normal, and the corresponding reflection-direction vector for the specular normal, both subject to $\mathbf n\cdot \boldsymbol\omega_i^k \ge 0$ to suppress interreflection and lens flare. The Ward BRDF, with a 2D Gaussian lobe in the half-vector frame, is the model that turns the refined normal into roughness and anisotropy estimates.

What would settle it

Capture a deeply concave, glossy object (for example, a polished metal bowl) whose true surface normals are known from a laser scanner, run the pipeline, and compare measured normals inside the concavity: if the claimed accuracy holds, cross-polarized intensities must follow the Lambertian cosine law even where interreflections are strong, so any normal error that grows with concavity or glossiness would show the load-bearing assumption is violated.

Watch

Extended reading notes

Core claim

At the core is an argument that a polarized OLAT sequence can be treated as a per-pixel signal, turning material capture into a sequence of statistical estimation steps. Cross-polarized frames isolate the diffuse reflection, and the difference between parallel- and cross-polarized frames isolates the specular reflection, with the separation factors derived through Mueller calculus and Malus's law. Overexposure is removed by treating each pixel's lighting-response curve as a signal and clipping anomalous pulses; interreflection and lens flare are suppressed through the constraint that the active light direction and the surface normal have nonnegative dot product; and occlusion is estimated from a visibility-weighted integral. Initial normals are synthesized from spherical gradient patterns, then refined by maximizing normalized cross-correlation between the observed intensity vector and the predicted Lambertian irradiance vector, or the specular reflection-direction vector for the specular normal. The same refined normals feed least-squares fits of the Ward model to obtain roughness and anisotropy, with albedo refit afterward. The paper's claim is that this produces physically consistent maps for objects that previously defeated polarization-based capture, from matte to mirrorlike.

Load-bearing premise

The load-bearing premise is that under cross-polarized light every surface point reflects like a uniform Lambertian surface, so the observed brightness across lighting directions is proportional to the cosine of the angle between each light direction and the surface normal; if polarization leaks, Fresnel effects, or interreflections break that proportionality, the optimized normals are biased and the bias propagates into every later map.

Editorial extensions

If this is right

  • Objects with clear coats, metals, and diffuse bases can be captured in a single studio session, producing separate diffuse and specular albedo and normal maps rather than a mixed estimate.
  • Physically based renderings built from the measured maps reproduce reference photographs under HDRI and area lighting, including anisotropic highlights that stay linear on cylindrical surfaces.
  • The released polarized reflectance fields of captured objects can serve as supervised training data for inverse-rendering and relighting methods.
  • The preprocessing and optimization steps remove overexposure, interreflection, and lens flare artifacts that static gradient-illumination capture leaves in albedo and normals.
  • The measured maps are view-consistent across the eight capture cameras, so the method supports multiview acquisition of shape with complex reflectance.

Reading between the lines

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

  • An untested extension is to apply the same per-pixel correlation refinement to non-polarized OLAT data, as long as specular contamination is removed by another means; the cost function itself does not require polarization hardware.
  • The paper's quantitative claim is a normal accuracy figure, but most validation is visual; a systematic comparison against an independent geometric ground truth on concave and multilayer objects would be the natural next step and would bound how the Lambertian assumption degrades with interreflection strength.
  • Because every albedo, roughness, and anisotropy map is refit using the optimized normals, the method's failure mode should appear first at grazing angles on brushed or anisotropic surfaces, where Fresnel effects break the cross-polarized Lambertian model; a targeted test on brushed metal at grazing views would reveal this.
  • The authors' own note that they do not currently measure geometry suggests the practical ceiling of the method on severely concave objects; integrating multiview geometry could turn the hard interreflection constraint into a learned or model-based correction.
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

4 major / 5 minor

Summary. The paper proposes a polarized reflectance-field capture and optimization pipeline for acquiring surface normals and spatially varying reflectance (diffuse/specular albedo, specular variance, anisotropy, roughness) of real, non-planar objects. The capture uses a Light Stage with 346 linearly polarized one-light-at-a-time (OLAT) directions and eight synchronized cameras; diffuse/specular separation is performed from cross- and parallel-polarized observations. The pipeline removes overexposure, inter-reflection, and lens-flare artifacts, synthesizes spherical gradient illuminations from the OLAT data, optimizes diffuse and specular normal maps via cross-correlation with a Lambertian template, and refits albedo, roughness, and anisotropy from those normals. Validation is mainly qualitative: ablations, rendered comparisons, and a relighting test whose reference is itself synthesized from the captured OLAT images. The abstract claims surface normals within 0.1 mm/px, but no ground-truth or error-metric experiment supports that number.

Significance. If the claims were substantiated, the contribution would be significant: a full-SVBRDF-plus-normal capture solution for concave, glossy, multi-material objects, together with a planned release of polarized reflectance-field data, would be a valuable resource for inverse rendering and material acquisition. The pipeline is coherently organized: dense OLAT capture, analytic artifact suppression, and coupled optimization are clearly specified, and the supplementary material provides derivations for the albedo initialization. However, the manuscript currently does not provide evidence for the headline precision claim, and the polarization model as derived omits the surface's effect on the polarization state of reflected light. The value of the contribution therefore remains conditional on additional, independent validation and on either correcting or quantitatively justifying the polarization assumptions.

major comments (4)
  1. [Abstract; §5] The central claim of "highly accurate surface normals (within 0.1mm/px)" is never tested. No ground-truth geometry, synthetic render with known normals, precision-machined phantom, or independent sensor is used; Figs. 7, 16, and 17 are qualitative ablations, Fig. 10 is a qualitative comparison against [35] with no error metric, and Table 3 reports only solver gradient norms, not measurement error against any reference. The pixel-to-millimeter mapping needed to interpret 0.1 mm/px is also not defined. A dedicated validation with known geometry or synthetic scenes, plus a per-pixel error metric (angular error, RMS normal error, or equivalent), is required to support the abstract's quantitative claim.
  2. [§5 (Relighting, Fig. 9)] The relighting validation is self-referential. The reference image under HDRI lighting is synthesized by weighting the same captured OLAT images that are used to fit the material parameters (Eqs. 12-17). Matching this reference can only attest to internal consistency of the fitting procedure, not to physical accuracy of the measured normals or reflectance. A valid test would compare against real photographs under independent illumination, use a held-out subset of OLAT directions for prediction, or render a synthetic object with known SVBRDF and compare against ground truth.
  3. [§4.1, Eqs. (2)-(3); Supplement A] The separation Id = 2I⊥ and Is = 2I∥ − 2I⊥ is derived in the supplement with a Mueller-calculus proof that places no reflective surface between the polarizer and the analyzer. In the actual capture, the incident light is already linearly polarized, and Fresnel reflection at non-normal incidence changes the polarization state because the s and p reflection coefficients differ; rough or multilayer surfaces additionally depolarize. The crossed analyzer therefore transmits part of the specular lobe, and the subtraction in Eq. (3) does not generally isolate the specular component. The paper should either derive the correct Mueller-matrix reflectance model including Fresnel rotation and depolarization, or provide a quantitative calibration (for example, a dielectric sphere at Brewster angle) showing that the leakage is negligible over the claimed material range.
  4. [§4.3, Eqs. (11)-(12)] The diffuse normal refinement assumes every cross-polarized OLAT observation is a Lambertian response proportional to n·ωi with uniform incident radiance Li across all 346 lights. No radiometric calibration of the LEDs is reported, and the constraints n·ωi ≥ 0 remove interreflection and lens-flare candidates but do not remove specular leakage or nonuniform-incidence bias. Because the optimized normal enters the subsequent albedo, roughness, and anisotropy fits (Eqs. 14-17), any bias propagates through the entire SVBRDF output. A synthetic sensitivity study - render surfaces with known normals and BRDFs, then deliberately add polarization leakage and nonuniform Li and rerun the optimization - is needed to bound the resulting normal error.
minor comments (5)
  1. [Algorithm 1] The algorithm initializes δ as the mean of the entire OLAT sequence, while the surrounding text describes δ as an ambient value with δ ≪ ε; the mean of a signal containing specular peaks is not generally an ambient value, and the threshold ε is never given a concrete value or calibration procedure.
  2. [Supplement B] The constraint is written as n × ωi ≥ 0; throughout the main paper the intended constraint is the dot product n · ωi ≥ 0 (Eqs. 12-17). The cross-product expression is dimensionally inconsistent with a scalar inequality and should be corrected.
  3. [Fig. 4] In the caption, the specular occlusion map is labeled τd; given the notation in Table 1 and the surrounding text, this should presumably be τs.
  4. [§2.3 and §4.1] The related-work section states that the capture is based on linearly polarized spherical gradient illumination [35], while the method actually captures OLAT and synthesizes gradient patterns offline (Eqs. 8-9). This distinction should be stated clearly in the method overview to avoid confusion about what is physically captured.
  5. [Abstract] The unit "0.1mm/px" needs a precise definition: is it a per-pixel normal perturbation expressed as surface displacement at the object's depth, and for what camera resolution and object distance? Without this definition the quantitative claim is not interpretable.

Circularity Check

1 steps flagged · score 4.0 of 10

Relighting validation is self-referential: the HDRI reference is a weighted sum of the same OLAT images used to fit all material parameters, so the agreement is partly forced by construction; the core estimation pipeline itself is not circular.

  1. fitted input called prediction [Section 5, 'Relighting' (page 7-8, Fig. 9)]
    "The reference image for HDRI illumination is synthesized by weighting the captured OLAT images for each light, with these weights derived from averaging the pixel values within corresponding spherical areas of the HDRI light probe. In Blender, we use the principal BSDF material, incorporating the measured diffuse albedo as the base color, the diffuse normal as the object's normal, and the measured specular albedo to set the specular IOR level along with the measured tangent. The measured roughness is also applied. Under HDRI lighting, our renderings closely match the reference image."

    The HDRI reference is a weighted sum of the same cross- and parallel-polarized OLAT images (I_d, I_s) against which the diffuse and specular albedos, normals, roughness, and anisotropy are optimized in Eqs. 12-17. A renderer using parameters fitted to reproduce those OLAT responses will approximately reproduce any linear combination of them, so the close match between rendering and reference is largely a self-consistency check of the fit rather than an independent measurement of accuracy. The abstract's quantitative claims ('within 0.1mm/px', 'accurate' reflectance) are therefore not independently validated by this comparison; no external geometry ground truth is provided, and the paper itself states in the Limitations that geometry is not measured.

full rationale

The core estimation chain is not circular: polarized OLAT separation (Eqs. 2-3), gradient-illumination initial normals (Eqs. 8-10), cross-correlation normal refinement (Eqs. 11-13), Ward-lobe variance fitting (Eq. 14), and albedo refits (Eqs. 16-17) all estimate unknowns from captured images under stated physical models. There is no load-bearing self-citation: the key prior work [35] has no overlapping authors, and no uniqueness theorem is imported from the present authors. The circularity is confined to the validation of the headline accuracy claim: the relighting reference image is synthesized by weighting the very OLAT images used to fit all material parameters, so the reported rendering agreement is substantially forced by the fitting procedure. Separately, the physical assumptions in Eq. 11 (uniform Lambertian response under cross-polarized OLAT) and Eq. 2 (perfect polarization preservation by specular reflection) are assumptions whose violation would bias the outputs, but that is a correctness risk, not circularity. The paper also acknowledges a relevant limitation: 'the method does not currently measure geometry, leading to difficulties in managing shadows caused by self-occlusion.' For these reasons, the central algorithm has independent content and only the accuracy validation is partially circular, giving a score of 4 rather than 6 or higher.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The pipeline is not a parameter-free derivation: it depends on a physical polarization model, a Lambertian diffuse model, a Ward specular model, uniform-light integration, and two artifact heuristics (n·ω_i>=0 and pulse replacement), plus several hand-set or fitted numbers. Each of these assumptions can shift the final normal and reflectance maps.

free parameters (4)
  • Overexposure detection threshold ε = not reported
    Algorithm 1 detects abnormal intensity pulses by comparing sorted-channel differences to ε; the paper says it can be easily determined during capture but gives no value or calibration procedure. This affects the diffuse and specular sequences used everywhere downstream.
  • Overexposure iteration count M = 2
    Set to 2 as a balance between artifact removal and preserving the original intensity distribution (Appendix C.1). The choice is hand-set and not optimized or justified with data.
  • Albedo scale constant κ = (L0·A0)^-1 = not reported
    Albedo initialization in Eqs. 6-7 and 30-31 depends on κ from light intensity and solid angle. The paper states it is device-determined but does not report the calibration or value, so absolute albedo scale is not independently reproducible.
  • Ward specular variance σ = (σx, σy) = per-pixel fit via Eq. 14
    The Ward lobe width is fit per pixel to the observed OLAT response; anisotropy and roughness are then derived from this fit, and specular albedo is refined with it (Eq. 17). These outputs are model-fitting results presented as measured quantities.
assumptions (6)
  • domain assumption Specular reflection preserves the polarization state of incident light while diffuse reflection is unpolarized, so I_d=2I_perp and I_s=2I_parallel-2I_perp.
    Section 3 and the Mueller calculus proof in Appendix A. The whole diffuse/specular decomposition rests on ideal polarizers and this Fresnel polarization model; non-ideal polarizers or partial depolarization violate it.
  • domain assumption Under cross-polarized OLAT, every visible surface point is a uniform Lambertian reflector with observed radiance proportional to (ω_i·n) for constant incident radiance.
    Invoked in Eq. 11 and the normal refinement objective Eq. 12; glossy, multilayer, or concave surfaces with residual specular leakage or interreflection break it.
  • domain assumption Specular reflection is described by the Ward BRDF with a 2D Gaussian lobe parameterized by σ=(σx,σy).
    Section 4.3 and Eq. 14; roughness and anisotropy are derived from the fitted σ, so the output maps inherit this model assumption.
  • domain assumption Incident radiance is uniform over the sphere and OLAT directions sample the sphere uniformly enough to replace integrals by sample means (law of large numbers).
    Used in Appendix A Eqs. 23-29 to derive albedo initialization; any spatial or radiometric nonuniformity of the light stage biases absolute albedo.
  • ad hoc to paper Inter-reflection and lens flare occur only for observations with n·ω_i < 0, so these observations can be discarded.
    Section 4.2; this is a heuristic exclusion rule, not a physical model of interreflection, and it also discards valid observations on concave or shadowed surfaces. The appendix concedes severe interreflection remains unsolved.
  • ad hoc to paper Overexposure appears as isolated high-intensity pulses in the sorted per-pixel OLAT signal and can be replaced by an ambient value.
    Algorithm 1; the threshold ε and iteration count M are hand-set, and the assumption fails for broad specular highlights.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Acquisition of Spatially-Varying Reflectance and Surface Normals via Polarized Reflectance Fields." pith.science (2026). https://pith.science/paper/R6ERLUNC

@misc{pith2026241209772,
  author       = {Pith},
  title        = {Pith review of: Acquisition of Spatially-Varying Reflectance and Surface Normals via Polarized Reflectance Fields},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/R6ERLUNC}},
  note         = {Machine review of arXiv:2412.09772}
}
read the original abstract

Accurately measuring the geometry and spatially-varying reflectance of real-world objects is a complex task due to their intricate shapes formed by concave features, hollow engravings and diverse surfaces, resulting in inter-reflection and occlusion when photographed. Moreover, issues like lens flare and overexposure can arise from interference from secondary reflections and limitations of hardware even in professional studios. In this paper, we propose a novel approach using polarized reflectance field capture and a comprehensive statistical analysis algorithm to obtain highly accurate surface normals (within 0.1mm/px) and spatially-varying reflectance data, including albedo, specular separation, roughness, and anisotropy parameters for realistic rendering and analysis. Our algorithm removes image artifacts via analytical modeling and further employs both an initial step and an optimization step computed on the whole image collection to further enhance the precision of per-pixel surface reflectance and normal measurement. We showcase the captured shapes and reflectance of diverse objects with a wide material range, spanning from highly diffuse to highly glossy - a challenge unaddressed by prior techniques. Our approach enhances downstream applications by offering precise measurements for realistic rendering and provides a valuable training dataset for emerging research in inverse rendering. We will release the polarized reflectance fields of several captured objects with this work.

Figures

Figures reproduced from arXiv: 2412.09772 by the authors.

Figure 2
Figure 2. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Inter-reflection and Lens Flare. In row 1), we present the captured intensity distribution of a fixed surface point, indi￾cated with a cross (×) in the following example images. The inten￾sity patterns are identified as a) interreflection, b) regular specular reflection, and c) lens flare. Row 2) provides a false-color view, with a zoom-in view in the last row 3), as well as the raw capture. In the false-color view,… view at source ↗
Figure 6
Figure 6. Ablation study on overexposure removal. The proposed overexposure removal effectively mitigates the lighting baked-in effect from the acquired specular albedo [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figures from the paper (12 more)
Figure 9
Figure 9. Figure 9: Rendering Comparisons. We validate our results using physically-based renderings with the measured materials. For each object, we showcase the renderings with corresponding captures under 1) environmental lighting or 2) area lighting conditions. Also, the specular albe…
Figure 10
Figure 10. Figure 10: Qualitative Comparison. We compare results from 1) our method with 2) from [35] via static capture on objects with specular outer layers. Examined properties cover diffuse albedo ρd and specular albedo ρs, diffuse normal nd, and specular normal ns, with zoomed-in view…
Figure 12
Figure 12. Figure 12: Dataset Examples. we factor out the average of ωi · n, resulting in: τ = 1 2π Z Ω ν(ωi , ωo) · (ωi · n)dωi ·  1 2π Z Ω (ωi · n)dωi −1 = 1 2π lim m→∞  2π P m I i m · 2−1  · ( 1 2 ) −1 ≈ 4 N X N k=0 ⌈I k − ζ⌉ · max(ω k i · n, 0) (32) where ζ is the average ambient n…
Figure 13
Figure 13. Figure 13: Ablation study on overexposure removal: Specular Albedo. The proposed overexposure removal effectively mitigates the lighting baked-in effect from the acquired specular albedo [PITH_FULL_IMAGE:figures/full_fig_p015_13.png]
Figure 14
Figure 14. Figure 14: Ablation Study on overexposure removal: Diffuse Albedo. Similar to the improvements over specular albedo, overexposure removal can also improve diffuse albedo [PITH_FULL_IMAGE:figures/full_fig_p015_14.png]
Figure 15
Figure 15. Figure 15: Ablation Study on Overexposure Removal: Iterations We showcase the diffuse albedo and specular albedo obtained with overexposure removal via various iterations in (b, d), and the corresponding removed overexposure values in (a, c), where M is the total number of itera…
Figure 16
Figure 16. Figure 16: Ablation Study on acquired diffuse normal with and without optimization. [PITH_FULL_IMAGE:figures/full_fig_p015_16.png]
Figure 17
Figure 17. Figure 17: Ablation Study on acquired specular normal with and without optimization. [PITH_FULL_IMAGE:figures/full_fig_p016_17.png]
Figure 18
Figure 18. Figure 18: More Qualitative Comparison. We compare results from 1) our method with 2) from [35] via static capture on objects with specular outer layers. Examined properties cover diffuse albedo ρd and specular albedo ρs, diffuse normal nd, and specular normal ns, with zoomed-in…
Figure 19
Figure 19. Figure 19: More Qualitative Comparison. We compare results from 1) our method with 2) from [35] via static capture on objects with specular outer layers. Examined properties cover diffuse albedo ρd and specular albedo ρs, diffuse normal nd, and specular normal ns, with zoomed-in…
Figure 20
Figure 20. Figure 20: Polarized OLAT from Multiviews. We showcase an example captured object from multiview under cross-polarized and parallel-polarized OLAT. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_20.png]
Figure 21
Figure 21. Figure 21: More Optimization Results. We present a grey ball from a) right and b) front, a soda can from c) left and d) right, and a specular cup from e) right and f) front. For each object, we showcase 1) original image, 2) diffuse albedo ρˆd, 3) specular albedo ρˆs, 4) diffuse…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. A Real-world Display Inverse Rendering Dataset

    cs.GR 2025-08 conditional novelty 7.0 of 10

    The paper introduces and evaluates the first real-world dataset for inverse rendering with a display-camera system, including a baseline method that recovers normals more accurately than prior methods.

Reference graph

Works this paper leans on

49 extracted references · 45 canonical work pages · cited by 1 Pith paper

  1. [35]

    Rapid acqui- sition of specular and diffuse normal maps from polarized spherical gradient illumination

    Wan-Chun Ma, Tim Hawkins, Pieter Peers, Charles-Felix Chabert, Malte Weiss, Paul E Debevec, et al. Rapid acqui- sition of specular and diffuse normal maps from polarized spherical gradient illumination. Rendering Techniques, 2007 (9):10, 2007. 1, 2, 3, 6, 8, 14, 16, 17

    work page 2007

  2. [1]

    Shape, illumination, and reflectance from shading

    Jonathan T Barron and Jitendra Malik. Shape, illumination, and reflectance from shading. IEEE transactions on pattern analysis and machine intelligence , 37(8):1670–1687, 2014. 1

    work page 2014

  3. [2]

    Photo- metric stereo with general, unknown lighting

    Ronen Basri, David Jacobs, and Ira Kemelmacher. Photo- metric stereo with general, unknown lighting. International Journal of computer vision, 72:239–257, 2007. 1

    work page 2007

  4. [3]

    Nonlinear programming: theory and algorithms

    Mokhtar S Bazaraa, Hanif D Sherali, and Chitharanjan M Shetty. Nonlinear programming: theory and algorithms . John wiley & sons, 2013. 14

    work page 2013

  5. [4]

    Efficient and modular implicit differentiation

    Mathieu Blondel, Quentin Berthet, Marco Cuturi, Roy Frostig, Stephan Hoyer, Felipe Llinares-L ´opez, Fabian Pe- dregosa, and Jean-Philippe Vert. Efficient and modular implicit differentiation. arXiv preprint arXiv:2105.15183 ,

    arXiv

  6. [5]

    Nerd: Neural reflectance decomposition from image collections

    Mark Boss, Raphael Braun, Varun Jampani, Jonathan T Bar- ron, Ce Liu, and Hendrik Lensch. Nerd: Neural reflectance decomposition from image collections. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 12684–12694, 2021. 1

    work page 2021

  7. [6]

    Neural-pil: Neu- ral pre-integrated lighting for reflectance decomposition

    Mark Boss, Varun Jampani, Raphael Braun, Ce Liu, Jonathan Barron, and Hendrik Lensch. Neural-pil: Neu- ral pre-integrated lighting for reflectance decomposition. Advances in Neural Information Processing Systems , 34: 10691–10704, 2021. 1

    work page 2021

  8. [7]

    JAX: composable transformations of Python+NumPy programs, 2018

    James Bradbury, Roy Frostig, Peter Hawkins, Matthew James Johnson, Chris Leary, Dougal Maclau- rin, George Necula, Adam Paszke, Jake VanderPlas, Skye Wanderman-Milne, and Qiao Zhang. JAX: composable transformations of Python+NumPy programs, 2018. 13

    work page 2018

Show all 49 references
  1. [8]

    Physics-based inverse rendering using combined implicit and explicit geometries

    Guangyan Cai, Kai Yan, Zhao Dong, Ioannis Gkioulekas, and Shuang Zhao. Physics-based inverse rendering using combined implicit and explicit geometries. In Computer Graphics Forum , pages 129–138. Wiley Online Library,

  2. [9]

    Field guide to polarization

    Edward Collett. Field guide to polarization. Spie Belling- ham, W A, 2005. 11

    2005

  3. [10]

    Torrance

    Robert L Cook and Kenneth E. Torrance. A reflectance model for computer graphics. ACM Transactions on Graph- ics (ToG), 1(1):7–24, 1982. 2

    1982

  4. [11]

    Reflectance and texture of real-world sur- faces

    Kristin J Dana, Bram Van Ginneken, Shree K Nayar, and Jan J Koenderink. Reflectance and texture of real-world sur- faces. ACM Transactions On Graphics (TOG), 18(1):1–34,

  5. [12]

    Acquiring the reflectance field of a human face

    Paul Debevec, Tim Hawkins, Chris Tchou, Haarm-Pieter Duiker, Westley Sarokin, and Mark Sagar. Acquiring the reflectance field of a human face. In Proceedings of the 27th annual conference on Computer graphics and interac- tive techniques, pages 145–156, 2000. 2

    2000

  6. [13]

    Single-image svbrdf cap- ture with a rendering-aware deep network

    Valentin Deschaintre, Miika Aittala, Fredo Durand, George Drettakis, and Adrien Bousseau. Single-image svbrdf cap- ture with a rendering-aware deep network. ACM Transac- tions on Graphics (ToG), 37(4):1–15, 2018. 2

    2018

  7. [14]

    Deep polarization imaging for 3d shape and svbrdf acquisi- tion

    Valentin Deschaintre, Yiming Lin, and Abhijeet Ghosh. Deep polarization imaging for 3d shape and svbrdf acquisi- tion. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 15567–15576,

  8. [15]

    Neural biplane representation for btf rendering and acquisition

    Jiahui Fan, Beibei Wang, Milos Hasan, Jian Yang, and Ling- Qi Yan. Neural biplane representation for btf rendering and acquisition. In ACM SIGGRAPH 2023 Conference Proceed- ings, pages 1–11, 2023. 1, 2

    2023

  9. [16]

    Single-shot reflectance measurement from polarized color gradient illumination

    Graham Fyffe and Paul Debevec. Single-shot reflectance measurement from polarized color gradient illumination. In 2015 IEEE International Conference on Computational Pho- tography (ICCP), pages 1–10. IEEE, 2015. 3

    2015

  10. [17]

    Deep inverse rendering for high-resolution svbrdf estimation from an arbitrary number of images

    Duan Gao, Xiao Li, Yue Dong, Pieter Peers, Kun Xu, and Xin Tong. Deep inverse rendering for high-resolution svbrdf estimation from an arbitrary number of images. ACM Trans. Graph., 38(4):134–1, 2019. 2

    2019

  11. [18]

    Towards material digitization with a dual-scale op- tical system

    Elena Garces, Victor Arellano, Carlos Rodriguez-Pardo, David Pascual-Hernandez, Sergio Suja, and Jorge Lopez- Moreno. Towards material digitization with a dual-scale op- tical system. ACM Transactions on Graphics (TOG), 42(4): 1–13, 2023. 1, 2

    2023

  12. [19]

    Estimating specular roughness and anisotropy from second order spherical gradient illumina- tion

    Abhijeet Ghosh, Tongbo Chen, Pieter Peers, Cyrus A Wil- son, and Paul Debevec. Estimating specular roughness and anisotropy from second order spherical gradient illumina- tion. In Computer Graphics Forum, pages 1161–1170. Wiley Online Library, 2009. 1, 3, 12

    2009

  13. [20]

    Circularly polarized spherical illumi- nation reflectometry

    Abhijeet Ghosh, Tongbo Chen, Pieter Peers, Cyrus A Wil- son, and Paul Debevec. Circularly polarized spherical illumi- nation reflectometry. In ACM SIGGRAPH Asia 2010 papers, pages 1–12. 2010. 2, 3

    2010

  14. [21]

    Multiview face cap- ture using polarized spherical gradient illumination

    Abhijeet Ghosh, Graham Fyffe, Borom Tunwattanapong, Jay Busch, Xueming Yu, and Paul Debevec. Multiview face cap- ture using polarized spherical gradient illumination. ACM Transactions on Graphics (TOG), 30(6):1–10, 2011. 1, 3

    2011

  15. [22]

    Shape and spatially-varying brdfs from pho- tometric stereo

    Dan B Goldman, Brian Curless, Aaron Hertzmann, and Steven M Seitz. Shape and spatially-varying brdfs from pho- tometric stereo. IEEE Transactions on Pattern Analysis and Machine Intelligence, 32(6):1060–1071, 2009. 1

    2009

  16. [23]

    The re- lightables: V olumetric performance capture of humans with realistic relighting

    Kaiwen Guo, Peter Lincoln, Philip Davidson, Jay Busch, Xueming Yu, Matt Whalen, Geoff Harvey, Sergio Orts- Escolano, Rohit Pandey, Jason Dourgarian, et al. The re- lightables: V olumetric performance capture of humans with realistic relighting. ACM Transactions on Graphics (To...

    2019

  17. [24]

    A new conjugate gradient method with guaranteed descent and an efficient line search

    William W Hager and Hongchao Zhang. A new conjugate gradient method with guaranteed descent and an efficient line search. SIAM Journal on optimization, 16(1):170–192, 2005. 14

    2005

  18. [25]

    Re- flectance capture using univariate sampling of brdfs

    Zhuo Hui, Kalyan Sunkavalli, Joon-Young Lee, Sunil Hadap, Jian Wang, and Aswin C Sankaranarayanan. Re- flectance capture using univariate sampling of brdfs. In Pro- ceedings of the IEEE International Conference on Computer Vision, pages 5362–5370, 2017. 2

    2017

  19. [26]

    Sparse ellipsom- etry: portable acquisition of polarimetric svbrdf and shape with unstructured flash photography

    Inseung Hwang, Daniel S Jeon, Adolfo Munoz, Diego Gutierrez, Xin Tong, and Min H Kim. Sparse ellipsom- etry: portable acquisition of polarimetric svbrdf and shape with unstructured flash photography. ACM Transactions on Graphics (TOG), 41(4):1–14, 2022. 2 9

    2022

  20. [27]

    Diffuse-specular separation using binary spherical gradient illumination

    Christos Kampouris, Stefanos Zafeiriou, and Abhijeet Ghosh. Diffuse-specular separation using binary spherical gradient illumination. In EGSR (EI&I), pages 1–10, 2018. 3

    2018

  21. [28]

    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, 42 (4), 2023. 1

    2023

  22. [29]

    Neumip: Multi-resolution neural ma- terials

    Alexandr Kuznetsov. Neumip: Multi-resolution neural ma- terials. ACM Transactions on Graphics (TOG), 40(4), 2021. 2

    2021

  23. [30]

    Rendering neural materials on curved surfaces

    Alexandr Kuznetsov, Xuezheng Wang, Krishna Mullia, Fu- jun Luan, Zexiang Xu, Milos Hasan, and Ravi Ramamoorthi. Rendering neural materials on curved surfaces. InACM SIG- GRAPH 2022 conference proceedings, pages 1–9, 2022. 2

    2022

  24. [31]

    Efficient multispectral fa- cial capture with monochrome cameras

    Chloe LeGendre, Kalle Bladin, Bipin Kishore, Xinglei Ren, Xueming Yu, and Paul Debevec. Efficient multispectral fa- cial capture with monochrome cameras. InACM SIGGRAPH 2018 Posters, pages 1–2. 2018. 3

    2018

  25. [32]

    Differentiable monte carlo ray tracing through edge sampling

    Tzu-Mao Li, Miika Aittala, Fr ´edo Durand, and Jaakko Lehti- nen. Differentiable monte carlo ray tracing through edge sampling. ACM Transactions on Graphics (TOG), 37(6):1– 11, 2018. 2

    2018

  26. [33]

    Model- ing surface appearance from a single photograph using self- augmented convolutional neural networks

    Xiao Li, Yue Dong, Pieter Peers, and Xin Tong. Model- ing surface appearance from a single photograph using self- augmented convolutional neural networks. ACM Transac- tions on Graphics (ToG), 36(4):1–11, 2017. 2

    2017

  27. [34]

    Learning to recon- struct shape and spatially-varying reflectance from a single image

    Zhengqin Li, Zexiang Xu, Ravi Ramamoorthi, Kalyan Sunkavalli, and Manmohan Chandraker. Learning to recon- struct shape and spatially-varying reflectance from a single image. ACM Transactions on Graphics (TOG), 37(6):1–11,

  28. [36]

    Opensvbrdf: A database of measured spatially- varying reflectance

    Xiaohe Ma, Xianmin Xu, Leyao Zhang, Kun Zhou, and Hongzhi Wu. Opensvbrdf: A database of measured spatially- varying reflectance. ACM Trans. Graph, 42(6), 2023. 2

    2023

  29. [37]

    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 syn- thesis. In ECCV, 2020. 1

    2020

  30. [38]

    Practical svbrdf acquisition of 3d objects with unstructured flash photography

    Giljoo Nam, Joo Ho Lee, Diego Gutierrez, and Min H Kim. Practical svbrdf acquisition of 3d objects with unstructured flash photography. ACM Transactions on Graphics (TOG) , 37(6):1–12, 2018. 2

    2018

  31. [39]

    Numerical optimiza- tion

    Jorge Nocedal and Stephen J Wright. Numerical optimiza- tion. Springer, 1999. 14

    1999

  32. [40]

    Unified neural encoding of btfs

    Gilles Rainer, Abhijeet Ghosh, Wenzel Jakob, and Tim Weyrich. Unified neural encoding of btfs. In Computer Graphics Forum , pages 167–178. Wiley Online Library,

  33. [41]

    Extracting the shape and roughness of specular lobe objects using four light pho- tometric stereo

    Fredric Solomon and Katsushi Ikeuchi. Extracting the shape and roughness of specular lobe objects using four light pho- tometric stereo. IEEE Transactions on Pattern Analysis and Machine Intelligence, 18(4):449–454, 1996. 2

    1996

  34. [42]

    Nerv: Neural reflectance and visibility fields for relighting and view synthesis

    Pratul P Srinivasan, Boyang Deng, Xiuming Zhang, Matthew Tancik, Ben Mildenhall, and Jonathan T Barron. Nerv: Neural reflectance and visibility fields for relighting and view synthesis. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pag...

    2021

  35. [43]

    Acquiring reflectance and shape from continuous spheri- cal harmonic illumination

    Borom Tunwattanapong, Graham Fyffe, Paul Graham, Jay Busch, Xueming Yu, Abhijeet Ghosh, and Paul Debevec. Acquiring reflectance and shape from continuous spheri- cal harmonic illumination. ACM Transactions on graphics (TOG), 32(4):1–12, 2013. 2, 3

    2013

  36. [44]

    Measuring and modeling anisotropic re- flection

    Gregory J Ward. Measuring and modeling anisotropic re- flection. In Proceedings of the 19th annual conference on Computer graphics and interactive techniques , pages 265– 272, 1992. 2, 7, 12

    1992

  37. [45]

    Ma- terial classification based on training data synthesized us- ing a btf database

    Michael Weinmann, Juergen Gall, and Reinhard Klein. Ma- terial classification based on training data synthesized us- ing a btf database. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part III 13 , pages 156–171....

    2014

  38. [46]

    Photometric stereo: A reflectance map technique for determining surface orientation from image in- tensity

    Robert J Woodham. Photometric stereo: A reflectance map technique for determining surface orientation from image in- tensity. In Image understanding systems and industrial ap- plications I, pages 136–143. SPIE, 1979. 1

    1979

  39. [47]

    Light sampling field and brdf representation for physically-based neural rendering

    Jing Yang, Hanyuan Xiao, Wenbin Teng, Yunxuan Cai, and Yajie Zhao. Light sampling field and brdf representation for physically-based neural rendering. In The Eleventh Interna- tional Conference on Learning Representations, 2022. 1

    2022

  40. [48]

    Single image surface appearance modeling with self- augmented cnns and inexact supervision

    Wenjie Ye, Xiao Li, Yue Dong, Pieter Peers, and Xin Tong. Single image surface appearance modeling with self- augmented cnns and inexact supervision. In Computer Graphics Forum , pages 201–211. Wiley Online Library,

  41. [49]

    A dif- ferential theory of radiative transfer

    Cheng Zhang, Lifan Wu, Changxi Zheng, Ioannis Gkioulekas, Ravi Ramamoorthi, and Shuang Zhao. A dif- ferential theory of radiative transfer. ACM Transactions on Graphics (TOG), 38(6):1–16, 2019. 2 10 A. Proof In this section, we provide detailed explanations for the equations d...

    2019

Pith tools

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