Pith. sign in

REVIEW 2 major objections 4 minor 29 references

Incremental Online Scene Reconstruction by 3D Gaussian Triangulation

T0 review · 2 major / 4 minor · reviewed 2026-07-14 · grok-4.5

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

Pith's one-line read 3D Gaussians can be triangulated online into high-fidelity meshes without offline implicit conversion.

desk verdict Solid engineering fix for the offline-meshing bottleneck in Gaussian surface reconstruction; useful incremental RGB-D pipeline with clean ablations, not a fundamental leap. read the letter →

arxiv 2607.10690 v1 pith:7RFWMUFB submitted 2026-07-12 cs.CV

Yanjin Zhu , Shaofan Liu , Jianke Zhu This is my paper
classification cs.CV
keywords 3DGaussianSplattingincrementalreconstructiontriangulationsurfacemeshingonlinemappingRGB-Dplanarsurfels
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

Most 3D Gaussian methods produce beautiful novel views but still convert the finished Gaussians offline into an implicit field before a mesh can be extracted. This paper claims that the conversion step is unnecessary. By forcing each Gaussian to behave as a flat, nearly opaque surfel and by pulling its center onto a local plane estimated from oriented depth points, the set of Gaussians becomes a triangulation-ready surface. A lightweight greedy algorithm then builds and updates triangle meshes on the fly as new RGB-D frames arrive, while previously optimized regions are frozen so memory and compute stay bounded. On standard indoor sequences the resulting meshes are both more accurate and more complete than those of prior Gaussian SLAM systems, and the same representation still renders photorealistic images. The practical payoff is continuous, expandable surface models that can be handed to downstream robotics or AR tasks without waiting for a global recompute.

What carries the argument

Direct Gaussian triangulation: after plane-based pulling and opacity/depth filtering produce a geometric Gaussian set, neighbors are projected onto the local tangent plane, sorted by angle, and connected by a greedy 10-degree-angle rule, with local remeshing and region freezing keeping the mesh consistent and memory-bounded.

What would settle it

Run the full pipeline on an RGB-D sequence that contains large depth holes or missing views and measure whether the resulting mesh still improves accuracy and completion over ball-pivoting or Marching-Cubes baselines; any degradation to worse-than-baseline geometry would falsify the claim that the constrained Gaussians remain triangulation-ready.

Watch

Extended reading notes

Core claim

A dense geometric Gaussian representation whose primitives are constrained to be planar elliptical surfels can be triangulated directly into high-fidelity explicit meshes while the scene is still being scanned, eliminating the offline implicit-field conversion that has previously blocked seamless online reconstruction.

Load-bearing premise

The method needs accurate depth maps and known camera poses; without them the plane-pulling step cannot produce a clean, hole-free set of surfels ready for triangulation.

Editorial extensions

If this is right

  • Downstream robotics and AR systems can receive an expanding, watertight mesh without waiting for a global offline extraction step.
  • Memory and compute remain roughly constant for long sequences because fully optimized historical regions are frozen out of the active graph.
  • The same Gaussian set continues to support real-time novel-view rendering, so reconstruction and appearance stay synchronized.
  • Local remeshing after each new frame removes topological conflicts that would otherwise accumulate when meshes from different viewpoints are fused.

Reading between the lines

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

  • If the plane-pulling idea generalizes, pure-RGB monocular streams might later be handled by replacing measured depth with multi-view stereo or monocular depth estimates, though that extension is left open by the authors.
  • The freezing heuristic could be reused by other incremental Gaussian SLAM systems that currently keep every primitive active, offering an immediate route to longer trajectories.
  • Because the mesh is built directly from the same primitives used for rendering, any future improvement in Gaussian density control would automatically improve both surface fidelity and image quality.
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

2 major / 4 minor

Summary. The paper proposes an online incremental RGB-D reconstruction framework that maintains a dense geometric Gaussian representation (planar elliptical surfels with high opacity) and directly triangulates a filtered Geometric Gaussian Set G_geo to produce and update explicit meshes without an intermediate implicit field. Key technical pieces are: (i) dual-branch rendering with opaque-plane depth (Eq. 3), (ii) a plane-based pulling loss (Eq. 5) plus normal consistency (Eq. 6) and sparsity (Eq. 7) that align Gaussians to local surfaces, (iii) an angle-based greedy triangulation with local remeshing (Botsch-style edge split/collapse/flip + Laplacian), and (iv) dynamic freezing of fully optimized historical regions once observation count and loss thresholds are met. Experiments on Replica and ScanNet++ report superior Acc./Comp. ratios and PSNR/SSIM/LPIPS versus KinectFusion, NICE-SLAM, MonoGS and RTG-SLAM, together with higher FPS and lower peak memory, under ground-truth poses.

Significance. If the claims hold, the work removes a practical bottleneck of recent Gaussian-based surface methods (SuGaR, 2DGS, GOF, GS-Pull) that still rely on offline Poisson/Marching Cubes extraction. Direct, incremental triangulation of mesh-ready surfels plus region freezing yields a single representation that simultaneously supports high-fidelity novel-view synthesis and continuously updated watertight meshes, which is directly useful for AR and robotic perception. Strengths that raise the contribution above pure engineering are the concrete efficiency numbers (Tabs. 4–5: 10.34 FPS, 2325 MB, 5.34 s mesh extraction) and the ablations (Tab. 6, Figs. 6–7) that isolate each geometric constraint and the selection threshold τ. The approach is empirical rather than theoretical, yet the operating regime (RGB-D + known poses) is clearly stated and the limitation of unobserved regions is acknowledged.

major comments (2)
  1. All quantitative tables (Tabs. 1–3) and efficiency numbers (Tab. 4) are obtained with ground-truth camera poses (explicitly stated in §4.1). Because the plane-pulling loss (Eq. 5), geometric-Gaussian selection (Eq. 9) and freezing criteria all depend on multi-view consistency of depth and normals, the reported gains may shrink under realistic pose noise. A controlled experiment with noisy or SLAM-estimated poses (or at least a sensitivity study) is needed to substantiate the claim of practical online reconstruction.
  2. The geometric selection threshold τ=0.001 (Eq. 9) and the freezing thresholds N_obs=10, ε_gs=0.1 are fixed after a limited sweep (Fig. 7) performed only on Replica. No cross-dataset validation or automatic schedule is provided; the same values are used for ScanNet++. Because both mesh fidelity and memory savings hinge on these cut-offs, a short sensitivity analysis on ScanNet++ (or a statement of how they transfer) is required before the efficiency claims can be considered robust.
minor comments (4)
  1. Fig. 4 caption states that RTG-SLAM meshes are obtained by ball-pivoting; the main text should also note whether the same post-processing is applied to MonoGS or whether only point sampling is used, to keep the Acc./Comp. comparison fully transparent.
  2. Notation for the dual-branch depth (Eq. 3) is dense; a short clarifying sentence or diagram inset would help readers distinguish the three cases (no intersection, front-facing opaque, otherwise).
  3. The active-window sizes (6 frames / 50 iters for Replica, 3 / 75 for ScanNet++) appear only in the implementation paragraph; moving them into a small hyper-parameter table would improve reproducibility.
  4. A few typographical issues remain (e.g., missing spaces after commas in several equations, “Gaussiansurfels” in §3.4). A careful proof-read is recommended.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the pipeline is an empirical engineering method whose losses, selection criteria, and triangulation are defined independently of the reported accuracy/rendering metrics.

full rationale

The paper presents an online RGB-D Gaussian framework whose core steps (planar/opacity constraints on primitives, L_plane + L_n + L_sparse supervision, geometric-set filtering via opacity/depth thresholds, angle-based greedy triangulation, local remeshing, and freezing of converged regions) are constructive algorithms and loss terms, not mathematical derivations that reduce to their own inputs. Evaluation metrics (Acc., Acc. Ratio, Comp. Ratio, PSNR/SSIM/LPIPS) are computed against external ground-truth meshes and images on Replica and ScanNet++; ablations (Tab. 6, Figs. 6–7) isolate each component without circular forcing. Hyper-parameters (λ weights, τ=0.001, N_obs=10) are tuned but do not redefine the target quantities. Citations to GS-Pull and GaussianShader supply prior ideas for pulling and normals; they are external works by different authors and are not load-bearing uniqueness claims. No self-definitional equations, fitted-input-as-prediction, or renamed known results appear. The derivation chain is therefore self-contained against the stated benchmarks.

Assumptions & free parameters 4 free parameters · 3 assumptions · 2 invented entities

The central claim rests on standard 3DGS rendering, the empirical effectiveness of planar-opaque constraints, hand-chosen loss weights and selection thresholds, and the assumption that local angular triangulation plus remeshing yields watertight incremental meshes. No new physical entities are postulated; free parameters are the usual hyper-parameters of a modern reconstruction pipeline.

free parameters (4)
  • loss weights λc, λd, λp, λn, λs = 0.8, 1.0, 0.05, 0.2, 0.001
    Set to 0.8/1.0/0.05/0.2/0.001; control the relative influence of color, depth, plane, normal and sparsity terms that drive the final mesh accuracy.
  • geometric selection threshold τ = 0.001
    Depth-fidelity cutoff that decides which Gaussians enter the triangulation set (Eq. 9); ablated but chosen by hand for best Acc./Comp. trade-off.
  • freezing thresholds Nobs and ϵgs = 10 and 0.1
    Observation count and loss floor that decide when a region is frozen; directly control memory/compute claims.
  • active window size and optimization iterations = dataset-specific
    Set differently per dataset (6/50 for Replica, 3/75 for ScanNet++); affect both quality and reported FPS.
assumptions (3)
  • domain assumption 3D Gaussians with si3 o0 and αi≈1 behave as planar elliptical surfels that can be directly triangulated into accurate surfaces.
    Stated in Sec. 3.2; underpins the entire direct-meshing claim.
  • ad hoc to paper Local plane-pulling loss (Eq. 5) plus normal consistency (Eq. 6) suffice to correct noisy depth and produce mesh-ready centers.
    Introduced in Sec. 3.3; ablation shows benefit but no theoretical guarantee of global consistency.
  • domain assumption Ground-truth camera poses are available for all frames.
    Explicitly used for fair evaluation (Sec. 4.1); real systems would need online pose estimation.
invented entities (2)
  • Dense Geometric Gaussian representation / Geometric Gaussian Set Ggeo
    purpose: Subset of high-opacity, depth-faithful Gaussians that serve as the sole input to the triangulation algorithm.
    Defined by Eq. 9; the paper’s meshing claim stands or falls on this selection being both dense enough and free of outliers.
  • Angle-based greedy Gaussian triangulation with local remeshing and freezing
    purpose: Convert the selected Gaussians into an incrementally updatable watertight mesh while bounding memory.
    Core algorithmic contribution (Sec. 3.4, Fig. 3); no external validation outside the reported experiments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Incremental Online Scene Reconstruction by 3D Gaussian Triangulation." pith.science (2026). https://pith.science/paper/7RFWMUFB

@misc{pith2026260710690,
  author       = {Pith},
  title        = {Pith review of: Incremental Online Scene Reconstruction by 3D Gaussian Triangulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7RFWMUFB}},
  note         = {Machine review of arXiv:2607.10690}
}
read the original abstract

Incremental scene reconstruction is essential for real-world applications. Although 3D Gaussian Splatting shows strong potential, most existing approaches require offline conversion of the optimized Gaussians into an intermediate implicit field for explicit mesh extraction, which hinders seamless integration with downstream tasks. To address this limitation, we propose a novel online framework that incrementally reconstructs and updates high-fidelity explicit meshes by directly triangulating a dense geometric Gaussian representation, which supports both high-quality rendering and incremental surface reconstruction. Moreover, we present a direct meshing algorithm that efficiently extracts and updates the mesh from the Gaussian set. To ensure mesh accuracy, we enforce a plane-based pulling constraint that dynamically aligns 3D Gaussian primitives to the approximated local surface. Furthermore, our framework significantly reduces memory and computational overhead during long-sequence processing by dynamically freezing fully optimized historical regions. Experiments on public datasets demonstrate that our method outperforms conventional Gaussian-based methods on both rendering quality and reconstruction accuracy.

Figures

Figures reproduced from arXiv: 2607.10690 by the authors.

Figure 1
Figure 1. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of our proposed approach. We first initialize Gaussians from a con￾tinuous stream of oriented point clouds. The dense geometric Gaussian representation is progressively refined by loss-guided densification and optimization against depth, color, and geometric constraints. Next, a geometric Gaussian set is selected for mesh￾ing, during which process redundant primitives are pruned to maintain compactness. A n… view at source ↗
Figure 3
Figure 3. Illustration of Gaussian triangulation pipeline. (a) Neighbors within an adaptive radius are identified and projected onto the tangent plane. (b) Valid candidates (green) are retained, while invisible (red) and normal-inconsistent (purple) ones are pruned. (c) The central Gaussian (blue) is shifted to a refined position (black) via weighted fusion. (d) Neighbors are re-projected onto the new tangent space and sorted… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Comparison of Mesh Reconstruction results across different Replica sequences. For RTG-SLAM [19], we employ the ball pivoting algorithm [2] to generate meshes from Gaussian balls [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Qualitative comparisons of novel view synthesis across different [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Qualitative effect of geometric constraints on mesh reconstruction. We display the reconstruction details. When only depth constraints are applied without geometric constraints, the reconstruction captures the general outline of the scene but suffers from low surface f…
Figure 7
Figure 7. Figure 7: Ablation Study on parameter τ to explore its impact on reconstruction quality. Geometric Constraints. To further validate the two geometric constraints Lplane+Ln, we provide a visual comparison of the reconstruction details in [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

29 extracted references · 2 linked inside Pith

  1. [1]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Barron, J.T., Mildenhall, B., Tancik, M., Hedman, P., Martin-Brualla, R., Srini- vasan, P.P.: Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 5855–5864 (2021)

    2021

  2. [2]

    IEEE transactions on visualization and computer graphics5(4), 349–359 (2002)

    Bernardini, F., Mittleman, J., Rushmeier, H., Silva, C., Taubin, G.: The ball- pivoting algorithm for surface reconstruction. IEEE transactions on visualization and computer graphics5(4), 349–359 (2002)

    2002

  3. [3]

    In: Proceedings of the 2004 Eurographics/ACM SIGGRAPH symposium on Geometry processing

    Botsch, M., Kobbelt, L.: A remeshing approach to multiresolution modeling. In: Proceedings of the 2004 Eurographics/ACM SIGGRAPH symposium on Geometry processing. pp. 185–192 (2004)

    2004

  4. [4]

    In: Forty-first International Conference on Machine Learning (2024)

    Cheng, K., Long, X., Yang, K., Yao, Y., Yin, W., Ma, Y., Wang, W., Chen, X.: Gaussianpro: 3d gaussian splatting with progressive propagation. In: Forty-first International Conference on Machine Learning (2024)

    2024

  5. [5]

    In: ACM SIGGRAPH 2024 conference papers

    Dai, P., Xu, J., Xie, W., Liu, X., Wang, H., Xu, W.: High-quality surface recon- struction using gaussian surfels. In: ACM SIGGRAPH 2024 conference papers. pp. 1–11 (2024)

    2024

  6. [6]

    Gopi, M.: A fast and efficient projection-based approach for surface reconstruction. Int. J. of High Performance Computer Graphics, Multimedia and Visualization 1(1), 1–12 (2000) 16 Y. Zhu et al

    2000

  7. [7]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Guédon, A., Lepetit, V.: Sugar: Surface-aligned gaussian splatting for efficient 3d mesh reconstruction and high-quality mesh rendering. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 5354– 5363 (2024)

    2024

  8. [8]

    Image and Vision Computing 68, 14–27 (2017)

    Häne, C., Heng, L., Lee, G.H., Fraundorfer, F., Furgale, P., Sattler, T., Pollefeys, M.: 3d visual perception for self-driving cars using a multi-camera system: Calibra- tion, mapping, localization, and obstacle detection. Image and Vision Computing 68, 14–27 (2017)

    2017

Show all 29 references
  1. [9]

    In: ACM SIGGRAPH 2024 conference papers

    Huang, B., Yu, Z., Chen, A., Geiger, A., Gao, S.: 2d gaussian splatting for geo- metrically accurate radiance fields. In: ACM SIGGRAPH 2024 conference papers. pp. 1–11 (2024)

    2024

  2. [10]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Jiang, Y., Tu, J., Liu, Y., Gao, X., Long, X., Wang, W., Ma, Y.: Gaussianshader: 3d gaussian splatting with shading functions for reflective surfaces. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 5322–5332 (2024)

    2024

  3. [11]

    In: Proceed- ings of the fourth Eurographics symposium on Geometry processing

    Kazhdan, M., Bolitho, M., Hoppe, H.: Poisson surface reconstruction. In: Proceed- ings of the fourth Eurographics symposium on Geometry processing. pp. 61–70 (2006)

    2006

  4. [12]

    ACM Trans

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph.42(4), 139–1 (2023)

    2023

  5. [13]

    ACM siggraph computer graphics21(4), 163–169 (1987)

    Lorensen, W.E., Cline, H.E.: Marching cubes: A high resolution 3d surface con- struction algorithm. ACM siggraph computer graphics21(4), 163–169 (1987)

    1987

  6. [14]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Lu, T., Yu, M., Xu, L., Xiangli, Y., Wang, L., Lin, D., Dai, B.: Scaffold-gs: Struc- tured 3d gaussians for view-adaptive rendering. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 20654–20664 (2024)

    2024

  7. [15]

    In: 2009 IEEE international conference on robotics and automation

    Marton, Z.C., Rusu, R.B., Beetz, M.: On fast surface reconstruction methods for large and noisy point clouds. In: 2009 IEEE international conference on robotics and automation. pp. 3218–3223. IEEE (2009)

    2009

  8. [16]

    In: ProceedingsoftheIEEE/CVFConferenceonComputerVisionandPatternRecog- nition

    Matsuki, H., Murai, R., Kelly, P.H., Davison, A.J.: Gaussian splatting slam. In: ProceedingsoftheIEEE/CVFConferenceonComputerVisionandPatternRecog- nition. pp. 18039–18048 (2024)

    2024

  9. [17]

    Commu- nications of the ACM65(1), 99–106 (2021)

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. Commu- nications of the ACM65(1), 99–106 (2021)

    2021

  10. [18]

    In: 2011 10th IEEE international symposium on mixed and augmented reality

    Newcombe, R.A., Izadi, S., Hilliges, O., Molyneaux, D., Kim, D., Davison, A.J., Kohi, P., Shotton, J., Hodges, S., Fitzgibbon, A.: Kinectfusion: Real-time dense surface mapping and tracking. In: 2011 10th IEEE international symposium on mixed and augmented reality. pp. 127–136...

    2011

  11. [19]

    Peng, Z., Shao, T., Liu, Y., Zhou, J., Yang, Y., Wang, J., Zhou, K.: Rtg-slam: Real- time3dreconstructionatscaleusinggaussiansplatting.In:ACMSIGGRAPH2024 Conference Papers. pp. 1–11 (2024)

    2024

  12. [20]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Sandström, E., Li, Y., Van Gool, L., Oswald, M.R.: Point-slam: Dense neural point cloud-based slam. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 18433–18444 (2023)

    2023

  13. [21]

    IEEE transactions on visualization and computer graphics23(11), 2455–2462 (2017)

    Schöps, T., Oswald, M.R., Speciale, P., Yang, S., Pollefeys, M.: Real-time view correction for mobile devices. IEEE transactions on visualization and computer graphics23(11), 2455–2462 (2017)

    2017

  14. [22]

    arXiv preprint arXiv:1906.05797 (2019) Incremental Online Scene Reconstruction by 3D Gaussian Triangulation 17

    Straub, J., Whelan, T., Ma, L., Chen, Y., Wijmans, E., Green, S., Engel, J.J., Mur-Artal, R., Ren, C., Verma, S., et al.: The replica dataset: A digital replica of indoor spaces. arXiv preprint arXiv:1906.05797 (2019) Incremental Online Scene Reconstruction by 3D Gaussian Tria...

    1906 arXiv

  15. [23]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Sucar, E., Liu, S., Ortiz, J., Davison, A.J.: imap: Implicit mapping and position- ing in real-time. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 6229–6238 (2021)

    2021

  16. [24]

    Waczyńska, J., Borycki, P., Tadeja, S., Tabor, J., Spurek, P.: Games: Mesh-based adapting and modification of gaussian splatting (2024),https://arxiv.org/abs/ 2402.01459

    2024 arXiv

  17. [25]

    In: Proceedings of the International Conference on Computer Vision (ICCV) (2023)

    Yeshwanth, C., Liu, Y.C., Nießner, M., Dai, A.: Scannet++: A high-fidelity dataset of 3d indoor scenes. In: Proceedings of the International Conference on Computer Vision (ICCV) (2023)

    2023

  18. [26]

    In: Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition

    Yu, Z., Chen, A., Huang, B., Sattler, T., Geiger, A.: Mip-splatting: Alias-free 3d gaussian splatting. In: Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition. pp. 19447–19456 (2024)

    2024

  19. [27]

    ACM Transactions on Graphics (ToG)43(6), 1–13 (2024)

    Yu, Z., Sattler, T., Geiger, A.: Gaussian opacity fields: Efficient adaptive surface reconstruction in unbounded scenes. ACM Transactions on Graphics (ToG)43(6), 1–13 (2024)

    2024

  20. [28]

    Advances in Neural Information Processing Systems37, 101856–101879 (2024)

    Zhang, W., Liu, Y.S., Han, Z.: Neural signed distance function inference through splatting 3d gaussians pulled on zero-level set. Advances in Neural Information Processing Systems37, 101856–101879 (2024)

    2024

  21. [29]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Zhu, Z., Peng, S., Larsson, V., Xu, W., Bao, H., Cui, Z., Oswald, M.R., Pollefeys, M.: Nice-slam: Neural implicit scalable encoding for slam. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 12786– 12796 (2022)

    2022

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.