Pith. sign in

REVIEW 3 major objections 4 minor 78 references

MR-Compare: A Mixed-Reality Framework for Spatially Grounded Visual Comparison of 3D Gaussian Splatting and Mesh Reconstructions with the Physical Environment

T0 review · 3 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read MR-Compare registers 3D reconstructions with live physical scenes at centimetre-level accuracy, enabling direct mixed-reality visual comparison.

desk verdict Solid systems paper with a genuine contribution; the ArUco ground-truth protocol and the tuned anisotropy threshold are the soft spots, but it deserves a serious referee. read the letter →

arxiv 2607.20325 v1 pith:LZZT6XYE submitted 2026-07-22 cs.GR

Changrui Zhu , Ernst Kruijff , Pengju Zhang , Simon Julier This is my paper
classification cs.GR
keywords mixedreality3DGaussiansplattingspatialregistrationvisualcomparisonvideosee-throughanisotropyfilterpointcloudreconstructionevaluation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper introduces MR-Compare, a mixed-reality framework that aligns past 3D reconstructions—both meshes and 3D Gaussian splatting—with the live video see-through view of the physical environment. The core claim is that this spatially grounded comparison is feasible: across five representative reconstruction workflows in two real rooms, the system achieves centimetre-level registration error and lets users visually slide between the reconstruction and reality. The authors show that desktop 3D Gaussian splatting workflows, especially 3DGS-MCMC, produce the best alignment and visual consistency, and that a simple anisotropy filter can further improve registration by pruning non-surface Gaussians. A reader should care because this enables practical MR applications like construction inspection, cultural heritage, and change detection, where comparing a digital model against the real world in place is essential.

What carries the argument

The central mechanism is representation-agnostic point-cloud registration: any 3D reconstruction is converted to a source point cloud (mesh vertices or Gaussian centres), and the headset's depth-scan point cloud serves as the target. A coarse-to-fine pipeline (TEASER++ for robust global initial alignment, V-GICP for local refinement) computes the transform that binds the reconstruction to the physical environment. A 3D Slider—a hemisphere mesh that selectively reveals passthrough via an alpha mask—lets users visually slide between the reconstruction and live reality. The anisotropy filter uses the ratio of a Gaussian's smallest to largest principal scale (ρ = min(σ)/max(σ)) to prune near-sph

What would settle it

A dense ground-truth evaluation in a real room—for example, using a high-precision laser-scan or a dense fiducial target—would reveal whether the ArUco-referenced errors generalize. If dense alignment errors exceed the reported 0.89–4.35 cm in regions away from the markers, the feasibility claim would be weakened.

Watch

Extended reading notes

Core claim

The paper establishes system-level feasibility for spatially grounded visual comparison in mixed reality. By treating a reconstruction's point-based geometry (mesh vertices or Gaussian centres) as a source point cloud and the headset's depth scan as the target, MR-Compare registers both representations to the physical scene using a coarse-to-fine pipeline (TEASER++ followed by V-GICP). In a real-world benchmark, translation errors ranged from 0.89 to 4.35 cm across workflows and rooms, with the desktop 3DGS-MCMC workflow achieving the lowest errors and the strongest VST-referenced visual consistency. The paper also shows that a training-free anisotropy filter—pruning near-spherical Gaussians

Load-bearing premise

The real-world registration error claims rest on sparse ArUco markers (six per room) as a proxy for ground truth; if the markers or their placement are systematically biased, the reported centimetre-level errors may not reflect true alignment everywhere in the scene.

Editorial extensions

If this is right

  • If MR-Compare's feasibility claim holds, spatially grounded visual comparison becomes practical for inspection and change-detection tasks in static indoor environments, where users can directly compare a digital model against the physical scene in situ.
  • The centimetre-level registration accuracy indicates that VST headsets can serve as reliable reference frames for evaluating reconstruction fidelity, not just for display.
  • The anisotropy filter's success suggests that off-the-shelf 3DGS assets can be made more registration-friendly without retraining, lowering the barrier to using Gaussian splatting in MR workflows.
  • The desktop 3DGS-MCMC advantage over meshes and mobile reconstructions, if it generalizes, would guide practitioners toward specific reconstruction pipelines for MR applications where alignment and visual consistency matter.
  • The system's persistence of the registered transform via spatial anchors implies that repeated sessions can reuse a saved alignment, enabling longitudinal comparisons without re-registration.

Reading between the lines

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

  • The registration-error numbers are measured relative to six sparse ArUco markers per room; true dense alignment may vary in areas far from markers, so the centimetre-level claim is a sparse-reference estimate rather than global ground truth—though the controlled Replica evaluation with full ground truth supports the general trend.
  • The anisotropy filter's pruning criterion (flatness of Gaussians) could be extended to other point-based representations or to registration with time-varying scenes, though the paper only tests static indoor environments.
  • A testable extension is to deploy MR-Compare in a dynamic or outdoor environment to see whether the coarse-to-fine registration and VST-referenced consistency hold under changing illumination and moving objects; the authors flag this as future work.
  • The finding that SSIM/PSNR align well with subjective ratings for 3DGS workflows, but not for meshes, suggests that image-quality metrics may need to be interpreted per-representation when assessing cross-media agreement.
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 / 4 minor

Summary. The paper presents MR-Compare, a PC-tethered Unity MR framework for registering 3D reconstructions (mesh and 3D Gaussian splatting) to the live physical environment on the Meta Quest 3 via video see-through. It combines a coarse-to-fine registration pipeline (TEASER++/TurboReg + V-GICP) with a 3D Slider interaction for spatially grounded visual comparison. The authors evaluate five reconstruction workflows (RealityScan, Polycam, 3DGS, 3DGS-MCMC, Scaniverse) in two indoor rooms using an objective registration benchmark (ArUco-referenced) and an exploratory user study (n=30), reporting centimetre-level translation errors (0.89–4.35 cm) and strongest overall registration and VST-referenced visual consistency for desktop 3DGS workflows. A controlled Replica scene evaluation ablates the registration pipeline and introduces a zero-shot anisotropy filter that prunes Gaussian centres based on scale ratios, reporting improved robustness and reduced alignment error under moderate pruning. The paper frames the results as establishing system-level feasibility rather than task-level effectiveness or standalone deployment.

Significance. If the central claims hold, MR-Compare is a useful practical contribution: it addresses the underexplored problem of spatially grounded visual comparison of heterogeneous reconstructions against the live physical world, and it does so with a reproducible, self-contained pipeline. The real-world benchmark with five representative workflows and a 30-participant user study is a solid empirical contribution; the objective and perceived measures are appropriately framed as workflow-level feasibility evidence. The anisotropy filter is a lightweight, model-agnostic heuristic that could benefit 3DGS-to-scan registration, and the controlled Replica ablation provides a reproducible testbed. However, the load-bearing quantitative claim of centimetre-level accuracy rests on a sparsely documented ArUco protocol, and the anisotropy-filter evaluation uses per-scene best thresholds rather than the proposed zero-shot default. These issues do not invalidate the system-level feasibility message, but they require substantive revision before the quantitative contributions can be accepted at face value.

major comments (3)
  1. [§4.1.1, Table 1] The registration-error measurement with six ArUco markers per room is under-specified. The paper does not state how marker poses are extracted from each reconstruction (from source images, rendered views, or the SfM point cloud), how physical marker poses are estimated from the VST stream, whether the Quest 3 passthrough intrinsics/distortion were calibrated, or how VST latency and rolling-shutter effects were handled. The cited 1–2 mm ArUco accuracy is from controlled camera setups, not from a wide-FOV headset passthrough. Since the best-case error (0.89 cm) is within an order of magnitude of the measurement tool's expected noise, the reported centimetre-level accuracy could be dominated by measurement error. Please specify the complete marker-pose estimation protocol and provide a dense ground-truth check (e.g., high-precision scan comparison or synthetic ground truth) to support the c
  2. [§4.3.2, Fig. 6] The anisotropy-filter contribution is evaluated by sweeping the threshold τ on the same Replica scenes and reporting per-scene best results (e.g., the −0.35 cm/−37.6% improvement for 3DGS is computed at per-scene minima, not at a fixed setting). The paper then proposes the automatically extracted median anisotropy as a zero-shot default, but it does not report the performance of that median default on the Replica scenes, nor a held-out-scene protocol. As a result, the improvement figures are the outcome of an oracle sweep, not an evaluation of the proposed zero-shot filter. Please report the median-threshold (or otherwise fixed) performance on held-out scenes, and adjust the claims of 'zero-shot' and 'training-free' accordingly.
  3. [§5.2, §4.2.3] The text claims that 'both desktop 3DGS workflows showed strong subjective–objective consistency in both rooms' for registration, but the only statistical evidence in §4.2.3/Table B.7 is a workflow-level Spearman correlation across five methods (n=5), which does not quantify per-workflow consistency. The perfect/moderate correlations are aggregate rankings, not evidence about individual workflows. The ΔRank analysis in §B.5.2 refers to visual consistency, not registration. Please either provide a per-workflow analysis or restate the claim to match the actual analysis, e.g., 'the two desktop 3DGS workflows were the main drivers of the aggregate ranking agreement.'
minor comments (4)
  1. [§B.4 / Table B.8] Several post-hoc odds ratios are extreme (e.g., 863.77, 1203.90, 4361.13) with very wide confidence intervals. These are likely due to near-separation in the ordinal models. Please report a sensitivity analysis (e.g., penalized likelihood or exact methods) or downweight the interpretation of those ratios.
  2. [Appendix C.3] The 'sanity check' with the original 3DGS contradicts the earlier claim that standard 3DGS contains more isolated outlier noise removed by preprocessing, by attributing the far-field splats to the Nerfstudio pipeline rather than to 3DGS itself. This is a useful caveat, but the main text's interpretation in §5.3 should be reconciled with this appendix observation.
  3. [§4.1.1] For the ArUco measurements, please report the number of pose samples per marker, the temporal window (five seconds) relative to the headset frame rate, and whether the reported mean±SD is over time per marker or across the six repetitions. Currently 'five-second interval' and 'six times' are ambiguous.
  4. [§5.4 / Table C.2] The simulated depth scanner introduces many hand-set noise parameters. Please provide a rationale or sensitivity analysis for at least the dominant parameters (e.g., noise coefficients, quantisation steps, burst duration) to show the controlled evaluation is not strongly tuned to the simulator.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; central claims rest on external measurements rather than on the paper's own assumptions.

full rationale

MR-Compare's central quantitative claims are measured against independent references rather than derived from its own inputs. Registration error is evaluated with physical ArUco markers as a sparse external ground truth; VST-referenced visual consistency compares reconstruction renderings against live passthrough images; and the user study collects participant ratings. The registration pipeline uses external algorithms (TEASER++, V-GICP) with independent source and target point clouds, and the marker-based metric is not used in the registration itself. The anisotropy-filter contribution is presented as a transparent threshold sweep on Replica scenes, with the median anisotropy chosen as a heuristic default; this is an in-sample hyperparameter/selection analysis and a potential generalizability limitation, but it does not reduce by construction to the paper's own equations or rename a fitted parameter as a prediction. The only self-citations ([74], [75]) support the adoption of a 3D Slider interaction technique and are not load-bearing for the paper's conclusions. The paper explicitly acknowledges its own limitations, including the sparse ArUco reference and the exploratory nature of the user study, which further supports a non-circular interpretation.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

No new physical entities are introduced. The central dependencies are domain assumptions about the accuracy of depth scans, sparse ArUco ground truth, Gaussian centres as surface proxies, and the representativeness of the simulated depth scanner. Multiple hand-chosen preprocessing and registration parameters shape the reported numbers; the anisotropy threshold is the clearest fitted parameter because it is swept and then summarized as a default.

free parameters (5)
  • Anisotropy filter threshold τ = swept over [1, 1e-7]; median anisotropy proposed as default
    Controls which Gaussian centres are retained. The best-performing range is identified from the same Replica scenes used for evaluation, so the default is data-dependent.
  • Radius crop radius r = 10.00 m
    Hand-chosen preprocessing limit on source points; affects outlier content entering registration.
  • Voxel-hash density filter s, N_min = s=0.2, N_min=3
    Hand-set density filter parameters used to remove floating noise from source and target clouds.
  • TEASER++/V-GICP registration parameters = voxel 0.1, noise bound 0.1, FPFH radii 0.4/0.8, matcher ratio 0.8; V-GICP voxel 0.1, res 0.3, max correspondence 0.2
    Hand-chosen configuration; the paper states TEASER++ was selected because it empirically produced more robust initial alignments.
  • Simulated Quest 3 depth-scanner noise parameters = Table C.2 (e.g., β=0.002, σ₀=0.0015, p_burst=0.015)
    Many hand-set parameters in the simulated scanner; conclusions from the Replica ablation depend on these approximating real Quest 3 depth behaviour.
assumptions (6)
  • domain assumption Gaussian centres provide a sufficient geometric source point set for registering 3DGS reconstructions
    Used for all 3DGS registration in the real-world benchmark; the paper itself motivates the anisotropy filter because centres can lie off the surface (§4.3).
  • domain assumption ArUco markers are a valid sparse ground-truth reference for registration error
    Used to measure centimetre-level errors; pose error of 1–2 mm is cited from controlled setups, but markers are sparse and mounted only at eye level.
  • domain assumption Quest 3 Depth API point clouds provide a sufficiently accurate metric target reference
    Target clouds come from a modified Depth-Scanner-Project; no independent accuracy characterization of the depth scans is provided.
  • domain assumption The Replica simulated scanner captures the relevant noise and artefacts of real Quest 3 depth sensing
    Anisotropy-filter and coarse-registration conclusions are validated against simulated scans rather than real on-device scans.
  • domain assumption VST-referenced image metrics are meaningful despite different camera pipelines
    PSNR/SSIM/LPIPS/DISTS compare reconstruction renderings against live passthrough; authors note absolute values may reflect headset exposure and tone mapping.
  • standard math CLMM and FDR-BH statistical assumptions hold for ordinal ratings with n=30
    Standard ordinal regression assumptions are invoked; the sample is small and participants completed ten conditions each.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MR-Compare: A Mixed-Reality Framework for Spatially Grounded Visual Comparison of 3D Gaussian Splatting and Mesh Reconstructions with the Physical Environment." pith.science (2026). https://pith.science/paper/LZZT6XYE

@misc{pith2026260720325,
  author       = {Pith},
  title        = {Pith review of: MR-Compare: A Mixed-Reality Framework for Spatially Grounded Visual Comparison of 3D Gaussian Splatting and Mesh Reconstructions with the Physical Environment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LZZT6XYE}},
  note         = {Machine review of arXiv:2607.20325}
}
abstract

We introduce MR-Compare, a mixed reality framework for spatially grounded visual comparison between 3D Gaussian splatting and mesh reconstructions with live video see-through (VST). Implemented on a PC-tethered Meta Quest~3, it combines a two-stage registration pipeline with a 3D Slider for cross-media comparison. We evaluated five representative desktop and mobile reconstruction workflows through a real-world benchmark with an exploratory user study ($n=30$) in two static indoor rooms. MR-Compare achieved centimetre-level translation error across all workflows. The two desktop 3DGS workflows showed the strongest overall pattern, with 3DGS-MCMC yielding the lowest registration error and strongest VST-referenced visual consistency. Room-session measures indicated high perceived usability and low workload. We further propose an anisotropy filter, a zero-shot module that leverages Gaussian anisotropies to improve 3DGS registration in MR-Compare. A controlled Replica threshold sweep shows that moderate pruning can improve robustness and reduce residual errors. These results establish system-level feasibility in the tested setting rather than task-level effectiveness or standalone deployment. The project is available at https://github.com/changruizhu96/MR-Compare.

Figures

Figures reproduced from arXiv: 2607.20325 by the authors.

Figure 1
Figure 1. Architecture of the MR-Compare: from data collection and extraction (a: source points; b: target points), through the spatial [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Reconstruction renderings and spatially registered live VST views in [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Participants’ predicted response distributions for perceived translation, rotation, and scale misalignment ( [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Participants’ predicted response distributions for each visual consistency score in [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Runtime and point retention across τ for 3DGS and 3DGS￾MCMC. Coloured lines show individual scenes; the black line and grey band indicate the mean and SD. Scene labels office0–office4 and room0– room2 are Replica identifiers. regions typically satisfy λ3 ≪ λ1,λ2, motiv…
Figure 6
Figure 6. Figure 6: τ-sweep registration error heatmaps. Rows show translation (top, cm) and rotation (bottom, °) errors across scenes and anisotropy thresholds τ. Numbers are shown for τ=1 (anisotropy filter disabled) and per-scene minima; bold enlarged numbers mark the minimum. Non￾conv…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

78 extracted references · 14 canonical work pages

  1. [2]

    J. T. Barron, B. Mildenhall, D. Verbin, P. P. Srinivasan, and P. Hedman. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition (CVPR), pp. 5470–5479. IEEE, 2022. doi: 10.1109/CVPR52688. 2022.00539 2

    arXiv 2022

  2. [3]

    Bellavia, C

    F. Bellavia, C. Colombo, L. Morelli, and F. Remondino. Challenges in image matching for cultural heritage: An overview and perspective. In International Conference on Image Analysis and Processing, pp. 210–222. Springer, 2022. doi: 10.1007/978-3-031-13321-3_19 1

    work page doi:10.1007/978-3-031-13321-3_19 2022

  3. [4]

    Benjamini and Y

    Y . Benjamini and Y . Hochberg. Controlling the false discovery rate: a practical and powerful approach to multiple testing.Journal of the Royal statistical society: series B (Methodological), 57(1):289–300, 1995. doi: 10.1111/j.2517-6161.1995.tb02031.x 5

    arXiv 1995

  4. [5]

    P. J. Besl and N. D. McKay. Method for registration of 3-d shapes. In Sensor fusion IV: control paradigms and data structures, vol. 1611, pp. 586–606. Spie, 1992. doi: 10.1117/12.57955 3

    doi:10.1117/12.57955 1992

  5. [6]

    Bonopera, J

    S. Bonopera, J. Esnault, S. Prakash, S. Rodriguez, T. Thonat, M. Benadel et al. sibr: A system for image based rendering, 2020. 2

    2020

  6. [7]

    Chang, Y

    J. Chang, Y . Xu, Y . Li, Y . Chen, W. Feng, and X. Han. Gaussreg: Fast 3d registration with gaussian splatting. InEuropean Conference on Computer Vision, pp. 407–423. Springer, 2024. doi: 10.1007/978-3-031-72633-0_23 3

    work page doi:10.1007/978-3-031-72633-0_23 2024

  7. [8]

    Z. Chen, K. Sun, F. Yang, and W. Tao. Sc2-pcr: A second order spatial compatibility for efficient and robust point cloud registration. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 13221–13231, 2022. doi: 10.1109/CVPR52688. 2022.01287 3

    arXiv 2022

  8. [9]

    R. H. B. Christensen. ordinal—regression models for ordinal data.R pack- age version, 10(2019):54, 2019. doi: 10.32614/CRAN.package.ordinal 5

    work page doi:10.32614/cran.package.ordinal 2019

Show all 78 references
  1. [10]

    DeTone, T

    D. DeTone, T. Malisiewicz, and A. Rabinovich. Superpoint: Self- supervised interest point detection and description. InProceedings of the IEEE conference on computer vision and pattern recognition work- shops, pp. 224–236, 2018. doi: 10.1109/CVPRW.2018.00060 2

    2018

  2. [11]

    K. Ding, K. Ma, S. Wang, and E. P. Simoncelli. Image quality assessment: Unifying structure and texture similarity.IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(5):2567–2581, 2022. doi: 10.1109/ TPAMI.2020.3045810 4

    2022

  3. [12]

    Realityscan: 3d models from images and laser scans, 2025

    Epic Games. Realityscan: 3d models from images and laser scans, 2025. RealityScan website. 2

    2025

  4. [13]

    M. J. Eppler. A comparison between concept maps, mind maps, conceptual diagrams, and visual metaphors as complementary tools for knowledge construction and sharing.Information visualization, 5(3):202–210, 2006. doi: 10.1057/palgrave.ivs.9500131 1

    2006 doi

  5. [14]

    B. J. Fernández-Palacios, D. Morabito, and F. Remondino. Access to complex reality-based 3d models using virtual reality solutions.Journal of cultural heritage, 23:40–48, 2017. doi: 10.1016/j.culher.2016.09.003 2

    2017 doi

  6. [15]

    T. Fick, J. Meulstee, M. Köllen, J. Van Doormaal, T. Van Doormaal, and E. Hoving. Comparing the influence of mixed reality, a 3d viewer, and mri on the spatial understanding of brain tumours.Frontiers in Virtual Reality, 4:1214520, 2023. doi: 10.3389/frvir.2023.1214520 2

    2023

  7. [16]

    Garrido-Jurado, R

    S. Garrido-Jurado, R. Muñoz-Salinas, F. J. Madrid-Cuevas, and M. J. Marín-Jiménez. Automatic generation and detection of highly reliable fiducial markers under occlusion.Pattern Recognition, 47(6):2280–2292,

  8. [17]

    Gleicher, D

    M. Gleicher, D. Albers, R. Walker, I. Jusufi, C. D. Hansen, and J. C. Roberts. Visual comparison for information visualization.Information Visualization, 10(4):289–309, 2011. doi: 10.1177/1473871611416549 2

    2011 doi

  9. [18]

    M. Grupp. evo: Python package for the evaluation of odometry and slam. https://github.com/MichaelGrupp/evo, 2017. 17

    2017

  10. [19]

    Guédon and V

    A. Guédon and V . Lepetit. Sugar: Surface-aligned gaussian splatting for efficient 3d mesh reconstruction and high-quality mesh rendering.CVPR,

  11. [20]

    S. G. Hart. Nasa-task load index (nasa-tlx); 20 years later. InProceedings of the human factors and ergonomics society annual meeting, vol. 50, pp. 904–908. Sage publications Sage CA: Los Angeles, CA, 2006. doi: 10. 1177/154193120605000909 4

    2006

  12. [22]

    Huang, Z

    S. Huang, Z. Gojcic, M. Usvyatsov, A. Wieser, and K. Schindler. Predator: Registration of 3d point clouds with low overlap. InProceedings of the IEEE/CVF Conference on computer vision and pattern recognition, pp. 4267–4276, 2021. doi: 10.1109/CVPR46437.2021.00425 3

    2021

  13. [23]

    Huynh-Thu and M

    Q. Huynh-Thu and M. Ghanbari. Scope of validity of psnr in image/video quality assessment.Electronics letters, 44(13):800–801, 2008. doi: 10. 1049/el:20080522 4

    2008

  14. [24]

    Jawset: Postshot – high fidelity 3d reconstruc- tion software, 2025

    Jawset Visual Computing. Jawset: Postshot – high fidelity 3d reconstruc- tion software, 2025. Jawset website. 2

    2025

  15. [25]

    Jiang, C

    Y . Jiang, C. Yu, T. Xie, X. Li, Y . Feng, H. Wang et al. Vr-gs: A physical dynamics-aware interactive gaussian splatting system in virtual reality. InACM SIGGRAPH 2024 Conference Papers, pp. 1–1, 2024. doi: 10. 1145/3641519.3657448 2

    2024

  16. [26]

    L. Joos, S. Jaeger-Honz, F. Schreiber, D. A. Keim, and K. Klein. Visual comparison of networks in vr.IEEE Transactions on Visualization and Computer Graphics, 28(11):3651–3661, 2022. doi: 10.1109/TVCG.2022. 3203001 2

    2022 doi

  17. [27]

    Kazhdan, M

    M. Kazhdan, M. Bolitho, and H. Hoppe. Poisson surface reconstruc- tion. InProceedings of the fourth Eurographics symposium on Geometry processing, vol. 7, 2006. doi: 10.2312/SGP/SGP06/061-070 2

    2006 doi

  18. [28]

    R. S. Kennedy, J. M. Drexler, D. E. Compton, K. M. Stanney, D. S. Lanham, and D. L. Harm. Configural scoring of simulator sickness, cybersickness and space adaptation syndrome: similarities and differences. InVirtual and Adaptive Environments, pp. 247–278. CRC Press, 2003. doi...

    2003 doi

  19. [29]

    R. S. Kennedy, N. E. Lane, K. S. Berbaum, and M. G. Lilienthal. Simulator sickness questionnaire: An enhanced method for quantifying simulator sickness.The international journal of aviation psychology, 3(3):203–220,

  20. [30]

    Kerbl, G

    B. Kerbl, G. Kopanas, T. Leimkühler, and G. Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics (TOG), 42(4):1–14, 2023. doi: 10.1145/3592433 2, 8, 18

    2023 doi

  21. [31]

    Kheradmand, D

    S. Kheradmand, D. Rebain, G. Sharma, W. Sun, Y .-C. Tseng, H. Isack et al. 3d gaussian splatting as markov chain monte carlo. InAdvances in Neural Information Processing Systems (NeurIPS), 2024. doi: 10.52202/ 079017-2573 2, 9

    2024

  22. [32]

    K. Kim, J. V . Carlis, and D. F. Keefe. Comparison techniques utilized in spatial 3d and 4d data visualizations: A survey and future directions. Computers & Graphics, 67:138–147, 2017. doi: 10.1016/j.cag.2017.05. 005 1, 2

    2017 doi

  23. [33]

    K. Kim, B. Jackson, I. Karamouzas, M. Adeagbo, S. J. Guy, R. Graff et al. Bema: A multimodal interface for expert experiential analysis of political assemblies at the pnyx in ancient greece. In2015 IEEE Symposium on 3D User Interfaces (3DUI), pp. 19–26. IEEE, 2015. doi: 10.110...

    2015 doi

  24. [34]

    Kleinbeck, H

    C. Kleinbeck, H. Schieber, K. Engel, R. Gutjahr, and D. Roth. Multi-layer gaussian splatting for immersive anatomy visualization.IEEE Transac- tions on Visualization and Computer Graphics, pp. 1–11, 2025. doi: 10. 1109/TVCG.2025.3549882 2

    2025

  25. [35]

    Koide, M

    K. Koide, M. Yokozuka, S. Oishi, and A. Banno. V oxelized GICP for Fast and Accurate 3D Point Cloud Registration. In2021 IEEE International Conference on Robotics and Automation (ICRA), pp. 11054–11059, May

  26. [36]

    V . J. León-Muñoz, F. Santonja-Medina, F. Lajara-Marco, A. J. Lisón- Almagro, J. Jiménez-Olivares, C. Marín-Martínez et al. The accuracy and absolute reliability of a knee surgery assistance system based on aruco-type sensors.Sensors, 23(19):8091, 2023. doi: 10.3390/s23198091 4

    2023 doi

  27. [37]

    J. R. Lewis, B. S. Utesch, and D. E. Maher. Umux-lite: when there’s no time for the sus. InProceedings of the SIGCHI conference on human factors in computing systems, pp. 2099–2102, 2013. doi: 10.1145/2470654 .2481287 4

    doi

  28. [38]

    H. Lim, D. Kim, G. Shin, J. Shi, I. Vizzo, H. Myung et al. Kiss-matcher: Fast and robust point cloud registration revisited. In2025 IEEE Interna- tional Conference on Robotics and Automation (ICRA), pp. 11104–11111. IEEE, 2025. doi: 10.1109/ICRA55743.2025.11127458 3

    2025

  29. [39]

    R. Liu, S. Ye, Z. Ding, G. Yang, S. Cheng, and K. Mueller. Drcmpvis: Visual comparison of physical targets in mobile diminished and mixed reality.IEEE Transactions on Visualization and Computer Graphics, 30(12):7672–7686, 2024. doi: 10.1109/TVCG.2024.3358419 2

    2024

  30. [40]

    S. Liu, D. Yang, Y . Gao, B. Ren, Y . Yang, and M. Fu. Automated 3d-gs reg- istration and fusion via skeleton alignment and gaussian-adaptive features. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 8158–8165. IEEE, 2025. doi: 10.1109/IR...

    2025

  31. [42]

    Interactive scenes – gorgeous, embeddable, and universally shareable 3d, 2025

    Luma AI, Inc. Interactive scenes – gorgeous, embeddable, and universally shareable 3d, 2025. Interactive Scenes page, Luma AI website. 2

    2025

  32. [43]

    S. LYi, J. Jo, and J. Seo. Comparative layouts revisited: Design space, guidelines, and future directions.IEEE Transactions on Visualization and Computer Graphics, 27(2):1525–1535, 2020. doi: 10.1109/TVCG.2020. 3030419 1, 2

    2020 doi

  33. [44]

    Marino, L

    E. Marino, L. Barbieri, B. Colacino, A. K. Fleri, and F. Bruno. An augmented reality inspection tool to support workers in industry 4.0 en- vironments.Computers in Industry, 127:103412, 2021. doi: 10.1016/j. compind.2021.103412 2

    2021

  34. [45]

    Mildenhall, P

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng. NeRF: Representing scenes as neural radiance fields for view synthesis. InProceedings of the European Conference on Computer Vision (ECCV), pp. 405–421. Springer, 2020. doi: 10.1007/978-3-030 ...

    2020 doi

  35. [46]

    Müller, A

    T. Müller, A. Evans, C. Schied, and A. Keller. Instant neural graphics primitives with a multiresolution hash encoding. InACM Transactions on Graphics (TOG), Proceedings of SIGGRAPH 2022, vol. 41, pp. 102:1– 102:15. ACM, 2022. doi: 10.1145/3528223.3530127 2

    2022

  36. [47]

    Scaniverse: Free 3 d scanner with lidar and gaussian splatting, 2025

    Niantic Spatial, Inc. Scaniverse: Free 3 d scanner with lidar and gaussian splatting, 2025. Scaniverse website. 2

    2025

  37. [48]

    Pauly, M

    M. Pauly, M. Gross, and L. P. Kobbelt. Efficient simplification of point- sampled surfaces. InIEEE Visualization, 2002. VIS 2002., pp. 163–170. IEEE, 2002. doi: 10.1109/VISUAL.2002.1183771 7

    2002 arXiv

  38. [49]

    Pharr, W

    M. Pharr, W. Jakob, and G. Humphreys.Physically Based Rendering: From Theory to Implementation. Morgan Kaufmann, 4 ed., 2023. 2

    2023

  39. [50]

    H.-P. Piepho. An algorithm for a letter-based representation of all- pairwise comparisons.Journal of Computational and Graphical Statistics, 13(2):456–466, 2004. doi: 10.1198/1061860043515 5

    2004 doi

  40. [51]

    Polycam, 2025

    Polycam Inc. Polycam, 2025. Mobile App. 2

    2025

  41. [52]

    Poroykov, P

    A. Poroykov, P. Kalugin, S. Shitov, and I. Lapitskaya. Modeling aruco markers images for accuracy analysis of their 3d pose estimation. In 30th international conference on computer graphics and machine vision (GraphiCon 2020). Part, vol. 2, 2020. doi: 10.51130/GRAPHICON-2020 -2-4-14 4

    2020 doi

  42. [53]

    Pranckevi ˇcius

    A. Pranckevi ˇcius. Gaussian splatting playground in Unity. https: //github.com/aras-p/UnityGaussianSplatting, 2023. Accessed: 2026-02-27. 2

    2023

  43. [55]

    R. B. Rusu, N. Blodow, and M. Beetz. Fast point feature histograms (fpfh) for 3d registration. InIEEE International Conference on Robotics and Automation (ICRA), pp. 3212–3217. IEEE, 2009. doi: 10.1109/ROBOT. 2009.5152473 3

    2009

  44. [56]

    Saredakis, A

    D. Saredakis, A. Szpak, B. Birckhead, H. A. Keage, A. Rizzo, and T. Loetscher. Factors associated with virtual reality sickness in head- mounted displays: a systematic review and meta-analysis.Frontiers in human neuroscience, 14:96, 2020. doi: 10.3389/fnhum.2020.00096 9

    2020

  45. [57]

    Schleising, C

    A. Schleising, C. Kunert, T. Schwandt, and W. Broll. Virtual pass-through: Evaluating 3d gaussian splatting as an alternative to conventional video pass-through in static environments. In2025 IEEE International Sympo- sium on Mixed and Augmented Reality (ISMAR), pp. 1257–1266. IEEE,

  46. [58]

    J. L. Schonberger and J.-M. Frahm. Structure-from-motion revisited. InProceedings of the IEEE conference on computer vision and pattern recognition, pp. 4104–4113, 2016. doi: 10.1109/CVPR.2016.445 2, 8

    2016 doi

  47. [59]

    Segal, D

    A. Segal, D. Haehnel, and S. Thrun. Generalized-icp. InRobotics: science and systems, vol. 2, p. 435. Seattle, W A, 2009. doi: 10.7551/mitpress/8727 .003.0022 3

    2009 doi

  48. [60]

    Snavely, S

    N. Snavely, S. M. Seitz, and R. Szeliski. Photo tourism: exploring photo collections in 3d. InACM siggraph 2006 papers, pp. 835–846, 2006. doi: 10.1145/1179352.1141964 2, 8

    2006

  49. [61]

    Speicher, B

    M. Speicher, B. D. Hall, and M. Nebeling. What is mixed reality? In Proceedings of the 2019 CHI conference on human factors in computing systems, pp. 1–15, 2019. doi: 10.1145/3290605.3300767 2

    2019

  50. [62]

    Straub, T

    J. Straub, T. Whelan, L. Ma, Y . Chen, E. Wijmans, S. Green et al. The replica dataset: A digital replica of indoor spaces.arXiv preprint arXiv:1906.05797, 2019. 2

    1906 arXiv

  51. [63]

    Tancik, E

    M. Tancik, E. Weber, E. Ng, R. Li, B. Yi, J. Kerr et al. Nerfstudio: A modular framework for neural radiance field development. InACM SIGGRAPH 2023 Conference Proceedings, SIGGRAPH ’23, 2023. doi: 10.1145/3588432.3591516 2

    2023

  52. [64]

    X. Tu, L. Radl, M. Steiner, M. Steinberger, B. Kerbl, and F. de la Torre. Vrsplat: Fast and robust gaussian splatting for virtual reality.Proceedings of the ACM on Computer Graphics and Interactive Techniques, 8(1):1–22,

  53. [65]

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli. Image quality assessment: From error visibility to structural similarity.IEEE Transac- tions on Image Processing, 13(4):600–612, 2004. doi: 10.1109/TIP.2003. 819861 4

    2004 doi

  54. [66]

    S. Yan, P. Shi, Z. Zhao, K. Wang, K. Cao, J. Wu et al. Turboreg: Turbo- clique for robust and efficient point cloud registration. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 26371–26381, 2025. doi: 10.1109/ICCV51701.2025.02447 3, 7

    2025

  55. [67]

    H. Yang, J. Shi, and L. Carlone. Teaser: Fast and certifiable point cloud registration.IEEE Transactions on Robotics, 37(2):314–333, 2020. doi: 10.1109/TRO.2020.3033695 3

    2020

  56. [68]

    B. Ye, S. Liu, H. Xu, X. Li, M. Pollefeys, M.-H. Yang et al. No pose, no problem: Surprisingly simple 3d gaussian splats from sparse unposed images. InInternational Conference on Learning Representations, vol. 2025, pp. 54009–54033, 2025. 3

    2025

  57. [69]

    V . Ye, R. Li, J. Kerr, M. Turkulainen, B. Yi, Z. Pan et al. gsplat: An open-source library for gaussian splatting.Journal of Machine Learning Research, 26(34):1–17, 2025. 2

    2025

  58. [70]

    doi: 10.1145/3728311 2

    doi

  59. [71]

    Zhang, P

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang. The unreason- able effectiveness of deep features as a perceptual metric. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 586–595, 2018. doi: 10.1109/CVPR.2018.00068 4

    2018

  60. [72]

    Zhang, H

    X. Zhang, H. Wu, M. Wu, and C. Wu. Extended motion diffusion-based change detection for airport ground surveillance.IEEE Transactions on Image Processing, 29:5677–5686, 2020. doi: 10.1109/TIP.2020.2984854 1

    2020

  61. [73]

    Q.-Y . Zhou, J. Park, and V . Koltun. Fast global registration. InEuropean Conference on Computer Vision (ECCV), pp. 766–782. Springer, 2016. doi: 10.1007/978-3-319-46475-6_47 3, 17

    2016 doi

  62. [74]

    C. Zhu, E. Kruijff, V . M. Pawar, and S. Julier. Evaluating 3d visual compar- ison techniques for change detection in virtual reality.IEEE Transactions on Visualization and Computer Graphics, 2025. doi: 10.1109/TVCG.2025 .3549578 3

    2025 doi

  63. [75]

    C. Zhu, E. Kruijff, H. Stedman, V . M. Pawar, and S. Julier. Does dis- play type matter for change detection? comparing immersive and non- immersive displays under low and high semantic availability.Displays, 93:103379, 2026. doi: 10.1016/j.displa.2026.103379 1

    2026

  64. [76]

    Z. Yu, A. Chen, B. Huang, T. Sattler, and A. Geiger. Mip-splatting: Alias- free 3d gaussian splatting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2024. doi: 10.1109/CVPR52733.2024.01839 2

    2024

  65. [82]

    Z. Zhu, S. Peng, V . Larsson, W. Xu, H. Bao, Z. Cui et al. Nice-slam: Neural implicit scalable encoding for slam. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 12786–12796,

  66. [1993]

    doi: 10.1207/s15327108ijap0303_3 4

    doi

  67. [2014]

    doi: 10.1016/j.patcog.2014.01.005 2

    2014 doi

  68. [2021]

    doi: 10.1109/ICRA48506.2021.9560835 3

    2021

  69. [2022]

    doi: 10.1109/CVPR52688.2022.01245 7 A SCENEDESIGN The floor plans of theOfficeandReceptionstudy spaces are shown in Fig. A.1. 4.20 m 3.27 m 4.20 m 3.27 m Approximately 14 m2 total Floorplan 1 m (a)Officefloor plan. 6.12 m 4.34 m 3.13 m 6.12 m 3.81 m 1.48 m Approximately 20 m2 ...

    2022

  70. [2025]

    doi: 10.1109/ISMAR67309.2025.00131 2

    2025

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.