Pith. sign in

REVIEW 3 major objections 3 minor 41 references

CLONE: Continuous Latent Optimization for Normal Estimation via 3D Gaussian Splatting

T0 review · 3 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Photometric-only training recovers surface normals from 3D Gaussian splatting.

desk verdict Abstract-only, plausibly novel weakly supervised normal-from-3DGS method whose core identifiability claim is unexamined—worth a referee, but only to extract the experiments. read the letter →

arxiv 2508.05950 v3 pith:37HUH62A submitted 2025-08-08 cs.CV cs.AI

Yanxing Liang , Yinghui Wang , Wei Li , Tao Yan , Jiaxing Shen This is my paper
classification cs.CVcs.AI
keywords normalestimation3DGaussiansplattingweaklysupervisedlearningphotometricconsistencydifferentiablerenderingsurfacereconstructionmulti-view
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper proposes CLONE, a weakly supervised framework that estimates surface normals from 3D Gaussian splatting (3DGS) without any ground-truth normal labels. The central claim is that by reparameterizing 3DGS geometry through a differentiable light interaction model, normals become an explicit and stable function of the Gaussian parameters, making the photometric reprojection loss act as internal supervision. Two additional components, a single-step deterministic refinement network and a cross-domain gating fusion, recover high-frequency detail and enforce multi-view consistency. If correct, this means accurate normals can be produced as a byproduct of photometric 3D reconstruction, removing a key annotation bottleneck.

What carries the argument

The load-bearing object is the differentiable light interaction model with a learnable modulation kernel. It provides a unified reparameterization of the 3DGS parameter space so that surface normals are a stable, explicit function of the Gaussian geometry; this is what converts photometric error into gradient signals that supervise normals directly.

What would settle it

Train CLONE on a multi-view dataset with known ground-truth normals and vary lighting conditions; if the photometric loss stays low while the estimated normals deviate from ground truth (or change with lighting), the claimed stable geometric-to-normal mapping does not hold.

Watch

Extended reading notes

Core claim

CLONE establishes an explicit and stable mapping between 3DGS geometric parameters and surface normals via a differentiable light interaction model with a learnable modulation kernel. This reparameterization turns the photometric loss into an internal supervision signal within an image-geometry-image consistency loop. A conditional refinement network then denoises the smooth Gaussian-derived normals and restores high-frequency detail, while a cross-domain gating fusion combines the geometrically consistent 3DGS estimate with the detailed refinement, jointly optimized under a unified photometric reprojection objective with geometric consistency regularizations. The result is an end-to-end, fu

Load-bearing premise

That photometric consistency under the learnable light interaction model provides enough signal to pin down true surface normals; if the light model absorbs shading variation, the normals become under-constrained and may only match observed viewpoints.

Editorial extensions

If this is right

  • Normal estimation becomes label-free for 3DGS-based reconstruction: any photometric multi-view setup can produce normals.
  • The recovered normals are not limited by Gaussian smoothness; the refinement network restores high-frequency detail while reprojection keeps them consistent.
  • The gating fusion gives a principled way to combine a smooth but globally consistent normal field with a detailed but possibly inconsistent one.
  • The closed-loop optimization means the 3DGS geometry itself is regularized by normal consistency, potentially improving overall reconstruction quality.

Reading between the lines

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

  • The same light-interaction reparameterization might extend to other surface intrinsic properties (albedo, roughness), making photometric-only inverse rendering recover a fuller scene model.
  • The main open risk is light-model ambiguity: if the learnable modulation kernel absorbs shading variation, the recovered normals may be correct only for the observed viewpoints; a controlled test with ground-truth normals under varying lighting would settle this.
  • One could expect that integrating CLONE into real-time SLAM or novel-view synthesis systems would give them surface orientation for free, enabling better geometry-aware effects.
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 / 3 minor

Summary. The paper proposes CLONE, a framework for estimating surface normals from 3D Gaussian splatting (3DGS) without ground-truth normal labels. The abstract describes four components: (1) a differentiable light interaction model with a learnable modulation kernel that reparameterizes the 3DGS parameter space to establish a mapping to surface normals; (2) a conditional single-step deterministic refinement network that recovers high-frequency detail; (3) a cross-domain gating fusion mechanism that combines the 3DGS estimate with the refinement; and (4) joint optimization under a unified photometric reprojection objective with geometric consistency regularizations. The central claim is that this closed-loop image-geometry-image consistency enables weakly supervised, photometric-only normal estimation. The manuscript as provided consists solely of the abstract; no method details, equations, experiments, or quantitative results are included.

Significance. If the claimed results are substantiated, the work would be significant: normal estimation from 3DGS without normal labels would remove a major annotation bottleneck and could improve downstream geometry-aware rendering and editing. The proposed closed-loop architecture is plausible and contemporary, and the differentiable rendering pathway is a natural fit for 3DGS. However, the significance cannot be evaluated from the current submission because no evidence is presented. In particular, the core identifiability issue—whether photometric loss with a jointly learned light model can uniquely determine surface normals—is neither analyzed nor experimentally addressed. The significance is therefore potential rather than demonstrated.

major comments (3)
  1. [Abstract / Full Text] The manuscript contains only the abstract; the full text is empty in the provided file. There is no method section, no mathematical formalization, no experimental protocol, no datasets, no baselines, and no quantitative results. The central claims of accurate normal estimation and weakly supervised learning are therefore unsupported. Even for a short paper, the core algorithm and evaluations must be present to be verifiable.
  2. [Abstract, 'explicit and stable mapping'] The claimed 'explicit and stable mapping between 3DGS geometric parameters and surface normals' is not demonstrated. The abstract attributes this mapping to a differentiable light interaction model with a learnable modulation kernel, but no derivation or equation is supplied. More critically, because the light interaction parameters are learned jointly with the normals from the same photometric loss, there is a gauge ambiguity: shading changes due to altered normals can be compensated by altered light parameters (and by other photometry-affecting parameters typical of 3DGS), preserving the rendered images. The manuscript provides neither an identifiability analysis nor a synthetic experiment that rules out such compensation. This is a load-bearing gap, as the photometric loss may then be minimized by normal fields that are image-consistent but not geometrically correct.
  3. [Abstract, 'internal supervision signal'] The phrase 'turning the photometric loss into an internal supervision signal' conflates an optimization objective with an independent supervision signal. Photometric consistency is an objective that many normal fields can satisfy; it does not by itself certify accuracy of the recovered normals. The paper needs to demonstrate, via synthetic experiments with known ground truth and real-world cross-checks, that the optimized normals correspond to true surface geometry rather than merely to rendering-consistent shading. Without such evidence, the weakly supervised claim remains an assertion.
minor comments (3)
  1. [Abstract] The term 'weakly supervised' is used, but the abstract states the method does not rely on external normal labels. If only photometric supervision is used, the method is unsupervised with respect to normals; the nature of the weak supervision should be clarified.
  2. [Abstract] The acronym 3DGS is introduced as '3D Gaussian splatting', which is fine, but the key concept 'continuous latent optimization' is not defined: what is the latent variable being optimized? The abstract should state the optimization variable and its relation to the 3DGS parameters.
  3. [Abstract] No related work is referenced, so the abstract does not position CLONE against existing weakly supervised normal estimation methods (e.g., photometric stereo, shape-from-shading, or self-supervised depth/normal learning). A brief comparison would help assess novelty.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the photometric objective is an optimization target, not a disguised input.

full rationale

The reviewed text (abstract and available manuscript fragments) contains no derivation chain in which a prediction reduces by construction to its own inputs. The central claim is that CLONE reparameterizes 3DGS parameters into surface normals and jointly optimizes them under a photometric reprojection loss with a differentiable light interaction model. This is a weakly supervised learning objective, not a logical derivation of normals from independent geometric labels. The concern that a learnable light model and per-Gaussian shading parameters could absorb photometric residuals, leaving normals underdetermined, is an identifiability/correctness risk, not circularity: the paper does not assert that normals are uniquely forced by an independent first-principles constraint. No fitted parameter is relabeled as a prediction, no load-bearing self-citation appears, no uniqueness theorem is imported from the authors' prior work, and no known result is merely renamed. Without equations or specific claims showing that the loss or regularizer is defined in terms of the target output itself, no circular step can be exhibited under the stated hard rules. Therefore the appropriate finding is no significant circularity.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central method relies entirely on learned components (kernel, refinement net, gating net) and a domain assumption that photometric consistency can supervise normals. No new physical entities are introduced.

free parameters (4)
  • learnable modulation kernel
    Part of the differentiable light interaction model; its parameters are trained against photometric loss, and it is central to the claimed mapping to normals.
  • refinement network weights
    The conditional single-step deterministic refinement network is trained jointly; its weights determine the high-frequency normal estimates.
  • cross-domain gating fusion weights
    The gating mechanism is learned to blend the two normal estimates, and its parameters affect the final output.
  • 3D Gaussian parameters
    The splatting scene parameters are fitted to the input images and are reparameterized to yield normals.
assumptions (3)
  • domain assumption Photometric consistency under a learned light model is a sufficient signal for geometric normal estimation.
    The entire weakly supervised loop depends on this. The abstract states the photometric loss is used as internal supervision, which implies this assumption.
  • domain assumption 3D Gaussian primitives, after reparameterization, can represent true surface geometry and normals.
    The paper claims a mapping between 3DGS geometric parameters and normals, which requires that the representation is expressive enough to encode real normals.
  • domain assumption The light interaction model is differentiable and learnable through photometric loss.
    All components are jointly optimized in a fully differentiable pathway, requiring this to hold.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CLONE: Continuous Latent Optimization for Normal Estimation via 3D Gaussian Splatting." pith.science (2026). https://pith.science/paper/37HUH62A

@misc{pith2026250805950,
  author       = {Pith},
  title        = {Pith review of: CLONE: Continuous Latent Optimization for Normal Estimation via 3D Gaussian Splatting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/37HUH62A}},
  note         = {Machine review of arXiv:2508.05950}
}
read the original abstract

We propose CLONE, a Continuous Latent Optimization framework for Normal Estimation via 3D Gaussian splatting. The core idea is to construct an image-geometry-image consistency loop that unifies explicit geometric representation with differentiable rendering, thereby enabling weakly supervised learning without normal ground truth. Specifically, CLONE comprises four components. First, by introducing a differentiable light interaction model with a learnable modulation kernel, we perform a unified reparameterization of the 3DGS parameter space, establishing an explicit and stable mapping between 3DGS geometric parameters and surface normals and turning the photometric loss into an internal supervision signal. Second, the conditional single-step deterministic refinement network integrates denoising architectures with differentiable reprojection constraints to refine the initial normals, thereby adaptively recovering the high-frequency details erased by the inherently smooth Gaussian primitives. Third, the cross-domain gating fusion mechanism adaptively combines the two complementary normal estimates while imposing multi-view reprojection consistency and implicit geometric regularization, reconciling the geometrically consistent yet over-smooth 3DGS estimate with the detailed yet potentially geometry-inconsistent refinement. Finally, all components are jointly optimized under a unified photometric reprojection objective with geometric consistency regularizations in a fully differentiable pathway, and the directional regularization aligns the learnable principal directions with the geometric normals, achieving an end-to-end optimization closed loop without relying on external normal labels.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 37 canonical work pages

  1. [1]

    Omnidata: A scalable pipeline for making multi-task mid-level vision datasets from 3d scans

    Ahmad Eftekhar, Abhinav Sax, Jitendra Malik, and Arsha Zamir. Omnidata: A scalable pipeline for making multi-task mid-level vision datasets from 3d scans. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 10786--10796, Nashville, TN, USA, 2021

    work page 2021

  2. [2]

    Objaverse: A universe of annotated 3d objects

    Matt Deitke, Dustin Schwenk, Jordi Salvador, Luca Weihs, Oscar Michel, Eli VanderBilt, Ludwig Schmidt, Kiana Ehsani, Aniruddha Kembhavi, and Ali Farhadi. Objaverse: A universe of annotated 3d objects. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , Vancouver, Canada, 2023. CVPR

    work page 2023

  3. [3]

    Discriminatively trained dense surface normal estimation

    Lubor Ladicky, Bernhard Zeisl, and Marc Pollefeys. Discriminatively trained dense surface normal estimation. In Bastian Leibe, Jiri Matas, Nicu Sebe, and Max Welling, editors, Computer Vision -- ECCV 2014 , volume 8689 of Lecture Notes in Computer Science , pages 468--484, Zurich, Switzerland, 2014. Springer

    work page 2014

  4. [4]

    Designing deep networks for surface normal estimation

    Xiaolong Wang, David Fouhey, and Abhinav Gupta. Designing deep networks for surface normal estimation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 539--547, Boston, MA, USA, 2015

    work page 2015

  5. [5]

    Predicting depth, surface normals and semantic labels with a common multi-scale convolutional architecture

    David Eigen and Rob Fergus. Predicting depth, surface normals and semantic labels with a common multi-scale convolutional architecture. In Proceedings of the IEEE International Conference on Computer Vision , pages 2650--2658, Santiago, Chile, 2015

    work page 2015

  6. [6]

    Depth and surface normal estimation from monocular images using regression on deep features and hierarchical crfs

    Boyang Li, Chunhua Shen, Yong Dai, Anton van den Hengel, and Ming-Ming He. Depth and surface normal estimation from monocular images using regression on deep features and hierarchical crfs. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 1119--1127, Boston, MA, USA, 2015

    work page 2015

  7. [7]

    Russell, and Abhinav Gupta

    Aayush Bansal, Bryan C. Russell, and Abhinav Gupta. Marr revisited: 2d–3d alignment via surface normal prediction. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 5965--5974, Las Vegas, NV, USA, 2016

    work page 2016

  8. [8]

    Enforcing geometric constraints of virtual normal for depth prediction

    Wen Yin, Yanyu Liu, Chunhua Shen, and Yonggang Yan. Enforcing geometric constraints of virtual normal for depth prediction. In Proceedings of the IEEE International Conference on Computer Vision , pages 5684--5693, Seoul, Republic of Korea, 2019

    work page 2019

Show all 41 references
  1. [9]

    Polytransform: Deep polygon transformer for instance segmentation

    Justin Liang, Namdar Homayounfar, Wei-Chiu Ma, Yuwen Xiong, Rui Hu, and Raquel Urtasun. Polytransform: Deep polygon transformer for instance segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 9128--9137, Seattle, WA, USA, 2020

    2020

  2. [10]

    Planetr: Structure-guided transformers for 3d plane recovery

    Zhi Shi, Yanyu Liu, and Xi Zhang. Planetr: Structure-guided transformers for 3d plane recovery. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 1234--1245, Montreal, Canada, 2021

    2021

  3. [11]

    Estimating and exploiting the aleatoric uncertainty in surface normal estimation

    Geonmo Bae, Ignas Budvytis, and Roberto Cipolla. Estimating and exploiting the aleatoric uncertainty in surface normal estimation. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 13137--13146, Montreal, Canada, 2021

    2021

  4. [12]

    Geonet: Geometric neural network for joint depth and surface normal estimation

    Xiaoqing Qi, Ruofan Liao, Zhe Liu, Raquel Urtasun, and Jiaya Jia. Geonet: Geometric neural network for joint depth and surface normal estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 283--291, Salt Lake City, UT, USA, 2018

    2018

  5. [13]

    Xiaoqing Qi, Zhe Liu, Ruofan Liao, Philip H. S. Torr, Raquel Urtasun, and Jiaya Jia. Geonet++: Iterative geometric neural network with edge-aware refinement for joint depth and surface normal estimation. IEEE Transactions on Pattern Analysis and Machine Intelligence , 44(2):96...

    2020

  6. [14]

    Adaptive surface normal constraint for geometric estimation from monocular images

    Xiaoxiao Long, Yiru Zheng, Yuchen Zheng, Bowen Tian, Chenglei Lin, Li Liu, Hu Zhao, Guang Zhou, and Wei Wang. Adaptive surface normal constraint for geometric estimation from monocular images. arXiv preprint arXiv:2402.05869, 2024

    2024 arXiv

  7. [15]

    Wonder3d: Single image to 3d using cross-domain diffusion

    Xianzhi Long, Yizhi Guo, Cheng Lin, and et al. Wonder3d: Single image to 3d using cross-domain diffusion. arXiv preprint arXiv:2310.15008, 2023

    2023 arXiv

  8. [16]

    McHugh, and Vincent Vanhoucke

    Laura Downs, Anthony Francis, Nate Koenig, Brandon Kinman, Ryan Hickman, Krista Reymann, Thomas B. McHugh, and Vincent Vanhoucke. Google scanned objects: A high-quality dataset of 3d scanned household items. In Proceedings of the IEEE International Conference on Robotics and A...

    2022

  9. [17]

    Syncdreamer: Generating multiview-consistent images from a single-view image

    Yuan Liu, Cheng Lin, Zijiao Zeng, Xiaoxiao Long, Lingjie Liu, Taku Komura, and Wenping Wang. Syncdreamer: Generating multiview-consistent images from a single-view image. arXiv preprint arXiv:2309.03453, 2023

    2023 arXiv

  10. [18]

    Magic3d: High-resolution text-to-3d content creation

    Chen-Hsuan Lin, Jun Gao, Luming Tang, Towaki Takikawa, Xiaohui Zeng, Xun Huang, Karsten Kreis, Sanja Fidler, Ming-Yu Liu, and Tsung-Yi Lin. Magic3d: High-resolution text-to-3d content creation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognitio...

    1998

  11. [19]

    Zero-1-to-3: Zero-shot one image to 3d object

    Rui Liu, Rui Wu, Bram Van Hoorick, and et al. Zero-1-to-3: Zero-shot one image to 3d object. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 9298--9309, Paris, France, 2023

    2023

  12. [20]

    Zero123++: A single image to consistent multi-view diffusion base model

    Rui Shi, Hao Chen, Zhi Zhang, and et al. Zero123++: A single image to consistent multi-view diffusion base model. arXiv preprint arXiv:2310.15110, 2023

    2023 arXiv

  13. [21]

    Realfusion: A reliable deep learning-based spatiotemporal fusion framework for generating seamless fine-resolution imagery

    Da Guo, Zhaoxin Li, Xiaowei Gao, and et al. Realfusion: A reliable deep learning-based spatiotemporal fusion framework for generating seamless fine-resolution imagery. Remote Sensing of Environment , 321:114689, 2025

    2025

  14. [22]

    One-2-3-45: Any single image to 3d mesh in 45 seconds without per-shape optimization

    Minghua Liu, Chao Xu, Haian Jin, Linghao Chen, Zexiang Xu, and Hao Su. One-2-3-45: Any single image to 3d mesh in 45 seconds without per-shape optimization. arXiv preprint arXiv:2306.16928, 2023

    2023 arXiv

  15. [23]

    Metric3d: Towards zero-shot metric 3d prediction from a single image

    Wen Yin, Cheng Zhang, Hao Chen, and et al. Metric3d: Towards zero-shot metric 3d prediction from a single image. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 1--12, Paris, France, 2023

    2023

  16. [24]

    Metric3d v2: A versatile monocular geometric foundation model for zero-shot metric depth and surface normal estimation

    Wen Yin, Cheng Zhang, Hao Chen, and et al. Metric3d v2: A versatile monocular geometric foundation model for zero-shot metric depth and surface normal estimation. 2024

    2024

  17. [25]

    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. Communications of the ACM , 65(1):99--106, 2021

    2021

  18. [26]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl and et al. 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics , 42(4):Article 1, 2023

    2023

  19. [27]

    Principles of Optics

    Max Born and Emil Wolf. Principles of Optics . Cambridge University Press, Cambridge, UK, 7th edition, 1999

    1999

  20. [28]

    Physically Based Rendering: From Theory to Implementation

    Matt Pharr, Wenzel Jakob, and Greg Humphreys. Physically Based Rendering: From Theory to Implementation . The MIT Press, Cambridge, MA, USA, 4th edition, 2023

    2023

  21. [29]

    Rumelhart, Geoffrey E

    David E. Rumelhart, Geoffrey E. Hinton, and Ronald J. Williams. Learning representations by back-propagating errors. Nature , 323(6088):533--536, 1986

    1986

  22. [30]

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

    Alexey Dosovitskiy and et al. An image is worth 16×16 words: Transformers for image recognition at scale. In International Conference on Learning Representations , 2021

    2021

  23. [31]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 770--778, Las Vegas, NV, USA, 2016

    2016

  24. [32]

    Feature pyramid networks for object detection

    Tsung - Yi Lin, Piotr Doll \'a r, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyramid networks for object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 2117--2125, Honolulu, HI, USA, 2017

    2017

  25. [33]

    Jolliffe

    Ian T. Jolliffe. Principal Component Analysis . Springer, New York, NY, USA, 2nd edition, 2002

    2002

  26. [34]

    P. C. Mahalanobis. On the generalized distance in statistics. Proceedings of the National Institute of Sciences of India , 2(1):49--55, 1936

    1936

  27. [35]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems , pages 6840--6851, 2020

    2020

  28. [36]

    Towards practical physical-optics rendering

    Shlomi Steinberg, Pradeep Sen, and Ling-Qi Yan. Towards practical physical-optics rendering. ACM Transactions on Graphics (TOG) , 41(4):1--24, 2022. Proceedings of SIGGRAPH 2022

    2022

  29. [37]

    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. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages ...

    2022

  30. [38]

    Normal-gs: 3d gaussian splatting with normal-involved rendering

    Meng Wei, Qianyi Wu, Jianmin Zheng, Hamid Rezatofighi, and Jianfei Cai. Normal-gs: 3d gaussian splatting with normal-involved rendering. In Advances in Neural Information Processing Systems (NeurIPS) , 2024. arXiv preprint arXiv:2410.20593

    2024 arXiv

  31. [39]

    Position-normal distributions for efficient rendering of specular microstructure

    Ling-Qi Yan, Miloš Hašan, Steve Marschner, and Ravi Ramamoorthi. Position-normal distributions for efficient rendering of specular microstructure. ACM Transactions on Graphics (TOG) , 35(4):1--9, 2016. Proceedings of SIGGRAPH 2016

    2016

  32. [40]

    Rendering specular microgeometry with wave optics

    Ling-Qi Yan, Miloš Hašan, Bruce Walter, Steve Marschner, and Ravi Ramamoorthi. Rendering specular microgeometry with wave optics. ACM Transactions on Graphics (TOG) , 37(4):1--10, 2018. Proceedings of SIGGRAPH 2018

    2018

  33. [41]

    Unified gaussian primitives for scene representation and rendering

    Yang Zhou, Songyin Wu, and Ling-Qi Yan. Unified gaussian primitives for scene representation and rendering. arXiv preprint arXiv:2406.09733 , 2024

    2024 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.