pith. sign in

arxiv: 2511.19202 · v2 · submitted 2025-11-24 · 💻 cs.CV · cs.GR

NVGS: Neural Visibility for Occlusion Culling in 3D Gaussian Splatting

Brent Zoomers , Florian Hahlbohm , Joni Vanherck , Lode Jorissen , Marcus Magnor , Nick Michiels This is my paper

Pith reviewed 2026-05-17 06:08 UTC · model grok-4.3

classification 💻 cs.CV cs.GR
keywords 3D Gaussian SplattingOcclusion CullingNeural VisibilityInstanced RasterizerMLPVRAM OptimizationReal-time RenderingComputer Vision
0
0 comments X

The pith

A small shared MLP learns the viewpoint-dependent visibility of Gaussians to discard occluded primitives before rasterization in 3D Gaussian Splatting.

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

The paper proposes training a compact MLP to approximate which Gaussians are visible from any viewpoint in a scene. Because the network is shared across all copies of the same asset, its size remains modest even when many instances appear together. The MLP is queried after frustum culling but before the rasterizer runs, so fully hidden Gaussians can be dropped from the pipeline. This visibility check is folded into a new instanced software rasterizer that exploits Tensor Cores for speed. The resulting system uses less VRAM and produces higher-quality images than prior methods on composed scenes while working alongside existing level-of-detail techniques.

Core claim

By training a small shared MLP to model the viewpoint-dependent visibility function of all Gaussians across instances of an asset, occluded primitives can be identified and discarded prior to rasterization when the queries are integrated into an instanced software rasterizer that leverages Tensor Cores for efficient computation.

What carries the argument

The occlusion culling MLP that learns the viewpoint-dependent visibility function of Gaussians, queried before the instanced rasterization step.

If this is right

  • Lower VRAM consumption for scenes built from repeated assets
  • Higher image quality than existing state-of-the-art methods for composed scenes
  • Complementary speed and memory gains when paired with level-of-detail strategies
  • Hardware-efficient queries through direct use of Tensor Cores inside the rasterizer

Where Pith is reading between the lines

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

  • The shared-MLP design could be tested on scenes containing many unique assets by training one network per asset class rather than one global network.
  • Because culling happens before rasterization, the technique may reduce per-frame computation time in addition to memory use for interactive walkthroughs.
  • The same visibility prediction pattern might apply to other semi-transparent primitive representations beyond Gaussians if the MLP input features are adapted.

Load-bearing premise

A small shared MLP can accurately learn the viewpoint-dependent visibility function of all Gaussians across instances of an asset so that querying it prior to rasterization safely discards occluded primitives without harming final image quality.

What would settle it

A side-by-side comparison of final rendered images with and without the MLP-based culling step, using PSNR or SSIM on test views containing partial occlusions.

Figures

Figures reproduced from arXiv: 2511.19202 by Brent Zoomers, Florian Hahlbohm, Joni Vanherck, Lode Jorissen, Marcus Magnor, Nick Michiels.

Figure 1
Figure 1. Figure 1: Shows a scene composed of multiple, separately trained 3DGS assets. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Left shows the donut asset from the front. Right we show the donut as if rendered from the front from a different viewpoint. We then show the Gaussians that our approach culls in red, along with what we would optimally render. sians that are not visible to the current viewpoint, which, similar to backface culling for meshes, are located mainly on the backside of the object, as shown in [PITH_FULL_IMAGE:fi… view at source ↗
Figure 3
Figure 3. Figure 3: Left: Avatar rendered at a near and far distance. Right: We show the original render, then the Gaussians that are not used at the far distance marked in red, and on the right the Gaussians that get used at the far distance. 2 [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: We extract per-Gaussian visibility by rendering to a set of training views and baking the visibility information into a lightweight [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Shows GT render using gsplat, V3DG, and our method for close, medium, and far distances. For V3DG and ours, we show the [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Shows how metrics evolve over distance for three scenes from V3DG [ [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Shows render time in milliseconds at 20 relative distances. [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: From left to right, we show the ground truth zero [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: From left to right, we show the ground truth zero [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: From left to right, we show the ground truth zero [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Shows GT render using gsplat, V3DG, and our method for close, medium, and far distances on the [PITH_FULL_IMAGE:figures/full_fig_p014_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Shows GT render using our method with and without MLP(solely using instanced rasterizer) for close, medium, and far distances [PITH_FULL_IMAGE:figures/full_fig_p015_13.png] view at source ↗
read the original abstract

3D Gaussian Splatting can exploit frustum culling and level-of-detail strategies to accelerate rendering of scenes containing a large number of primitives. However, the semi-transparent nature of Gaussians prevents the application of another highly effective technique: occlusion culling. We address this limitation by proposing a novel method to learn the viewpoint-dependent visibility function of all Gaussians in a trained model using a small, shared MLP across instances of an asset in a scene. By querying it for Gaussians within the viewing frustum prior to rasterization, our method can discard occluded primitives during rendering. Leveraging Tensor Cores for efficient computation, we integrate these neural queries directly into a novel instanced software rasterizer. Our approach outperforms the current state of the art for composed scenes in terms of VRAM usage and image quality, utilizing a combination of our instanced rasterizer and occlusion culling MLP, and exhibits complementary properties to existing LoD techniques.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript introduces NVGS, a technique for occlusion culling in 3D Gaussian Splatting by training a small shared MLP to predict viewpoint-dependent visibility of Gaussians across asset instances. This MLP is queried before an instanced software rasterizer to discard occluded primitives, aiming to reduce VRAM usage while maintaining or improving image quality in scenes with multiple composed instances. The approach is claimed to outperform existing methods and complement LoD techniques.

Significance. If validated, the method could significantly advance efficient rendering of complex, instance-heavy scenes in real-time applications by addressing the occlusion culling limitation in semi-transparent 3DGS representations. The use of a compact MLP and Tensor Core integration for neural queries represents a practical innovation that could be adopted in graphics pipelines, provided the culling preserves rendering fidelity.

major comments (3)
  1. The description of how the visibility MLP is supervised is insufficient. For semi-transparent Gaussians, binary visible/occluded labels from depth or ray marching may not ensure that culling leaves the alpha-blended output unchanged. Without explicit loss formulation or false-negative rate measurements on overlapping instances, the claim that querying the MLP safely discards primitives without harming quality cannot be assessed. This is central to the outperformance claim.
  2. No quantitative tables, ablation studies, or error analysis are referenced. The claim of outperforming SOTA in VRAM usage and image quality for composed scenes requires specific metrics (e.g., PSNR, SSIM, memory reduction) compared to baselines. Absence of these undermines verification of the central claim.
  3. The integration of neural queries into the instanced rasterizer is described at high level, but it is unclear how the MLP outputs modify the rasterization process without introducing artifacts in multi-instance occlusions. A concrete example or equation showing the culling decision would strengthen the method.
minor comments (2)
  1. The abstract claims outperformance but does not specify the datasets or baselines used, which would help contextualize the results.
  2. Clarify the input/output dimensions of the shared MLP and how viewpoint dependence is encoded.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment in turn below, clarifying our approach where possible and noting planned revisions to strengthen the presentation.

read point-by-point responses

  1. Referee: The description of how the visibility MLP is supervised is insufficient. For semi-transparent Gaussians, binary visible/occluded labels from depth or ray marching may not ensure that culling leaves the alpha-blended output unchanged. Without explicit loss formulation or false-negative rate measurements on overlapping instances, the claim that querying the MLP safely discards primitives without harming quality cannot be assessed. This is central to the outperformance claim.

    Authors: We agree that the supervision details merit a more explicit treatment. The manuscript currently describes the MLP training at a high level; we will revise the method section to include the precise loss formulation (a combination of visibility supervision and rendering-consistency terms) and report false-negative rates measured on overlapping instances. This addition will directly address concerns about whether culling preserves the alpha-blended result. revision: yes

  2. Referee: No quantitative tables, ablation studies, or error analysis are referenced. The claim of outperforming SOTA in VRAM usage and image quality for composed scenes requires specific metrics (e.g., PSNR, SSIM, memory reduction) compared to baselines. Absence of these undermines verification of the central claim.

    Authors: The manuscript contains quantitative comparisons of PSNR, SSIM, and VRAM usage against baselines on composed scenes, together with initial ablations. We acknowledge that these results are not sufficiently cross-referenced in the text. We will add explicit pointers to the relevant tables and figures and expand the ablation and error-analysis sections in the revision. revision: partial

  3. Referee: The integration of neural queries into the instanced rasterizer is described at high level, but it is unclear how the MLP outputs modify the rasterization process without introducing artifacts in multi-instance occlusions. A concrete example or equation showing the culling decision would strengthen the method.

    Authors: We concur that the integration description can be made more concrete. The MLP produces a per-Gaussian visibility score that is thresholded to decide culling before the instanced rasterizer processes the primitive; this decision is applied uniformly across instances. We will insert an explicit equation for the culling threshold and a worked example of multi-instance occlusion handling in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity; MLP visibility and instanced rasterizer are independent additions

full rationale

The paper proposes training a small shared MLP to predict viewpoint-dependent visibility for Gaussians and integrating it into a new instanced software rasterizer. No equations or steps in the provided abstract or description reduce the visibility output to a quantity already present in the training loss or final image by construction. The central claim relies on an external supervision signal for the MLP (not shown to be derived from the rendering itself) and a novel rasterizer whose behavior is not defined in terms of the culling result. No self-citations, uniqueness theorems, or ansatzes are invoked that would make the derivation load-bearing on prior author work. This is a standard case of an independent architectural addition evaluated on VRAM and quality metrics.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the domain assumption that visibility can be compactly represented by a small shared MLP and that the resulting culling decisions preserve image quality.

axioms (1)
  • domain assumption A small shared MLP can accurately approximate the viewpoint-dependent visibility function of Gaussians across multiple instances of the same asset.
    Invoked to justify training one network per asset type instead of per Gaussian or per scene.
invented entities (1)
  • Neural Visibility MLP no independent evidence
    purpose: Predicts per-Gaussian occlusion from viewpoint to enable culling before rasterization.
    New component introduced by the paper; no independent evidence outside the claimed rendering results is provided.

pith-pipeline@v0.9.0 · 5479 in / 1372 out tokens · 27984 ms · 2026-05-17T06:08:11.061911+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

  • IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    We address this limitation by proposing a novel method to learn the viewpoint-dependent visibility function of all Gaussians in a trained model using a small, shared MLP across instances of an asset in a scene. By querying it for Gaussians within the viewing frustum prior to rasterization, our method can discard occluded primitives during rendering.

  • IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    Our results show that standard volume rendering implicitly encodes a soft form of occlusion: once transmittance saturates, subsequent splats can be discarded without any significant loss in color.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [1]

    Fairchild

    Pontus Andersson, Jim Nilsson, Tomas Akenine-M¨oller, Mag- nus Oskarsson, Kalle ˚Astr¨om, and Mark D. Fairchild. Flip: A difference evaluator for alternating images.Proceedings of the ACM on Computer Graphics and Interactive Techniques, 3(2):15:1–15:23, 2020. 6

    work page 2020

  2. [2]

    Lightgaussian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps, 2023

    Zhiwen Fan, Kevin Wang, Kairun Wen, Zehao Zhu, Dejia Xu, and Zhangyang Wang. Lightgaussian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps, 2023. 2

    work page 2023

  3. [3]

    Ea- gles: Efficient accelerated 3d gaussians with lightweight en- codings

    Sharath Girish, Kamal Gupta, and Abhinav Shrivastava. Ea- gles: Efficient accelerated 3d gaussians with lightweight en- codings. InComputer Vision – ECCV 2024, pages 54–71, Cham, 2025. Springer Nature Switzerland. 3, 4

    work page 2024

  4. [4]

    Measurement of areas on a sphere using fibonacci and latitude–longitude lattices.Mathematical geo- sciences, 42:49–64, 2010

    ´Alvaro Gonz´alez. Measurement of areas on a sphere using fibonacci and latitude–longitude lattices.Mathematical geo- sciences, 42:49–64, 2010. 4, 6

    work page 2010

  5. [5]

    3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics, 42(4), 2023

    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, 2, 4

    work page 2023

  6. [6]

    A hierarchical 3d gaussian representation for real-time ren- dering of very large datasets.ACM Transactions on Graphics, 43(4), 2024

    Bernhard Kerbl, Andreas Meuleman, Georgios Kopanas, Michael Wimmer, Alexandre Lanvin, and George Drettakis. A hierarchical 3d gaussian representation for real-time ren- dering of very large datasets.ACM Transactions on Graphics, 43(4), 2024. 1, 3

    work page 2024

  7. [7]

    3d gaussian splatting as markov chain monte carlo

    Shakiba Kheradmand, Daniel Rebain, Gopal Sharma, Weiwei Sun, Yang-Che Tseng, Hossam Isack, Abhishek Kar, Andrea Tagliasacchi, and Kwang Moo Yi. 3d gaussian splatting as markov chain monte carlo. InAdvances in Neural Information Processing Systems (NeurIPS), 2024. Spotlight Presentation. 2

    work page 2024

  8. [8]

    Adam: A method for stochastic optimization.International Conference on Learn- ing Representations, 2014

    Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization.International Conference on Learn- ing Representations, 2014. 6

    work page 2014

  9. [9]

    LODGE: Level-of- detail large-scale Gaussian splatting with efficient rendering

    Jonas Kulhanek, Marie-Julie Rakotosaona, Fabian Manhardt, Christina Tsalicoglou, Michael Niemeyer, Torsten Sattler, Songyou Peng, and Federico Tombari. LODGE: Level-of- detail large-scale Gaussian splatting with efficient rendering. InProceedings of the 39th International Conference on Neu- ral Information Processing Systems, 2025. 1, 3

    work page 2025

  10. [10]

    Characterization and analysis of the 3d gaussian splatting rendering pipeline.IEEE Computer Architecture Letters, 24 (1):13–16, 2025

    Jiwon Lee, Yunjae Lee, Youngeun Kwon, and Minsoo Rhu. Characterization and analysis of the 3d gaussian splatting rendering pipeline.IEEE Computer Architecture Letters, 24 (1):13–16, 2025. 5

    work page 2025

  11. [11]

    Occlugaussian: Occlusion-aware gaussian splatting for large scene reconstruction and rendering

    Shiyong Liu, Xiao Tang, Zhihao Li, Yingfan He, Chongjie Ye, Jianzhuang Liu, Binxiao Huang, Shunbo Zhou, and Xiaofei Wu. Occlugaussian: Occlusion-aware gaussian splatting for large scene reconstruction and rendering. 2025. 3, 4

    work page 2025

  12. [12]

    Scaffold-gs: Structured 3d gaussians for view-adaptive rendering

    Tao Lu, Mulin Yu, Linning Xu, Yuanbo Xiangli, Limin Wang, Dahua Lin, and Bo Dai. Scaffold-gs: Structured 3d gaussians for view-adaptive rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20654–20664, 2024. 3

    work page 2024

  13. [13]

    Compact 3d scene representation via self- organizing gaussian grids

    Wieland Morgenstern, Florian Barthel, Anna Hilsmann, and Peter Eisert. Compact 3d scene representation via self- organizing gaussian grids. InComputer Vision – ECCV 2024, pages 18–34, Cham, 2025. Springer Nature Switzerland. 2

    work page 2024

  14. [14]

    tiny-cuda-nn, 2021

    Thomas M ¨uller. tiny-cuda-nn, 2021. 4

    work page 2021

  15. [15]

    Compgs: Smaller and faster gaussian splatting with vector quantization.ECCV,

    KL Navaneet, Kossar Pourahmadi Meibodi, Soroush Abbasi Koohpayegani, and Hamed Pirsiavash. Compgs: Smaller and faster gaussian splatting with vector quantization.ECCV,

    work page

  16. [16]

    Radsplat: Radiance field-informed gaussian splat- ting for robust real-time rendering with 900+ fps

    Michael Niemeyer, Fabian Manhardt, Marie-Julie Rakoto- saona, Michael Oechsle, Daniel Duckworth, Rama Gosula, Keisuke Tateno, John Bates, Dominik Kaeser, and Federico Tombari. Radsplat: Radiance field-informed gaussian splat- ting for robust real-time rendering with 900+ fps. InInterna- tional Conference on 3D Vision 2025, 2025. 3, 4

    work page 2025

  17. [17]

    Reducing the mem- ory footprint of 3d gaussian splatting.Proceedings of the ACM on Computer Graphics and Interactive Techniques, 7 (1), 2024

    Panagiotis Papantonakis, Georgios Kopanas, Bernhard Kerbl, Alexandre Lanvin, and George Drettakis. Reducing the mem- ory footprint of 3d gaussian splatting.Proceedings of the ACM on Computer Graphics and Interactive Techniques, 7 (1), 2024. 2

    work page 2024

  18. [18]

    StopThePop: Sorted Gaussian Splatting for View-Consistent Real-time Rendering.ACM Transactions on Graphics, 4(43),

    Lukas Radl, Michael Steiner, Mathias Parger, Alexan- der Weinrauch, Bernhard Kerbl, and Markus Steinberger. StopThePop: Sorted Gaussian Splatting for View-Consistent Real-time Rendering.ACM Transactions on Graphics, 4(43),

    work page

  19. [19]

    Octree-gs: Towards consistent real-time rendering with lod-structured 3d gaussians.IEEE Transactions on Pattern Analysis and Machine Intelligence, pages 1–15, 2025

    Kerui Ren, Lihan Jiang, Tao Lu, Mulin Yu, Linning Xu, Zhangkai Ni, and Bo Dai. Octree-gs: Towards consistent real-time rendering with lod-structured 3d gaussians.IEEE Transactions on Pattern Analysis and Machine Intelligence, pages 1–15, 2025. 1, 3

    work page 2025

  20. [20]

    Revising densification in gaussian splatting

    Samuel Rota Bul `o, Lorenzo Porzi, and Peter Kontschieder. Revising densification in gaussian splatting. InComputer Vision – ECCV 2024: 18th European Conference, Milan, Italy, September 29–October 4, 2024, Proceedings, Part LXIII, page 347–362, Berlin, Heidelberg, 2024. Springer-Verlag. 2

    work page 2024

  21. [21]

    Flod: Integrating flexible level of detail into 3d gaussian splatting for customizable rendering, 2024

    Yunji Seo, Young Sun Choi, Hyun Seung Son, and Youngjung Uh. Flod: Integrating flexible level of detail into 3d gaussian splatting for customizable rendering, 2024. 1, 3

    work page 2024

  22. [22]

    Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ra- mamoorthi, Jonathan T

    Matthew Tancik, Pratul P. Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ra- mamoorthi, Jonathan T. Barron, and Ren Ng. Fourier features let networks learn high frequency functions in low dimen- sional domains.NeurIPS, 2020. 4

    work page 2020

  23. [23]

    RTMV: A ray-traced multi-view synthetic dataset for novel view synthesis

    Jonathan Tremblay, Moustafa Meshry, Alex Evans, Jan Kautz, Alexander Keller, Sameh Khamis, Thomas M¨uller, Charles Loop, Nathan Morrical, Koki Nagano, Towaki Takikawa, and Stan Birchfield. RTMV: A ray-traced multi-view synthetic dataset for novel view synthesis. InECCV Workshop on Learning to Generate 3D Shapes and Scenes, 2022. 6, 1, 3 9

    work page 2022

  24. [24]

    Donut 3d model

    Karan Vagadia. Donut 3d model. https : / / www . cgtrader . com / free - 3d - models / food / beverage/realistic- donut- fa8648b6- f2f6- 4fb3-93d9-ce531d2ba013, 2021. 3D model, accessed on May 20, 2024. 7, 2, 3

    work page 2021

  25. [25]

    Image quality assessment: From error visibility to structural similarity.Image Processing, IEEE Transactions on, 13:600 – 612, 2004

    Zhou Wang, Alan Bovik, Hamid Sheikh, and Eero Simoncelli. Image quality assessment: From error visibility to structural similarity.Image Processing, IEEE Transactions on, 13:600 – 612, 2004. 6

    work page 2004

  26. [26]

    Mvhumannet: A large- scale dataset of multi-view daily dressing human captures

    Zhangyang Xiong, Chenghong Li, Kenkun Liu, Hongjie Liao, Jianqiao Hu, Junyi Zhu, Shuliang Ning, Lingteng Qiu, Chongjie Wang, Shijie Wang, et al. Mvhumannet: A large- scale dataset of multi-view daily dressing human captures. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19801–19811, 2024. 7, 1, 3

    work page 2024

  27. [27]

    Virtualized 3d gaussians: Flexible cluster-based level- of-detail system for real-time rendering of composed scenes

    Xijie Yang, Linning Xu, Lihan Jiang, Dahua Lin, and Bo Dai. Virtualized 3d gaussians: Flexible cluster-based level- of-detail system for real-time rendering of composed scenes. InACM SIGGRAPH 2025 Conference Papers. Association for Computing Machinery, 2025. 1, 3, 5, 6, 7, 4

    work page 2025

  28. [28]

    gsplat: An open- source library for gaussian splatting.Journal of Machine Learning Research, 26(34):1–17, 2025

    Vickie Ye, Ruilong Li, Justin Kerr, Matias Turkulainen, Brent Yi, Zhuoyang Pan, Otto Seiskari, Jianbo Ye, Jeffrey Hu, Matthew Tancik, and Angjoo Kanazawa. gsplat: An open- source library for gaussian splatting.Journal of Machine Learning Research, 26(34):1–17, 2025. 1, 6

    work page 2025

  29. [29]

    Mip-splatting: Alias-free 3d gaussian splat- ting

    Zehao Yu, Anpei Chen, Binbin Huang, Torsten Sattler, and Andreas Geiger. Mip-splatting: Alias-free 3d gaussian splat- ting. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 19447– 19456, 2024. 3

    work page 2024

  30. [30]

    LP-3DGS: Learning to prune 3d gaussian splatting

    Zhaoliang Zhang, Tianchen Song, Yongjae Lee, Li Yang, Cheng Peng, Rama Chellappa, and Deliang Fan. LP-3DGS: Learning to prune 3d gaussian splatting. InThe Thirty-eighth Annual Conference on Neural Information Processing Sys- tems, 2024. 2

    work page 2024

  31. [31]

    PRoGS: Progressive Rendering of Gaussian Splats

    Brent Zoomers, Maarten Wijnants, Ivan Molenaers, Joni Vanherck, Jeroen Put, Lode Jorissen, and Nick Michiels. PRoGS: Progressive Rendering of Gaussian Splats . In2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 3118–3127, Los Alamitos, CA, USA,

    work page

  32. [32]

    3, 4 10 NVGS: Neural Visibility for Occlusion Culling in 3D Gaussian Splatting Supplementary Material

    IEEE Computer Society. 3, 4 10 NVGS: Neural Visibility for Occlusion Culling in 3D Gaussian Splatting Supplementary Material

    work page

  33. [33]

    This includes the results on the full layout files compared with our baseline instanced rasterizer implementation

    Supplementary In our supplementary material, we present additional results for all the layout files and show renders and FLIP compar- isons for all layouts. This includes the results on the full layout files compared with our baseline instanced rasterizer implementation. Results for individual assets are presented, along with visual differences between th...

    work page