Pith. sign in

REVIEW 3 major objections 7 minor 50 references

EyeNavGS: A 6-DoF Navigation Dataset and Record-n-Replay Software for Real-World 3DGS Scenes in VR

T0 review · 3 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read EyeNavGS presents the first publicly available 6-DoF navigation dataset collected inside photorealistic 3DGS reconstructions of real-world scenes, with per-frame head and gaze traces from 46 participants.

desk verdict A genuinely new 6-DoF navigation dataset for real-world 3DGS scenes, but the gaze and FOV columns are shipped without validation or clear documentation, so it is not yet benchmark-grade. read the letter →

arxiv 2506.02380 v1 pith:33FJWHCP submitted 2025-06-03 cs.MM cs.CVcs.GRcs.HC

classification cs.MMcs.CVcs.GRcs.HC
keywords 3DGaussianSplatting6-DoFnavigationvirtualrealityeyegazedatasetviewportpredictionadaptivestreamingfoveatedrenderingrecord-and-replay
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

EyeNavGS introduces the first publicly available 6-DoF navigation dataset for photorealistic 3DGS reconstructions of real-world scenes, collected from 46 participants exploring twelve indoor and outdoor scenes at two sites. Each trace records per-frame head pose, eye gaze, field of view, and timestamps for both eyes during free, standing navigation in a 3 m × 3 m area. The paper also releases a record-and-replay fork of the SIBR viewer and utility tools so other systems can replay and convert the traces. If the dataset holds up, researchers in viewport prediction, view-adaptive streaming, 3D saliency, and foveated rendering gain a reusable source of realistic human behavior in 3DGS scenes.

What carries the argument

The load-bearing mechanism is a record-and-replay extension of the SIBR viewer's OpenXR branch, used in free world standing mode. During recording, per-frame stereo pose, FOV, and gaze data are written to CSV; during replay, loadViewData() reads the CSV and overrides the headset pose before rendering. Scene initialization applies a tilt-correcting quaternion, a real-world scale factor, and a chosen starting viewpoint without modifying the trained 3DGS .ply files. Utility tools convert traces between virtual-world and physical-stage coordinates and between CSV and JSON pose formats.

What would settle it

Run a calibration audit on the published traces: for each participant, project the recorded gaze ray into the corresponding rendered stereo frame and check that fixation points rest on scene surfaces without a constant per-participant offset. The paper provides no such accuracy numbers, so evidence of systematic gaze bias or tracking dropout would falsify the usefulness of the gaze half of the dataset.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that authentic human 6-DoF navigation in 3DGS scenes can be captured as per-frame traces and faithfully replayed in a modified SIBR viewer. The dataset contains left- and right-eye views for every rendered frame, including eye position, head orientation, gaze position and gaze orientation in world coordinates, FOV angles, and timestamps. Each scene was initialized with a corrective quaternion, a metric scale factor, and a manually selected starting viewpoint so the virtual world matches gravity and physical proportions. The authors position this as filling the gap between synthetic navigation traces and traces collected from less realistic representations like point clouds.

Load-bearing premise

The load-bearing premise is that the eye gaze traces recorded from the Meta Quest Pro are accurate and usable without per-participant calibration or filtering.

Editorial extensions

If this is right

  • 6-DoF viewport prediction models can be trained and tested on real human exploration patterns instead of synthetic paths.
  • Adaptive streaming systems for 3DGS can be benchmarked against recorded traces with known frame rates and movement distances.
  • Aggregated gaze fixations provide a basis for 3D saliency maps of real-world reconstructed scenes.
  • Foveated rendering can be evaluated with per-frame gaze data recorded during natural standing 6-DoF navigation.
  • Replay mode lets researchers regenerate exact stereo video from a trace, enabling reproducible rendering studies.

Reading between the lines

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

  • A consequence the authors do not state explicitly is that the gaze columns need an accuracy audit before being used as ground truth, since the paper reports no per-participant calibration or tracking-loss handling.
  • Because the traces include scene-scale and tilt parameters, one could transfer traces across scenes by applying the inverse initialization transform, which the paper only hints at through its physical-stage coordinate conversion.
  • The dataset could be extended into a benchmark by adding a held-out participant split and standardized viewport-prediction error metrics, which the paper does not define.
  • The recorded frame-rate gap between sites means downstream results may need to account for whether rendering lag altered navigation behavior.
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 / 7 minor

Summary. The paper introduces EyeNavGS, a dataset of 6-DoF navigation traces from 46 participants exploring twelve real-world 3D Gaussian Splatting (3DGS) scenes in VR, collected with Meta Quest Pro headsets at two sites. Each trace contains per-frame head pose, eye gaze, field-of-view, and timestamp data. The authors also release an open-source fork of the SIBR viewer with record-and-replay functionality and utility tools for conversion and visualization. The paper describes the data collection procedure, the CSV data format, per-scene initialization parameters, and basic statistics of the collected traces.

Significance. If the dataset is of high quality, it addresses a genuine gap: no publicly available navigation dataset currently targets photorealistic 3DGS reconstructions of real-world scenes. The accompanying open-source software and utility tools increase the practical value, and the multi-site, 46-participant collection with IRB approval suggests a reasonable diversity of users. The paper also provides concrete data-format specifications and per-scene initialization parameters, which support reproducibility. However, the paper's central claim—that this is a trustworthy, reusable dataset for viewport prediction, saliency, and foveated rendering—currently hinges on sensor data that is not validated.

major comments (3)
  1. [Section 3.3, Table 3] The sample rows in Table 3 list FOV1 = -0.942 and FOV3 = -0.942; field-of-view angles are physically non-negative. This suggests either an undocumented sign convention (e.g., angles measured relative to the view axis) or unsanitized raw values in the released data. Because this table is the only concrete illustration of the dataset format, the paper must clarify the sign convention, verify that all FOV entries in the released traces satisfy the expected range, and document a validation step that checks FOV signs and magnitudes during recording or post-processing.
  2. [Section 4.1 and Section 3.3] The paper never reports that per-participant eye calibration was performed on the Meta Quest Pro, nor does it provide any gaze accuracy or precision metrics. The gaze columns are half of the dataset's novelty, and the claimed downstream uses (3D saliency, foveated rendering, gaze-contingent streaming) depend on accurate fixation data. The authors should document the eye-calibration procedure (including whether the headset's built-in calibration was used), report gaze validation results (e.g., against a known fixation target or a held calibration slide), and add per-session or per-row quality flags so that consumers can exclude sessions with invalid or lost eye tracking.
  3. [Section 6] The dataset analysis validates neither the head-pose nor the eye-gaze streams; it reports only average frame rates and total distances walked. For the paper's use-case claims to be credible, the paper should include basic physical plausibility checks: FOV consistency across left and right eyes, smoothness of head orientation and position over time, consistency between head and gaze orientation during fixation, and a cross-check of the virtual-to-physical coordinate conversion described in Section 5. Such validation would also help identify sessions that should be excluded from the release, which is particularly important given the absence of any per-row quality tagging.
minor comments (7)
  1. [Section 2.1] The phrase "their their relatively low rendered visual quality" contains a duplicated "their"; please correct the typo.
  2. [Section 3.1] The text refers to the "Blende-KIRI add-on"; this should be "Blender-KIRI add-on".
  3. [Section 4.1] The sentence "Each participants explored twelve scenes" uses incorrect subject-verb agreement; it should be "Each participant explored twelve scenes" or "All participants each explored twelve scenes."
  4. [Section 3.3] The description of the GazePos columns is ambiguous: the text says these values are "very similar to, but distinct from" the eye position, which leaves unclear whether GazePos is the origin of the gaze ray (i.e., the eyeball position) or the 3D fixation point on the scene. Please clarify the physical meaning and, if it is the ray origin, consider renaming the column to "GazeOrigin" to avoid confusion.
  5. [Table 3] The sample table should include explicit units and sign annotations for the FOV columns; currently the negative values are unexplained and could easily be misread as data errors.
  6. [Abstract and Section 1] The claim of being "the first publicly available dataset" is hard to verify without a more comprehensive search of recent 3DGS navigation datasets; the authors should consider softening this to "to the best of our knowledge" in the abstract and supporting it with a more thorough related-work discussion.
  7. [Section 4.2 and Dataset Availability] The paper gives a URL for the dataset but no persistent identifier or checksum; providing a DOI or a versioned release with file hashes would strengthen the reproducibility of the resource.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: EyeNavGS is a measured dataset contribution, not a derivation; self-citations are motivational and not load-bearing.

full rationale

This paper presents a dataset and software release, so there is no derivation chain whose outputs could reduce to inputs by construction. The central claim—that EyeNavGS is the first publicly available 6-DoF navigation dataset for real-world 3DGS scenes—is an empirical claim-of-record assertion checked against prior datasets discussed in Section 2.1; it is not obtained by fitting a parameter or by a definitional identity. The per-scene initialization parameters (tilt quaternion, scale, initial position; Table 1) are manually calibrated values applied to scenes, not predicted outputs. The record-replay pipeline injects recorded traces into the SIBR renderer; replay fidelity is an engineering property, not a circular derivation. Self-citations to SGSS [47] and LTS [42] appear only as motivation for why the dataset is needed (Section 1, Section 7) and as example frameworks for compatibility; none of these citations is used to justify the dataset's existence, format, or validity. No uniqueness theorem or prior author result is invoked to forbid alternatives. The absence of reported eye-calibration and the negative FOV sample values in Table 3 are data-quality and validity concerns, not circularity; they do not make any claim equivalent to its inputs. Therefore the paper shows no significant circularity.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

This is a measurement paper, so the ledger holds calibration constants and domain assumptions rather than fitted model parameters. The per-scene quaternions, scale factors, and initial positions in Table 1 are hand-tuned via a manual Blender workflow and are load-bearing for the physical-stage-coordinate conversion and for the perceptual-comfort claim, but their accuracy is unquantified. The main domain assumption is that unvalidated Quest Pro gaze and pose data are accurate enough for research use; the paper provides no calibration, validation, or loss-filtering evidence.

free parameters (3)
  • per-scene scale factor (12 scenes) = 0.76, 12, 0.36, 1, 2, 2.7, 1, 1.25, 0.64, 0.53, 0.8, 0.64 (per scene order of Table 1)
    Hand-calibrated in Blender/KIRI Engine by comparing known real-world dimensions to the 3DGS representation, then applied at runtime to map virtual units to meters. Accuracy of the calibration is not quantified in the paper.
  • per-scene tilt-correction quaternion (12 scenes) = see Table 1; e.g., truck (-0.0896, 0, 0, 0.9960), london (0, 0, 0, 1)
    Manually adjusted in Blender to align the scene ground plane with gravity, then exported and applied via XrReferenceSpaceCreateInfo. This transform is undone in the utility that converts traces to physical stage coordinates (Section 5).
  • per-scene initial view position (12 scenes) = see Table 1; e.g., truck (0, 2.1, -4), nyc (-1.6, 4.4, 4)
    Manually selected to avoid starting inside geometry and to emulate natural human perspective. It affects first-person viewport statistics but not the trace geometry itself.
assumptions (5)
  • domain assumption COLMAP-reconstructed 3DGS scenes lack gravity alignment and intrinsic metric scale.
    Section 3.1 motivates the tilt and scale corrections with this; it is standard behavior for structure-from-motion pipelines and is treated as an input to the calibration workflow rather than proven in the paper.
  • domain assumption OpenXR Stage Space maps physical movement 1:1 to virtual movement at the calibrated scale.
    Section 2.2 describes free world standing mode as a 1:1 mapping of tracked physical movements to the virtual scene; the utility that converts traces to physical stage coordinates inverts the scene transforms (Section 5).
  • domain assumption The Meta Quest Pro head and eye tracking outputs are accurate enough for research use without per-participant validation.
    Section 4.1 enables eye gaze tracking but reports no eye-calibration procedure, no gaze-accuracy validation, and no tracking-loss filtering; the dataset's gaze columns are shipped as-is (Section 3.3).
  • ad hoc to paper One minute of task-free exploration yields representative navigation behavior.
    Section 4.1 fixes each session to one minute with no assigned task; no justification is given for the duration, and the analysis in Section 6 reports only fps and distance, so session adequacy is assumed rather than demonstrated.
  • standard math Quaternion pose transforms applied at runtime via XrReferenceSpaceCreateInfo correctly resolve the OpenXR and COLMAP coordinate mismatch.
    Section 3.1 and Figure 1 describe rotating the camera 180 degrees around the X axis; this is standard rigid-body math and is not the load-bearing risk.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EyeNavGS: A 6-DoF Navigation Dataset and Record-n-Replay Software for Real-World 3DGS Scenes in VR." pith.science (2026). https://pith.science/paper/33FJWHCP

@misc{pith2026250602380,
  author       = {Pith},
  title        = {Pith review of: EyeNavGS: A 6-DoF Navigation Dataset and Record-n-Replay Software for Real-World 3DGS Scenes in VR},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/33FJWHCP}},
  note         = {Machine review of arXiv:2506.02380}
}
read the original abstract

3D Gaussian Splatting (3DGS) is an emerging media representation that reconstructs real-world 3D scenes in high fidelity, enabling 6-degrees-of-freedom (6-DoF) navigation in virtual reality (VR). However, developing and evaluating 3DGS-enabled applications and optimizing their rendering performance, require realistic user navigation data. Such data is currently unavailable for photorealistic 3DGS reconstructions of real-world scenes. This paper introduces EyeNavGS (EyeNavGS), the first publicly available 6-DoF navigation dataset featuring traces from 46 participants exploring twelve diverse, real-world 3DGS scenes. The dataset was collected at two sites, using the Meta Quest Pro headsets, recording the head pose and eye gaze data for each rendered frame during free world standing 6-DoF navigation. For each of the twelve scenes, we performed careful scene initialization to correct for scene tilt and scale, ensuring a perceptually-comfortable VR experience. We also release our open-source SIBR viewer software fork with record-and-replay functionalities and a suite of utility tools for data processing, conversion, and visualization. The EyeNavGS dataset and its accompanying software tools provide valuable resources for advancing research in 6-DoF viewport prediction, adaptive streaming, 3D saliency, and foveated rendering for 3DGS scenes. The EyeNavGS dataset is available at: https://symmru.github.io/EyeNavGS/.

Figures

Figures reproduced from arXiv: 2506.02380 by the authors.

Figure 1
Figure 1. The SIBR viewer source code resolves the mismatch [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Real-time user trace collection setup showing the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Pre-trained 3DGS scenes from the 3DGS paper [ [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Newly trained 3DGS scenes from the ZipNeRF [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Example eye gaze visualization of the bicycle scene. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: User movement trajectories for an outdoor and an [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 36 canonical work pages

  1. [1]

    The Khronos Group Inc. [n. d.].OpenXR Overview. The Khronos Group Inc. https://www.khronos.org/openxr/

  2. [2]

    2023.gsplat — 3D Gaussian Splatting WebGL viewer

  3. [3]

    antimatter15. 2023. Splat: WebGL Gaussian Splatting Renderer. https://github. com/antimatter15/splat

  4. [4]

    Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P Srinivasan, and Peter Hedman. 2022. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion. 5470–5479

  5. [5]

    Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P Srinivasan, and Pe- ter Hedman. 2023. Zip-nerf: Anti-aliased grid-based neural radiance fields. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 19697– 19705

  6. [6]

    Harun Barutcu. 2024. 404 Gen: Bringing Text to 3D Gaussian Splatting to Unity. Radiance Fields(17 April 2024). https://radiancefields.com/404-gen-bringing- text-to-3d-gaussian-splatting-to-unity

  7. [7]

    Blender Foundation. [n. d.].blender.org. https://www.blender.org/

  8. [8]

    Sebastien Bonopera, Jerome Esnault, Siddhant Prakash, Simon Rodriguez, Theo Thonat, Mehdi Benadel, Gaurav Chaurasia, Julien Philip, and George Drettakis

Show all 50 references
  1. [9]

    Jacob Chakareski, Mahmudur Khan, Tanguy Ropitault, and Steve Blandino. 2020. 6DOF virtual reality dataset and performance evaluation of millimeter wave vs. free-space-optical indoor communications systems for lifelike mobile VR streaming. In2020 54th Asilomar Conference on Sig...

  2. [10]

    Ying Chen, Hojung Kwon, Hazer Inaltekin, and Maria Gorlatova. 2022. VR Viewport Pose Model for Quantifying and Exploiting Frame Correlations. In Proc. IEEE INFOCOM

  3. [11]

    Daniel Duckworth, Peter Hedman, Christian Reiser, Peter Zhizhin, Jean-François Thibert, Mario Lučić, Richard Szeliski, and Jonathan T Barron. 2024. Smerf: Streamable memory efficient radiance fields for real-time large-scene exploration. ACM Transactions on Graphics (TOG)43, 4...

  4. [12]

    Eugene d’Eon, Bob Harrison, Taos Myers, and Philip A Chou. 2017. 8i voxelized full bodies-a voxelized point cloud dataset.ISO/IEC JTC1/SC29 Joint WG11/WG1 (MPEG/JPEG) input document WG11M40059/WG1M740067, 8 (2017), 11

  5. [13]

    Elias Ennadifi, Thierry Ravet, Matei Mancas, Mohammed El Amine Mokhtari, and Bernard Gosselin. 2023. Enhancing VR Gaming Experience using Computational Attention Models and Eye-Tracking. InProceedings of the 2023 ACM international conference on interactive media experiences. 194–198

  6. [14]

    Guangchi Fang and Bing Wang. 2024. Mini-splatting: Representing scenes with a constrained number of gaussians. InEuropean Conference on Computer Vision. Springer, 165–181

  7. [15]

    Linus Franke, Laura Fink, and Marc Stamminger. 2025. Vr-splatting: Foveated radiance field rendering via 3d gaussian splatting and neural points.Proceedings of the ACM on Computer Graphics and Interactive Techniques8, 1 (2025), 1–21

  8. [16]

    Serhan Gül, Sebastian Bosse, Dimitri Podborski, Thomas Schierl, and Cornelius Hellge. 2020. Kalman filter-based head motion prediction for cloud-based mixed reality. InProceedings of the 28th ACM international conference on multimedia. 3632–3641

  9. [17]

    Peter Hedman, Julien Philip, True Price, Jan-Michael Frahm, George Drettakis, and Gabriel Brostow. 2018. Deep blending for free-viewpoint image-based rendering.ACM Transactions on Graphics (ToG)37, 6 (2018), 1–15

  10. [18]

    Kaiyuan Hu, Yili Jin, Haowen Yang, Junhua Liu, and Fangxin Wang. 2023. FSVVD: A dataset of full scene volumetric video. InProceedings of the 14th Conference on ACM Multimedia Systems. 410–415

  11. [19]

    Kaiyuan Hu, Haowen Yang, Yili Jin, Junhua Liu, Yongting Chen, Miao Zhang, and Fangxin Wang. 2023. Understanding user behavior in volumetric video watching: Dataset, analysis and prediction. InProceedings of the 31st ACM International Conference on Multimedia. 1108–1116

  12. [20]

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao

  13. [21]

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis

  14. [22]

    Bernhard Kerbl, Andreas Meuleman, Georgios Kopanas, Michael Wimmer, Alexandre Lanvin, and George Drettakis. 2024. A hierarchical 3d gaussian representation for real-time rendering of very large datasets.ACM Transactions on Graphics (TOG)43, 4 (2024), 1–15

  15. [23]

    Khan and J

    M. Khan and J. Chakareski. 2020. NJIT 6DOF VR Navigation Dataset. https: //www.jakov.org. https://www.jakov.org Accessed: 2025-05-06

  16. [24]

    2025.3DGS Render Blender Addon - KIRI Engine

    KIRI Engine. 2025.3DGS Render Blender Addon - KIRI Engine. https://www. kiriengine.app/blender-addon/3dgs-render

  17. [25]

    Arno Knapitsch, Jaesik Park, Qian-Yi Zhou, and Vladlen Koltun. 2017. Tanks and temples: Benchmarking large-scale scene reconstruction.ACM Transactions on Graphics (ToG)36, 4 (2017), 1–13

  18. [26]

    Muhammed Kocabas, Jen-Hao Rick Chang, James Gabriel, Oncel Tuzel, and Anurag Ranjan. 2024. Hugs: Human gaussian splats. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 505–515

  19. [27]

    Brooke Krajancich, Petr Kellnhofer, and Gordon Wetzstein. 2023. Towards attention–aware foveated rendering.ACM Transactions on Graphics (TOG)42, 4 (2023), 1–10

  20. [28]

    Simon Lessels and Roy A Ruddle. 2004. Changes in navigational behaviour produced by a wide field of view and a high fidelity visual scene. InProceedings of the 10th Eurographics Symposium on Virtual Environments. Eurographics, 71– 78

  21. [29]

    Weikai Lin, Yu Feng, and Yuhao Zhu. 2025. MetaSapiens: Real-Time Neural Rendering with Efficiency-Aware Pruning and Accelerated Foveated Rendering. InProceedings of the 30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, ...

  22. [30]

    Jean-Eudes Marvie and Pascal Gautron. 2025. Real-Time GPU-Accelerated Gaussian Splatting with NVIDIA DesignWorks Sample vk_gaussian_splatting. NVIDIA Technical Blog(23 April 2025). https://developer.nvidia.com/blog/real- time-gpu-accelerated-gaussian-splatting-with-nvidia-desi...

  23. [31]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. 2020. NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis. InECCV

  24. [32]

    mkkellogg. 2024. GaussianSplats3D: Three.js-based Gaussian Splatting renderer. https://github.com/mkkellogg/GaussianSplats3D

  25. [33]

    Jérémy Ouellette, Jashanjot Singh Sidhu, and Abdelhak Bentaleb. 2025. Maze- Lab: A Large-Scale Dynamic Volumetric Point Cloud Video Dataset With User Behavior Traces. InProceedings of the 16th ACM Multimedia Systems Conference. 298–304

  26. [34]

    Panagiotis Papantonakis, Georgios Kopanas, Bernhard Kerbl, Alexandre Lanvin, and George Drettakis. 2024. Reducing the memory footprint of 3d gaussian splatting.Proceedings of the ACM on Computer Graphics and Interactive Techniques 7, 1 (2024), 1–17

  27. [35]

    Anjul Patney, Marco Salvi, Joohwan Kim, Anton Kaplanyan, Chris Wyman, Nir Benty, David Luebke, and Aaron Lefohn. 2016. Towards foveated rendering for gaze-tracked virtual reality.ACM Transactions On Graphics (TOG)35, 6 (2016), 1–12. Ding et al

  28. [36]

    Johannes Lutz Schönberger and Jan-Michael Frahm. 2016. Structure-from-Motion Revisited. InConference on Computer Vision and Pattern Recognition (CVPR)

  29. [37]

    Johannes Lutz Schönberger, Enliang Zheng, Marc Pollefeys, and Jan-Michael Frahm. 2016. Pixelwise View Selection for Unstructured Multi-View Stereo. In European Conference on Computer Vision (ECCV)

  30. [38]

    Yuang Shi, Géraldine Morin, Simone Gasparini, and Wei Tsang Ooi. 2024. Lapisgs: Layered progressive 3d gaussian splatting for adaptive streaming.arXiv preprint arXiv:2408.14823(2024)

  31. [39]

    Rahul Singh, Muhammad Huzaifa, Jeffrey Liu, Anjul Patney, Hashim Sharif, Yifan Zhao, and Sarita Adve. 2023. Power, performance, and image quality tradeoffs in foveated rendering. In2023 IEEE Conference Virtual Reality and 3D User Interfaces (VR). IEEE, 205–214

  32. [40]

    Vincent Sitzmann, Ana Serrano, Amy Pavel, Maneesh Agrawala, Diego Gutierrez, Belen Masia, and Gordon Wetzstein. 2018. Saliency in VR: How do people explore virtual environments?IEEE transactions on visualization and computer graphics 24, 4 (2018), 1633–1642

  33. [41]

    Shishir Subramanyam, Irene Viola, Alan Hanjalic, and Pablo Cesar. 2020. User centered adaptive streaming of dynamic point clouds with low complexity tiling. InProceedings of the 28th ACM international conference on multimedia. 3669–3677

  34. [42]

    Yuan-Chun Sun, Yuang Shi, Cheng-Tse Lee, Mufeng Zhu, Wei Tsang Ooi, Yao Liu, Chun-Ying Huang, and Cheng-Hsin Hsu. 2025. LTS: A DASH streaming system for dynamic multi-layer 3D Gaussian splatting scenes. InProceedings of the 16th ACM Multimedia Systems Conference. 136–147

  35. [43]

    Matthew Tancik, Ethan Weber, Evonne Ng, Ruilong Li, Brent Yi, Justin Kerr, Terrance Wang, Alexander Kristoffersen, Jake Austin, Kamyar Salahi, Abhik Ahuja, David McAllister, and Angjoo Kanazawa. 2023. Nerfstudio: A Modular Framework for Neural Radiance Field Development. InACM...

  36. [44]

    Yi-Zhen Tsai, Xuechen Zhang, Zheng Li, and Jiasi Chen. 2025. L3GS: Layered 3D Gaussian Splats for Efficient 3D Scene Delivery.arXiv preprint arXiv:2504.05517 (2025)

  37. [45]

    Yao Wang, Qi Dai, Mihai Bâce, Karsten Klein, and Andreas Bulling. 2024. Saliency3D: a 3D saliency dataset collected on screen. InProceedings of the 2024 Symposium on Eye Tracking Research and Applications. 1–6

  38. [46]

    Zehao Yu, Anpei Chen, Binbin Huang, Torsten Sattler, and Andreas Geiger. 2024. Mip-splatting: Alias-free 3d gaussian splatting. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 19447–19456

  39. [47]

    Mufeng Zhu, Mingju Liu, Cunxi Yu, Cheng-Hsin Hsu, and Yao Liu. 2025. SGSS: Streaming 6-DoF Navigation of Gaussian Splat Scenes. InProceedings of the 16th ACM Multimedia Systems Conference. 46–56

  40. [2020]

    https://gitlab.inria.fr/sibr/sibr_ core

    sibr: A System for Image Based Rendering. https://gitlab.inria.fr/sibr/sibr_ core

  41. [2023]

    3d gaussian splatting for real-time radiance field rendering.ACM Transac- tions on Graphics42, 4 (2023), 1–14

  42. [2024]

    InACM SIGGRAPH 2024 conference papers

    2d gaussian splatting for geometrically accurate radiance fields. InACM SIGGRAPH 2024 conference papers. 1–11

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.