Pith. sign in

REVIEW 3 major objections 5 minor 39 references

The paper claims that a single hemisphere of reflective views is enough to recover a high-frequency environment map, with sub-5° normal error and a nearly 5 dB gain over prior reflective reconstruction methods.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review

2026-08-01 02:37 UTC pith:332WDTFH

load-bearing objection A solid, honest partial-view environment map method that only works for mirror-like surfaces; the 5 dB claim rests on a small custom benchmark and no code, and the glass-facade framing overreaches. the 3 major comments →

arxiv 2607.25362 v1 pith:332WDTFH submitted 2026-07-28 cs.CV

PanoLess: Environment Reconstruction from Partial Reflective Views

Ahitagni Das , Ashok Veeraraghavan , Vivek Boominathan This is my paper
classification cs.CV
keywords environment map reconstructionGaussian splattingreflective surfacesneural cubemapvisibility mapinverse renderingpartial-view reconstructionspecular reflections
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

PanoLess aims to establish that the reflections on one side of a shiny surface act as an indirect camera: from images taken only in a single hemisphere, the surrounding environment can be reconstructed as a distant-illumination map without panning or panoramic capture. The method couples surface-aligned 2D Gaussian splats—which provide explicit, consistent surface normals—to a learnable neural cubemap that is queried directly by each pixel's mirror reflection direction, so high-frequency lighting is stored without an intermediate network. It also produces a visibility map that records which environment directions were actually observed, letting users distinguish trustworthy lighting from extrapolated regions. On the paper's synthetic Shiny Partial benchmark, PanoLess reports roughly 5 dB higher environment-map PSNR than the best baseline and mean normal error below 5°, and real handheld captures indicate the approach transfers beyond synthetic data. A reader should care because this turns ordinary shiny objects into cheap environment sensors for relighting, AR, and scene understanding.

Core claim

PanoLess discovers that partial reflective views can recover coherent surrounding illumination, provided the scene uses surface-aligned primitives. The environment is stored as a jointly optimized six-face neural cubemap; each pixel's mirror reflection direction, from an explicit normal, samples the cubemap directly, and rendered color is diffuse albedo plus that radiance. A visibility map marks which cubemap directions observations support. On its benchmark the method reports a nearly 5 dB environment-map PSNR gain over reflection-aware baselines, about 89% of normals within 5°, and generalization to real handheld captures. The paper argues via ablations and a pose-jitter sweep that normal

What carries the argument

The load-bearing mechanism is the direct, unattenuated coupling between a per-pixel mirror reflection direction and a learnable six-face cubemap. Surface-aligned 2D Gaussian splats give explicit normals n; the specular reflection direction is ω_r = normalize(v − 2(v·n)n), computed in screen space and used to bilinearly sample a trainable 6×3×L×L cubemap, with final color C = A + L_env(ω_r)—no roughness attenuation, no MLP blending. Two supporting mechanisms carry the geometry: a depth-normal consistency loss that stabilizes the normals, and a visibility map that accumulates per-texel evidence during training so observed and extrapolated environment directions can be distinguished.

Load-bearing premise

The entire pipeline rests on the assumption that the observed surface is mirror-like: each pixel's color is the diffuse albedo plus the environment radiance along one mirror direction, with no roughness attenuation, so on matte or rough surfaces the environment map receives almost no supervision and the central reconstruction claim weakens.

What would settle it

Render Shiny Partial Vase with a rough or matte BRDF of increasing roughness while keeping everything else fixed: if environment-map PSNR does not drop sharply even as surface geometry stays accurate, the unit-reflection assumption is not what carries the result; if it does, the assumption is confirmed as load-bearing. A complementary check is the paper's own pose-jitter sweep, which drops envmap PSNR from 18.4 to 12.0 dB at only 0.5° camera rotation.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Environment reconstruction no longer requires 360° or panoramic capture for mirror-like surfaces; a single hemisphere of views suffices.
  • Curved reflectors act as wide-angle mirrors, illuminating large portions of the environment from limited viewpoints, while planar mirrors illuminate a narrow band—knowledge the visibility map makes explicit.
  • The visibility map gives downstream relighting, AR insertion, and scene understanding a direct confidence signal for each environment direction.
  • Removing roughness-dependent attenuation improves supervision of the cubemap; small roughness values otherwise starve the environment map of gradients.
  • Normal accuracy is the bottleneck: mean normal error below 5° is needed to keep reflection queries coherent, and pose accuracy is a practical prerequisite.

Where Pith is reading between the lines

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

  • The same visibility signal could gate a generative prior for unseen environment regions: only apply diffusion-based inpainting where the map flags low confidence, keeping observed directions untouched.
  • The method's direct cubemap supervision suggests a general recipe: for other physically based shading terms, avoid learned multiplicative weights when the target material is near one end of the parameter range, since attenuation weakens gradients exactly where supervision is sparse.
  • A natural stress test is to sweep roughness or diffuse fraction on the Shiny Partial benchmark; the expected failure boundary should track where the albedo channel absorbs the photometric signal.
  • Because reflection directions compress viewpoint motion, the framework may transfer to mapping environments from moving specular objects in the wild, with the visibility map acting as an angular coverage sensor.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes PanoLess, a Gaussian-splatting framework that reconstructs a distant environment map from partial, single-hemisphere views of reflective surfaces. The method represents the reflector with surface-aligned 2D Gaussian splats, rasterizes per-pixel normal/albedo/roughness buffers, and defers shading through a learnable neural cubemap queried at the specular reflection direction. A visibility map records which environment directions are supported by observations, and an early silhouette loss anchors geometry. Experiments on a custom synthetic benchmark (Shiny Partial, three scenes) report a nearly 5 dB improvement over the strongest baseline in recovered environment PSNR and mean normal angular error below 5°, alongside ablations isolating normal consistency, silhouette supervision, and the direct (unattenuated) cubemap pathway. Qualitative results on real captures and a mixed-specularity subset (Partial Shiny Blender) are also presented.

Significance. If the results hold, PanoLess demonstrates a practically useful capability: recovering coherent high-frequency surrounding illumination from one-sided observations of highly specular surfaces, a setting where existing reflection-aware Gaussian-splatting methods degrade. The paper is honestly written: the ablations are informative, the matched-silhouette baseline (Table 2) improves fairness, and the pose-jitter study (S5) makes a predicted failure mode explicit. The visibility map is a potentially valuable output for downstream applications. However, the central claim is only established for the near-mirror regime; the paper's own experiments show sharp degradation outside it, and the quantitative benchmark is small and without error bars. The strengths are real but the scope of the claim needs to be reconciled with the evidence.

major comments (3)
  1. [Abstract; Sec. 1; Sec. 3.4, Eq. (8); Sec. 4.5; Fig. 8; S3] The abstract and introduction motivate 'glass facades', but the shading model in Eq. (8), C(p)=A(p)+L_env(ω_r(p)), assumes an opaque, perfectly specular surface with unit reflection strength and no Fresnel or transmission term. For glass, the Fresnel reflectance is well below 1 and the transmitted background is viewpoint-dependent; the constant albedo channel cannot absorb that variation, so the photometric loss will push transmitted signal into the cubemap and corrupt the recovered map. The paper's own results confirm this: Sec. 4.5 states the method 'breaks down as surfaces become more diffuse,' Fig. 8 shows the coffee cup produces no usable env signal, and S3 reports darker, less faithful maps on Partial Shiny Blender. The nearly 5 dB improvement is therefore established only for mirror-like surfaces; the abstract's 'glass facades' claim is unsupported. Either add a quantitative glass
  2. [Table 1; Table 3; S5] The main quantitative claim rests on a custom benchmark with only three scenes, and all numbers are single-run point estimates with no standard deviations, seeds, or significance tests. This is particularly concerning given the demonstrated sensitivity to pose error (S5: envmap PSNR drops from 18.4 to 12.0 dB under a 0.5° rotation perturbation). The 'nearly 5 dB' average is dominated by one favorable configuration. Please report repeated runs (e.g., at least 3 seeds per scene and per ablation) and state the variance, or otherwise justify stability.
  3. [Sec. 3.6; Sec. 4.2; S4] The visibility map is presented as a key contribution, but its only quantitative evaluation is a small post-processing gain on a single scene (Cola, S4: +0.28 dB envmap PSNR). The claim that the map is 'well-calibrated' is not supported by any calibration metric, such as correlation between V and per-texel reconstruction error. As a named contribution, it needs more direct evidence; otherwise it should be repositioned as a qualitative byproduct.
minor comments (5)
  1. [Sec. 3.4, Eq. (6)] The statement that the reflection-direction error is 'largest at grazing angles' appears to contradict the bound in Eq. (6), where ∥δω_r∥ ≤ 2(|v·n|+1)∥δn∥ is maximized at normal incidence (|v·n|=1) and minimized at grazing. Please check the direction of this remark.
  2. [Sec. 4.4] The observation that albedo maps are 'low-energy and nearly uniform' is presented as validation of Eq. (8), but it only shows that the optimizer exploits the direct cubemap pathway; it does not establish physical correctness. Consider tempering this claim.
  3. [Sec. 4.3; Fig. 6] The Shiny Real results are qualitative only; no ground-truth environment map or quantitative metric is reported. The main text should state this explicitly to avoid implying quantitative validation on real data.
  4. [Fig. 3] The visibility map panels lack a color scale or legend. Please add one so the reader can interpret the confidence values.
  5. [Sec. 4.1] The per-scene gains over MaterialRefGS range from about 2.5 dB (Cola) to 7.5 dB (Mirror), so 'nearly 5 dB' is an average over three scenes. Reporting per-scene deltas alongside the average would give a clearer picture.

Circularity Check

0 steps flagged

No significant circularity: the environment map is optimized from images and evaluated against held-out ground truth; Eq. 6 is a derived bound, not a fitted prediction.

full rationale

PanoLess's derivation chain is self-contained. The environment map E is a free parameter optimized through the photometric loss (Eqs. 9-14) against training images; no term in the objective supervises E directly with the ground-truth cubemap, and the reported environment-map metrics are evaluated on held-out test views and against ground-truth environment maps (Table 1). Eq. (6) is a first-order error-propagation bound obtained algebraically from the reflection equation (Eq. 5), not a fitted quantity, and the pose-jitter experiment (S5) is an independent empirical probe. The visibility map is explicitly defined as accumulated evidence (Sec. 3.6) and is used as a confidence diagnostic, not disguised as an independent prediction. All load-bearing building blocks (2DGS [10], 3DGS [16], normal propagation [36], Ref-NeRF [30]) are cited external works with no author overlap; there is no self-citation chain, imported uniqueness theorem, or ansatz smuggled via self-citation. The admitted limitation in Sec. 4.5—that the unit-reflectance model breaks down as surfaces become more diffuse—is an external-validity boundary, not a circular step. The abstract's glass-facade generalization is likewise a validation gap rather than a derivation that reduces to its own inputs. Therefore no step reduces to its own inputs by construction, and the circularity score is 0.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 0 invented entities

The method rests on a deliberately scoped mirror-like, far-field shading model and on the accuracy of estimated surface normals and poses. The free parameters are mostly hand-set hyperparameters; the cubemap and visibility map are learned representations rather than new physical entities.

free parameters (6)
  • lambda_d = 0.5
    Depth blending weight in Eq. 3 mixing expected and median depth; chosen by hand.
  • lambda_s = 0.2
    Weight of the SSIM term in the photometric loss (Eq. 9); standard value.
  • lambda_nd = not reported
    Weight of the depth-normal consistency regularizer (Eq. 12); value absent from the paper.
  • silhouette_loss_window = 3000 iterations
    BCE alpha loss is active only for the first 3000 iterations (Eq. 13); chosen to stabilize early training.
  • visibility_threshold_tau = not reported
    Opacity threshold used to scatter visibility increments (Sec. 3.6); needed to reproduce the visibility map.
  • cubemap_resolution_L = not reported
    Grid size per cube face for the environment and visibility maps; affects high-frequency fidelity and is not specified.
axioms (5)
  • domain assumption Environment radiance depends only on direction (far-field/distant illumination); the reflector is small relative to scene distance.
    Invoked in Sec. 3.4 to justify the cubemap representation; invalid for nearby light sources and large reflectors.
  • ad hoc to paper Additive shading with unit reflection strength and negligible roughness: C = A + L_env(omega_r), with no full BRDF integral.
    Eq. 8 is the core rendering model; the authors acknowledge it diverts supervision on diffuse surfaces (S3, Fig. 8).
  • domain assumption Surface-aligned 2DGS normals, after depth-normal regularization, are accurate enough to index the cubemap.
    The whole pipeline depends on Eq. 5-6 reflection directions from estimated normals; support is empirical (Table 1, L_n ablation) rather than theoretical.
  • domain assumption Camera poses from COLMAP and object masks from SAM are accurate; the scene is static.
    Used in all real and synthetic experiments; S5 shows sharp degradation at 0.5 degrees of pose rotation.
  • standard math Standard alpha compositing, bilinear cubemap sampling, and the reflection law.
    Background math for Eqs. 1, 5, and 7; uncontroversial.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of PanoLess: Environment Reconstruction from Partial Reflective Views." pith.science (2026). https://pith.science/paper/332WDTFH

@misc{pith2026260725362,
  author       = {Pith},
  title        = {Pith review of: PanoLess: Environment Reconstruction from Partial Reflective Views},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/332WDTFH}},
  note         = {Machine review of arXiv:2607.25362}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Reflections from shiny objects and glass facades naturally extend the field of view of a camera, capturing the surrounding environment without the need to pan the camera or acquire a full panorama. We propose PanoLess, a Gaussian-splat-based framework that reconstructs the surrounding environment as a distant illumination map from images captured on only one side of a reflective surface. PanoLess leverages surface-aligned 2D Gaussian splats with deferred shading to recover accurate per-pixel normals and reflection cues, which are fused into a neural cubemap representation of the environment. In addition, PanoLess produces a visibility map that explicitly denotes which regions of the environment are supported by the partial reflective observations. Unlike existing inverse-rendering and reflection-aware Gaussian-splatting approaches, which typically require full 360-degree coverage and struggle under incomplete views, PanoLess enables consistent, physically grounded illumination estimation from partial-view input. We show that PanoLess achieves high-fidelity and geometrically consistent environment reconstruction, outperforming reflection-aware baselines on a new custom synthetic benchmark and publicly available datasets, and demonstrating generalization to real-world reflective captures.

Figures

Figures reproduced from arXiv: 2607.25362 by Ahitagni Das, Ashok Veeraraghavan, Vivek Boominathan.

Figure 1
Figure 1. Figure 1: PanoLess recovers the surrounding environment from partial views of a re￾flective object—without panoramic capture. From images taken on one side of a shiny surface, PanoLess estimates an environment map alongside a visibility map that dis￾tinguishes well-observed regions from unsupported ones. 1 Introduction Reflective surfaces such as glass façades, metals, and polished objects offer a unique opportunity… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our rendering pipeline. Given a set of optimizable 2D Gaus￾sians, we rasterize per-Gaussian properties to obtain normal, albedo, roughness, depth, and alpha maps. Surface normals and view directions give reflection directions, which query a learnable cubemap encoder to produce the specular color; this is composed with the diffuse albedo to yield the final render. Our pipeline consists of three … view at source ↗
Figure 3
Figure 3. Figure 3: Recovered environment maps on Shiny Partial. For each scene we show the ground-truth environment map, our reconstruction, our visibility map (bright → well￾observed), and the environment maps extracted from each baseline. The PSNR against the G.T. environment map is mentioned for each on the top left. estimated normal, so even small angular errors redirect the cubemap query and corrupt the recovered illumi… view at source ↗
Figure 4
Figure 4. Figure 4: Per-pixel normal angular error on Shiny Partial. [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Novel-view renders and surface normals on Shiny Partial [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Environment reconstruction on Shiny Real. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: (a) Ablation study on the Vase scene. Each row removes or modifies one component of the full model: ρ-weighted E scales the cubemap contribution by (1−ρ) 2 rather than using it directly, w/o Lα removes early silhouette supervision, and w/o Ln removes normal consistency. We show the recovered environment map, visibility map, surface normals, and a novel-view render for each variant. (b) Output decomposition… view at source ↗
Figure 8
Figure 8. Figure 8: Failure case. On a coffee cup scene, the coffee surface lacks metallic reflections, providing no signal for environment reconstruction. Our method is best suited for highly specular surfaces. PanoLess is designed for mirror-like surfaces: the unity reflection strength of Eq. 8 holds for chrome, polished metal, and glass, but breaks down as surfaces become more diffuse. To evaluate how the method generalize… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

39 extracted references · 2 linked inside Pith

  1. [1]

    IEEE TPAMI25(2), 218–233 (2003)

    Basri, R., Jacobs, D.W.: Lambertian reflectance and linear subspaces. IEEE TPAMI25(2), 218–233 (2003)

    2003

  2. [2]

    Communications of the ACM19(10), 542–547 (1976) 16 Ahitagni Das , Ashok Veeraraghavan, and Vivek Boominathan

    Blinn, J.F., Newell, M.E.: Texture and reflection in computer generated images. Communications of the ACM19(10), 542–547 (1976) 16 Ahitagni Das , Ashok Veeraraghavan, and Vivek Boominathan

    1976

  3. [3]

    IEEE Transactions on Visualization and Computer Graphics (2025)

    Chen, D., Li, H., Ye, W., Wang, Y., Xie, W., Zhai, S., Wang, N., Liu, H., Bao, H., Zhang, G.: Pgsr: Planar-based gaussian splatting for efficient and high-fidelity sur- face reconstruction. IEEE Transactions on Visualization and Computer Graphics (2025)

    2025

  4. [4]

    In: SIGGRAPH (2024)

    Dai, P., Xu, J., Xie, W., Liu, X., Wang, H., Xu, W.: High-quality surface recon- struction using gaussian surfels. In: SIGGRAPH (2024)

    2024

  5. [5]

    In: SIGGRAPH

    Debevec, P.: Rendering synthetic objects into real scenes: global illumination and hdr photography. In: SIGGRAPH. pp. 189–198 (1998)

    1998

  6. [6]

    In: SIGGRAPH

    Debevec, P., Hawkins, T., Tchou, C., Duiker, H.P., Sarokin, W., Sagar, M.: Ac- quiring the reflectance field of a human face. In: SIGGRAPH. pp. 145–156 (2000)

    2000

  7. [7]

    In: ECCV (2024)

    Gao, J., Gu, C., Lin, Y., Li, Z., Zhu, H., Cao, X., Zhang, L., Yao, Y.: Relightable 3d gaussians: Realistic point cloud relighting with brdf decomposition and ray tracing. In: ECCV (2024)

    2024

  8. [8]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2025)

    Gu, C., Wei, X., Zeng, Z., Yao, Y., Zhang, L.: Irgs: Inter-reflective gaussian splat- ting with 2d gaussian ray tracing. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2025)

    2025

  9. [9]

    ACM Transactions on Graph- ics40(6), 1–11 (2021)

    Hadadan, S., Chen, S., Zwicker, M.: Neural radiosity. ACM Transactions on Graph- ics40(6), 1–11 (2021)

    2021

  10. [10]

    In: SIGGRAPH (2024)

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

    2024

  11. [11]

    IEEE Transactions on Pattern Analysis and Machine IntelligencePAMI-3(6), 661–669 (1981)

    Ikeuchi, K.: Determining surface orientations of specular surfaces by using the photometric stereo method. IEEE Transactions on Pattern Analysis and Machine IntelligencePAMI-3(6), 661–669 (1981)

    1981

  12. [12]

    In: CVPR

    Jensen, R., Dahl, A., Vogiatzis, G., Tola, E., Aanæs, H.: Large scale multi-view stereopsis evaluation. In: CVPR. pp. 406–413 (2014)

    2014

  13. [13]

    In: European Conference on Computer Vision (ECCV) (2024)

    Ji, C., Li, Y., Liao, Y.: Reframe: Reflective surface real-time rendering for mobile devices. In: European Conference on Computer Vision (ECCV) (2024)

    2024

  14. [14]

    In: CVPR (2024)

    Jiang, Y., Tu, J., Liu, Y., Gao, X., Long, X., Wang, W., Ma, Y.: Gaussianshader: 3d gaussian splatting with shading functions for reflective surfaces. In: CVPR (2024)

    2024

  15. [15]

    In: Eurographics Workshop on Rendering

    Kautz, J., Sloan, P.P., Snyder, J.: Fast, arbitrary brdf shading for low-frequency lighting using spherical harmonics. In: Eurographics Workshop on Rendering. pp. 291–296 (2002)

    2002

  16. [16]

    In: SIGGRAPH (2023)

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. In: SIGGRAPH (2023)

    2023

  17. [17]

    In: SIGGRAPH Asia (2018)

    Li,Z.,Xu,Z.,Ramamoorthi,R.,Sunkavalli,K.,Chandraker,M.:Learningtorecon- struct shape and spatially-varying reflectance from a single image. In: SIGGRAPH Asia (2018)

    2018

  18. [18]

    In: ICCV

    Liang, R., Chen, H., Li, C., Chen, F., Panneer, S., Vijaykumar, N.: Envidr: Im- plicit differentiable renderer with neural environment lighting. In: ICCV. pp. 79–89 (2023)

    2023

  19. [19]

    In: CVPR (2024)

    Liang, Z., Zhang, Q., Feng, Y., Shan, Y., Jia, K.: Gs-ir: 3d gaussian splatting for inverse rendering. In: CVPR (2024)

    2024

  20. [20]

    In: ACM SIGGRAPH 2023 Conference Proceedings (2023)

    Liu, Y., Wang, P., Lin, C., Long, X., Wang, J., Liu, L., Komura, T., Wang, W.: Nero: Neural geometry and brdf reconstruction of reflective objects from multiview images. In: ACM SIGGRAPH 2023 Conference Proceedings (2023)

    2023

  21. [21]

    arXiv preprint arXiv:2404.01168 (2024)

    Meng, J., Li, H., Wu, Y., Gao, Q., Yang, S., Zhang, J., Ma, S.: Mirror- 3dgs: Incorporating mirror reflections into 3d gaussian splatting. arXiv preprint arXiv:2404.01168 (2024)

    Pith/arXiv arXiv 2024

  22. [22]

    In: CVPR (2022) PanoLess: Environment Reconstruction from Partial Reflective Views 17

    Munkberg, J., et al.: Extracting triangular 3d models, materials, and lighting from images. In: CVPR (2022) PanoLess: Environment Reconstruction from Partial Reflective Views 17

    2022

  23. [23]

    In: CVPR (2014)

    Oxholm, G., Nishino, K.: Multiview shape and reflectance from natural illumina- tion. In: CVPR (2014)

    2014

  24. [24]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2024)

    Phongthawee, P., Chinchuthakun, W., Sinsunthithet, N., Jampani, V., Raj, A., Khungurn, P., Suwajanakorn, S.: Diffusionlight: Light probes for free by painting a chrome ball. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2024)

    2024

  25. [25]

    In: SIGGRAPH

    Ramamoorthi, R., Hanrahan, P.: A signal-processing framework for inverse ren- dering. In: SIGGRAPH. pp. 117–128 (2001)

    2001

  26. [26]

    In: SIGGRAPH

    Sloan, P.P., Kautz, J., Snyder, J.: Precomputed radiance transfer for real-time rendering in dynamic, low-frequency lighting environments. In: SIGGRAPH. pp. 527–536 (2002)

    2002

  27. [27]

    In: CVPR (2021)

    Somanath, G., Kurz, D.: Hdr environment map estimation for real-time augmented reality. In: CVPR (2021)

    2021

  28. [28]

    Springer (2010)

    Szeliski, R.: Computer Vision: Algorithms and Applications. Springer (2010)

    2010

  29. [29]

    arXiv preprint arXiv:2403.17822 (2024)

    Turkulainen, M., Ren, X., Melekhov, I., Seiskari, O., Rahtu, E., Kannala, J.: Dn-splatter: Depth and normal priors for gaussian splatting and meshing. arXiv preprint arXiv:2403.17822 (2024)

    Pith/arXiv arXiv 2024

  30. [30]

    In: CVPR

    Verbin, D., Hedman, P., Mildenhall, B., Zickler, T., Barron, J.T., Srinivasan, P.P.: Ref-nerf: Structured view-dependent appearance for neural radiance fields. In: CVPR. pp. 5481–5490 (2022)

    2022

  31. [31]

    In: IEEE International Conference on Robotics and Automation (ICRA) (2025)

    Xiang, H., Li, X., Cheng, K., Lai, X., Zhang, W., Liao, Z., Zeng, L., Liu, X.: Gaus- sianroom: Improving 3d gaussian splatting with sdf guidance and monocular cues for indoor scene reconstruction. In: IEEE International Conference on Robotics and Automation (ICRA) (2025)

    2025

  32. [32]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition (CVPR) (2025)

    Xie, T., Chen, X., Xu, Z., Xie, Y., Jin, Y., Shen, Y., Peng, S., Bao, H., Zhou, X.: Envgs: Modeling view-dependent appearance with environment gaussian. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition (CVPR) (2025)

    2025

  33. [33]

    In: ECCV (2024)

    Yan, Y., et al.: Street gaussians: Modeling dynamic urban scenes with gaussian splatting. In: ECCV (2024)

    2024

  34. [34]

    In: NeurIPS (2024)

    Yang, Z., Gao, X., Sun, Y., Huang, Y., Lyu, X., Zhou, W., Jiao, S., Qi, X., Jin, X.: Spec-gaussian: Anisotropic view-dependent appearance for 3d gaussian splatting. In: NeurIPS (2024)

    2024

  35. [35]

    In: International Conference on Learning Representations (ICLR) (2025)

    Yao, Y., Zeng, Z., Gu, C., Zhu, X., Zhang, L.: Reflective gaussian splatting. In: International Conference on Learning Representations (ICLR) (2025)

    2025

  36. [36]

    In: SIG- GRAPH (2024)

    Ye, K., Hou, Q., Zhou, K.: 3d gaussian splatting with deferred reflection. In: SIG- GRAPH (2024)

    2024

  37. [37]

    ACM Transactions on Graphics (TOG) (2024)

    Yu, Z., Sattler, T., Geiger, A.: Gaussian opacity fields: Efficient adaptive surface reconstruction in unbounded scenes. ACM Transactions on Graphics (TOG) (2024)

    2024

  38. [38]

    NeurIPS (2025)

    Zhang, W., Tang, J., Zhang, W., Fang, Y., Liu, Y.S., Han, Z.: Materialrefgs: Re- flective gaussian splatting with multi-view consistent material inference. NeurIPS (2025)

    2025

  39. [39]

    Zhang, Y., Chen, A., Wan, Y., Song, Z., Yu, J., Luo, Y., Yang, W.: Ref-gs: Direc- tional factorization for 2d gaussian splatting. In: CVPR (2025) 18 Ahitagni Das , Ashok Veeraraghavan, and Vivek Boominathan S1 Shiny Real Capture Details Capture pipeline.Each scene is recorded as a handheld video with the operator walking in an arc around one side of the r...

    2025

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.