Pith. sign in

REVIEW 5 major objections 5 minor 34 references

FGO-SLAM: Enhancing Gaussian SLAM with Globally Consistent Opacity Radiance Field

T0 review · 5 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read FGO-SLAM claims a Gaussian SLAM system that maintains a globally consistent opacity radiance field for mapping, uses global bundle adjustment for robust tracking, and extracts surfaces directly from 3D Gaussians via tetrahedral level sets,

desk verdict A credible integration of ORB-style global BA with Gaussian opacity fields, but the distinct mesh-extraction claim rests on an unproven level-set assumption and no quantitative mesh metrics. read the letter →

arxiv 2509.01547 v1 pith:MGV7CZ2O submitted 2025-09-01 cs.RO

Fan Zhu , Yifan Zhao , Ziyu Chen , Biao Yu , Hui Zhu This is my paper · ORCID
classification cs.RO
keywords GaussianSplattingSLAMopacityradiancefieldsurfaceextractionmarchingtetrahedracameraposeoptimizationbundleadjustmentmonoculardensemapping
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

FGO-SLAM is a visual SLAM system trying to solve two problems at once: Gaussian-based SLAM systems render and map quickly but lack effective camera-pose optimization and struggle to produce usable meshes. The paper argues that both problems yield to a single scene representation: a globally consistent opacity radiance field built from 3D Gaussian primitives. With this field, the system can optimize poses by global bundle adjustment over sparse feature points, refine geometry with depth-distortion and normal-consistency losses that do not need ground-truth depth, and then extract surfaces directly using tetrahedral grids and Marching Tetrahedra. If the claims hold, a monocular or RGB-D agent could track, map, and hand back a clean mesh in real time, which would matter for embodied AI simulation and downstream tasks. The paper reports improved tracking accuracy and rendering quality over NeRF- and Gaussian-based baselines on TUM, ScanNet, and Replica.

What carries the argument

The load-bearing object is the opacity radiance field: for each Gaussian primitive, the paper evaluates it as a 1D Gaussian along each camera ray (after ray-Gaussian intersection), volume-renders per-view opacity, then assigns the point a global opacity as the minimum over all views. This min-over-views construction is what makes the field globally consistent and thereby eligible for level-set geometry extraction. The second mechanism is the surface-extraction pipeline: per-Gaussian bounding boxes at the 3-sigma level, tetrahedral grids anchored at box centers and corners, edge-length filtering to remove connections between distant Gaussians, and Marching Tetrahedra with binary search to loc

What would settle it

On a real sequence with known poses (e.g. a ScanNet scene), render the per-view opacity O(o,r,d) at a dense grid of points near a reconstructed surface from two or more camera viewpoints, and measure how often the global min-over-views differs from the per-view values by more than a small threshold; persistent disagreement would refute the global-consistency premise. A complementary check: compute the F-score or Chamfer distance of the extracted mesh against ground-truth TSDF on Replica, which the paper does not report.

Watch

Extended reading notes

Core claim

The central claim, in the authors' terms, is that defining the opacity of any 3D point as the minimum of its per-view ray-Gaussian opacity values, O(P)=min over views, yields a view-consistent radiance field that behaves like a signed geometry field. This field lets the system optimize the Gaussian map without depth images — using only color loss plus depth-distortion and normal-consistency terms — and lets it extract surfaces directly from the Gaussians by building 3-sigma tetrahedral grids, filtering long edges, and running Marching Tetrahedra on binary-searched level sets. Combined with a classical feature-based visual odometry front-end and global bundle adjustment on loop closure, the s

Load-bearing premise

The load-bearing premise is that taking the minimum of per-view opacities at every 3D point yields a single view-consistent field whose level sets are actual surfaces; if the per-view opacity fields disagree enough, both the direct mesh extraction and the depth-free map optimization lose their foundation.

Editorial extensions

If this is right

  • A Gaussian SLAM system can output a usable mesh without a separate TSDF fusion or Poisson reconstruction step, because surfaces are read off the opacity field's level sets.
  • Global bundle adjustment on loop closure improves not only trajectory accuracy but also rendered image quality, since corrected poses feed a cleaner map.
  • Mapping does not require ground-truth depth: the two regularizers substitute for depth priors, which is what lets the system run with a monocular camera.
  • Keyframe-based mapping keeps per-frame cost low enough for 7.83 FPS on the evaluated Replica office scene, an order of magnitude faster than the compared NeRF-based systems.
  • On the tested datasets, the method beats or matches NeRF-based (NICE-SLAM, ESLAM) and Gaussian-based (Photo-SLAM, SplaTAM, MonoGS) baselines on tracking RMSE and rendering metrics.

Reading between the lines

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

  • If the min-over-views opacity is truly view-consistent, the same surface-extraction trick could be grafted onto other 3DGS SLAM front-ends that already produce good poses and maps, giving them mesh output for free.
  • The monocular results suggest depth-free geometric mapping might be more feasible in Gaussian SLAM than in NeRF-based systems; a direct comparison of monocular mesh accuracy against depth-based systems would quantify how much is gained.
  • A stress test that does not appear in the paper would be to measure the variance of per-view opacities on reflective or transparent surfaces, where the min operation may cut into the geometry; failure there would bound the field's consistency claim.
  • The qualitative-only mesh evaluation leaves open the possibility of replacing it with standard F-score/Chamfer metrics; if the mesh is as good as the rendering, FGO-SLAM would close a gap that 3DGS systems have left open.
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

5 major / 5 minor

Summary. The paper presents FGO-SLAM, a Gaussian Splatting SLAM system that combines feature-based visual odometry with global bundle adjustment and an 'opacity radiance field' defined over 3D Gaussians. The authors claim state-of-the-art tracking on TUM and ScanNet, high-quality rendering on Replica, real-time performance, and a distinctive ability to directly extract meshes from 3D Gaussians via marching tetrahedra, without requiring depth-based map optimization. The manuscript includes three main contributions: a global-adjustment tracking module, an opacity radiance field with depth-distortion and normal-consistency regularization, and a level-set surface extraction method.

Significance. If the technical claims were fully supported, the work would be a useful contribution to dense Gaussian SLAM: the tracking results are competitive, the rendering quality is high on the reported benchmarks, and direct mesh extraction from Gaussians would address a recognized limitation of existing 3DGS-based SLAM systems. The proposed global adjustment and regularization terms are plausible engineering choices that could benefit the community. However, the central theoretical claim of a 'globally consistent opacity radiance field' is not established, and the surface-extraction evaluation is qualitative only. The experimental reporting also contains inconsistencies that must be resolved before the significance of the results can be assessed.

major comments (5)
  1. [Sec. III-B, Eq. (7)] The transmittance product in Eq. (7) uses G1D_i(d) for all previous Gaussians j; standard volume rendering requires G1D_j(d). As written, O(o,r,d) is not a proper accumulated opacity. Moreover, O(P) in Eq. (6) is a minimum over a finite, incrementally growing set of keyframe views; no argument is given that this creates a well-defined scalar level-set function whose 0.5-isosurface is the scene surface. Since this field is the basis for the claimed direct surface extraction and for the 'globally consistent' representation, the central novelty is unsupported. Please correct the equation and provide either a proof or a systematic empirical validation of the view-consistency assumption.
  2. [Sec. III-B, Eq. (5) and text] The expression for the depth that maximizes G1D(d) contains a typographical error: the denominator should be o_g^T r_g, not 0_g^T r_g. More importantly, the text states that after reaching G1D_max the opacity 'remains constant'; this is false for a Gaussian, which decreases after its maximum. This error affects the intuition behind the depth-distortion loss and the opacity field. Please correct the formula and the accompanying explanation.
  3. [Sec. IV-A.3] The paper explicitly declines quantitative mesh evaluation ('we opted for a qualitative evaluation') and presents only images in Figs. 1, 3, and 6. Since direct surface extraction is a primary claimed contribution and a key differentiator from other Gaussian SLAM baselines, qualitative results are insufficient. Please report quantitative mesh metrics (e.g., F-score, accuracy, completeness, or Chamfer distance) on at least Replica and ScanNet, and compare against available mesh-producing baselines such as NICE-SLAM or ESLAM.
  4. [Sec. IV-D.1, Table V] There is a direct inconsistency: the main text states that Table V shows results on the TUM RGB-D dataset, while the table caption says the experiments were conducted on scene 0000 00 of the ScanNet dataset. This makes the global-adjustment ablation ambiguous and prevents the reader from verifying the claimed improvements. Please state the exact dataset and sequence for this ablation and correct the text/caption mismatch.
  5. [Sec. IV-A, metrics] The paper does not specify whether the rendering metrics (PSNR, SSIM, LPIPS) and depth L1 are computed on keyframes used during mapping optimization or on held-out views. If the reported metrics are on training keyframes, the comparison with baselines that report novel-view synthesis would be biased. Please clarify the evaluation protocol and, if necessary, report held-out or novel-view results to support the claimed state-of-the-art mapping performance.
minor comments (5)
  1. [Eq. (6)] The notation O(0, r, d) should presumably be O(o, r, d). Please fix the equation and ensure that the camera center variable is consistently defined.
  2. [Sec. III, cross-reference] The text in Section III says Sec. IV-B.2 describes the surface extraction method, but the actual description is in Sec. III-D. Please correct the cross-reference.
  3. [Sec. IV-B.2] Spelling typo: 'NICEC-SLAM' should be 'NICE-SLAM'.
  4. [Sec. IV-D.2] The ablation of the regularization terms is qualitative only (Fig. 7). Since the final loss includes weights α and β with α ranging from 100 to 1000, a quantitative ablation or sensitivity analysis would strengthen the claim that these terms are beneficial and not over-tuned.
  5. [Table IV] The real-time comparison in Table IV omits MonoGS and Photo-SLAM, which are in the other tables. For completeness, please include all compared baselines or explain why they are absent from this timing comparison.

Circularity Check

0 steps flagged · score 0.0 of 10
full rationale

The paper does not reduce any prediction to a fitted input or to a self-citation chain. Tracking uses ORB-SLAM-style feature reprojection and bundle adjustment (Eqs. 1-3), which are independent classical SLAM components. Mapping fits 3D Gaussians with color, depth-distortion, and normal-consistency losses (Eqs. 8-12); these are regularizers borrowed from external work [8,31], not from the authors' own prior claims. The surface-extraction module (Sec. III-D) takes the opacity field defined in Eqs. (5)-(7) and finds its level sets with marching tetrahedra; this is a standard definitional pipeline (an output is the isosurface of a fitted field), not a derivation that secretly assumes the conclusion. The paper's 'globally consistent opacity radiance field' is asserted through the min-over-views definition in Eq. (6), but no proof of view-invariance is given; that is an unsupported modeling assumption and a correctness risk, not a circular step. Similarly, the decision to evaluate meshes only qualitatively (Sec. IV-A.3) weakens the evidence but does not make the claim definitionally true. Finally, Eq. (7) contains an apparent transmittance-index typo (G1D_i instead of G1D_j) and the text's claim that opacity stays constant after the Gaussian maximum is technically wrong, but these are independent technical errors, not circularities. No load-bearing self-citations exist: reference [31] (Gaussian opacity fields) is by different authors, and all other key citations are external. The central claims are therefore not equivalent to their inputs by construction.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The system inherits most machinery from ORB-SLAM3, 3D Gaussian Splatting, and Gaussian Opacity Fields. Its own additions are heuristic choices, such as min-over-views opacity, loss weights, and tetrahedral grid thresholds, rather than derived parameters, and none is accompanied by sensitivity analysis or formal justification.

free parameters (4)
  • depth distortion weight alpha = 1000 (range 100-1000)
    Weights the depth distortion loss in Eq. 12; chosen by the authors on the evaluation datasets with no sensitivity analysis.
  • normal consistency weight beta = 0.05
    Weights the normal consistency loss in Eq. 12; fixed by hand.
  • tetrahedral bounding box scale = 3-sigma
    Section III-D builds tetrahedral grids around each Gaussian at the 3-sigma level; the threshold is heuristic.
  • tetrahedral edge filtering threshold = sum of maximum scales
    Section III-D removes cells whose edges exceed the sum of the maximum scales of the connected Gaussians; heuristic and unvalidated.
assumptions (6)
  • domain assumption ORB feature tracking and bundle adjustment from [2] provide reliable pose estimates and loop closures.
    Section III-A builds the entire tracking and global adjustment module on ORB-SLAM3's methodology.
  • domain assumption The 1D ray-Gaussian opacity composition from Gaussian Opacity Fields [31] approximates true scene occupancy.
    Section III-B adopts this model through Eqs. (4)-(7).
  • ad hoc to paper The min-over-views opacity at Eq. (6) yields a globally consistent opacity field.
    No theorem or external measurement supports this; it is the core representation assumption of the paper.
  • standard math Marching tetrahedra correctly extracts level sets from a scalar opacity field.
    Section III-D relies on the standard marching tetrahedra algorithm [23].
  • domain assumption Baseline numbers taken from other papers are directly comparable to the authors' runs.
    Section IV-A.2 states that some baseline results came from their published reports, which assumes identical evaluation protocols.
  • domain assumption Reconstruction metrics computed on the same sequence used for mapping are meaningful measures of mapping quality.
    Section IV-A.3 evaluates PSNR, SSIM, LPIPS, and depth on the reconstructed sequence itself, with no held-out frames or quantitative mesh metric.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FGO-SLAM: Enhancing Gaussian SLAM with Globally Consistent Opacity Radiance Field." pith.science (2026). https://pith.science/paper/MGV7CZ2O

@misc{pith2026250901547,
  author       = {Pith},
  title        = {Pith review of: FGO-SLAM: Enhancing Gaussian SLAM with Globally Consistent Opacity Radiance Field},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MGV7CZ2O}},
  note         = {Machine review of arXiv:2509.01547}
}
read the original abstract

Visual SLAM has regained attention due to its ability to provide perceptual capabilities and simulation test data for Embodied AI. However, traditional SLAM methods struggle to meet the demands of high-quality scene reconstruction, and Gaussian SLAM systems, despite their rapid rendering and high-quality mapping capabilities, lack effective pose optimization methods and face challenges in geometric reconstruction. To address these issues, we introduce FGO-SLAM, a Gaussian SLAM system that employs an opacity radiance field as the scene representation to enhance geometric mapping performance. After initial pose estimation, we apply global adjustment to optimize camera poses and sparse point cloud, ensuring robust tracking of our approach. Additionally, we maintain a globally consistent opacity radiance field based on 3D Gaussians and introduce depth distortion and normal consistency terms to refine the scene representation. Furthermore, after constructing tetrahedral grids, we identify level sets to directly extract surfaces from 3D Gaussians. Results across various real-world and large-scale synthetic datasets demonstrate that our method achieves state-of-the-art tracking accuracy and mapping performance.

Figures

Figures reproduced from arXiv: 2509.01547 by the authors.

Figure 1
Figure 1. Comparison of mapping effect. Our method con￾structs a complete mesh and provides higher-fidelity images along with more accurate camera pose tracking results. This method does not rely on truth depth images for map optimization, which allows us to work with a monocular camera. perceptrons (MLP) may result in catastrophic forgetting and over-smoothing, which challenges the ability of NeRF￾based dense Visual SLAM sys… view at source ↗
Figure 2
Figure 2. Overview. Our proposed system is composed of three modules divided into two stages: the tracking and mapping modules in the first stage, and the mesh extraction module in the second stage. The tracking and mapping modules maintain a globally consistent opacity radiance field as the foundational map. FGO-SLAM accepts RGB or RGB-D image frames, where the tracking module estimates the camera pose in real time. Upon det… view at source ↗
Figure 3
Figure 3. Surface Extraction. G 1D(d) = exp − 1 2  og + drg T  og + drg  (5) Clearly, when d = − r T g rg 0T g rg is reached, the function G 1D(d) reaches its maximum, denoted as G 1D max. This means that if there is only one Gaussian G along the ray, the opacity increases along the ray until it reaches G 1D max, after which it remains constant. Thus, we construct the opacity radiance field using each Gaussian’s opacity… view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Rendering results on the Replica dataset [24]. The dashed boxes highlight the differences between our approach and baselines. For the details of Windows and cabinets, our method significantly outperforms baselines. with other methods, we also use RGB-D mode for our exp…
Figure 6
Figure 6. Figure 6: Visual comparison of the mesh quality on the Replica dataset [24]. Previous Gaussian-based methods cannot generate mesh effectively, and therefore, cannot be compared here. (a) w/o regularization (b) w regularization [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Ablation of the regularization. 3) Evaluation on Replica [24]: We assessed our method’s scene reconstruction capabilities using the high-quality syn￾thetic dataset, Replica. As indicated in TABLE I, our ap￾proach achieves the best average rendering quality across all s…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 30 canonical work pages

  1. [1]

    Nerf2real: Sim2real transfer of vision-guided bipedal motion skills using neural radiance fields,

    A. Byravan, J. Humplik, L. Hasenclever, A. Brussee, F. Nori, T. Haarnoja, B. Moran, S. Bohez, F. Sadeghi, B. Vujatovic et al. , “Nerf2real: Sim2real transfer of vision-guided bipedal motion skills using neural radiance fields,” in 2023 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2023, pp. 9362–9369

    work page 2023

  2. [2]

    Orb-slam3: An accurate open-source library for visual, visual-inertial, and multimap slam,

    C. Campos, R. Elvira, J. J. G. Rodriguez, J. M. M. Montiel, and J. D. Tardos, “Orb-slam3: An accurate open-source library for visual, visual-inertial, and multimap slam,” IEEE Transactions on Robotics , vol. 37, no. 6, pp. 1874–1890, 2021

    work page 2021

  3. [3]

    Orbeez-slam: A real-time monocular visual slam with orb features and nerf-realized mapping,

    C.-M. Chung, Y .-C. Tseng, Y .-C. Hsu, X.-Q. Shi, Y .-H. Hua, J.-F. Yeh, W.-C. Chen, Y .-T. Chen, and W. H. Hsu, “Orbeez-slam: A real-time monocular visual slam with orb features and nerf-realized mapping,” in 2023 IEEE International Conference on Robotics and Automation (ICRA). IEEE, Conference Proceedings, pp. 9400–9406

    work page 2023

  4. [4]

    Scannet: Richly-annotated 3d reconstructions of indoor scenes,

    A. Dai, A. X. Chang, M. Savva, M. Halber, T. Funkhouser, and M. Nießner, “Scannet: Richly-annotated 3d reconstructions of indoor scenes,” in Proceedings of the IEEE conference on computer vision and pattern recognition , Conference Proceedings, pp. 5828–5839

    work page

  5. [5]

    Svo: Fast semi-direct monocular visual odometry,

    C. Forster, M. Pizzoli, and D. Scaramuzza, “Svo: Fast semi-direct monocular visual odometry,” in 2014 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2014, pp. 15–22

    work page 2014

  6. [6]

    Bags of binary words for fast place recognition in image sequences,

    D. G ´alvez-L´opez and J. D. Tardos, “Bags of binary words for fast place recognition in image sequences,” IEEE Transactions on robotics, vol. 28, no. 5, pp. 1188–1197, 2012

    work page 2012

  7. [7]

    Liv-gaussmap: Lidar-inertial-visual fusion for real-time 3d radiance field map render- ing,

    S. Hong, J. He, X. Zheng, C. Zheng, and S. Shen, “Liv-gaussmap: Lidar-inertial-visual fusion for real-time 3d radiance field map render- ing,” IEEE Robotics and Automation Letters , 2024

    work page 2024

  8. [8]

    2d gaussian splat- ting for geometrically accurate radiance fields,

    B. Huang, Z. Yu, A. Chen, A. Geiger, and S. Gao, “2d gaussian splat- ting for geometrically accurate radiance fields,” in ACM SIGGRAPH 2024, Conference Proceedings, pp. 1–11

    work page 2024

Show all 34 references
  1. [9]

    Photo-slam: Real- time simultaneous localization and photorealistic mapping for monoc- ular stereo and rgb-d cameras,

    H. Huang, L. Li, H. Cheng, and S.-K. Yeung, “Photo-slam: Real- time simultaneous localization and photorealistic mapping for monoc- ular stereo and rgb-d cameras,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , Conference Proceedings, pp...

  2. [10]

    Di-fusion: Online implicit 3d reconstruction with deep priors,

    J. Huang, S. S. Huang, H. Song, and S. M. Hu, “Di-fusion: Online implicit 3d reconstruction with deep priors,” in 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , Conference Proceedings, pp. 8928–8937

    2021

  3. [11]

    Eslam: Efficient dense slam system based on hybrid representation of signed distance fields,

    M. M. Johari, C. Carta, and F. Fleuret, “Eslam: Efficient dense slam system based on hybrid representation of signed distance fields,” in 2023 IEEE/Cvf Conference on Computer Vision and Pattern Recogni- tion (CVPR) , Conference Proceedings, pp. 17 408–17 419

    2023

  4. [12]

    Nerf-vins: A real-time neural radiance field map- based visual-inertial navigation system,

    S. Katragadda, W. Lee, Y . Peng, P. Geneva, C. Chen, C. Guo, M. Li, and G. Huang, “Nerf-vins: A real-time neural radiance field map- based visual-inertial navigation system,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2024, pp. 10 230–10 237

    2024

  5. [13]

    Splatam: Splat track & map 3d gaussians for dense rgb-d slam,

    N. Keetha, J. Karhade, K. M. Jatavallabhula, G. Yang, S. Scherer, D. Ramanan, and J. Luiten, “Splatam: Splat track & map 3d gaussians for dense rgb-d slam,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Conference Proceedings, pp. 21 357–21 366

  6. [14]

    3d gaussian splatting for real-time radiance field rendering,

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, and G. Drettakis, “3d gaussian splatting for real-time radiance field rendering,” ACM Transactions on Graphics, vol. 42, no. 4, pp. 1–14, 2023

    2023

  7. [15]

    Aligning cyber space with physical world: A com- prehensive survey on embodied ai,

    Y . Liu, W. Chen, Y . Bai, J. Luo, X. Song, K. Jiang, Z. Li, G. Zhao, J. Lin, and G. Li, “Aligning cyber space with physical world: A com- prehensive survey on embodied ai,” arXiv preprint arXiv:2407.06886 , 2024

    2024 arXiv

  8. [16]

    Ngel-slam: Neural implicit representation-based global consistent low-latency slam system,

    Y . Mao, X. Yu, Z. Zhang, K. Wang, Y . Wang, R. Xiong, and Y . Liao, “Ngel-slam: Neural implicit representation-based global consistent low-latency slam system,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2024, pp. 6952–6958

    2024

  9. [17]

    Gaussian splatting slam,

    H. Matsuki, R. Murai, P. H. Kelly, and A. J. Davison, “Gaussian splatting slam,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , Conference Proceedings, pp. 18 039–18 048

  10. [18]

    Nerf: Representing scenes as neural radiance fields for view synthesis,

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoor- thi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,” Communications of the Acm , vol. 65, no. 1, pp. 99–106, 2022

    2022

  11. [19]

    Instant neural graphics primitives with a multiresolution hash encoding,

    T. M ¨uller, A. Evans, C. Schied, and A. Keller, “Instant neural graphics primitives with a multiresolution hash encoding,” ACM transactions on graphics (TOG) , vol. 41, no. 4, pp. 1–15, 2022

    2022

  12. [20]

    Orb-slam2: An open-source slam system for monocular, stereo, and rgb-d cameras,

    R. Mur-Artal and J. D. Tardos, “Orb-slam2: An open-source slam system for monocular, stereo, and rgb-d cameras,” IEEE Transactions on Robotics , vol. 33, no. 5, pp. 1255–1262, 2017

    2017

  13. [21]

    Kinectfusion: Real-time dense surface mapping and tracking,

    R. A. Newcombe, S. Izadi, O. Hilliges, D. Molyneaux, D. Kim, A. J. Davison, P. Kohi, J. Shotton, S. Hodges, and A. Fitzgibbon, “Kinectfusion: Real-time dense surface mapping and tracking,” in 2011 10th IEEE International Symposium on Mixed and Augmented Reality, Conference Pro...

    2011

  14. [22]

    Dtam: Dense tracking and mapping in real-time,

    R. A. Newcombe, S. J. Lovegrove, and A. J. Davison, “Dtam: Dense tracking and mapping in real-time,” in 2011 International Conference on Computer Vision , Conference Proceedings, pp. 2320–2327

    2011

  15. [23]

    Deep marching tetrahedra: a hybrid representation for high-resolution 3d shape syn- thesis,

    T. Shen, J. Gao, K. Yin, M.-Y . Liu, and S. Fidler, “Deep marching tetrahedra: a hybrid representation for high-resolution 3d shape syn- thesis,” Advances in Neural Information Processing Systems , vol. 34, pp. 6087–6101, 2021

    2021

  16. [24]

    The replica dataset: A digital replica of indoor spaces,

    J. Straub, T. Whelan, L. Ma, Y . Chen, E. Wijmans, S. Green, J. J. Engel, R. Mur-Artal, C. Ren, and S. Verma, “The replica dataset: A digital replica of indoor spaces,” arXiv preprint arXiv:1906.05797 , 2019

    1906 arXiv

  17. [25]

    A benchmark for the evaluation of rgb-d slam systems,

    J. Sturm, N. Engelhard, F. Endres, W. Burgard, and D. Cremers, “A benchmark for the evaluation of rgb-d slam systems,” 2012 IEEE/Rsj International Conference on Intelligent Robots and Systems (IROS) , pp. 573–580, 2012

    2012

  18. [26]

    Imap: Implicit mapping and positioning in real-time,

    E. Sucar, S. K. Liu, J. Ortiz, and A. J. Davison, “Imap: Implicit mapping and positioning in real-time,” 2021 IEEE/Cvf International Conference on Computer Vision (ICCV 2021) , pp. 6209–6218, 2021

    2021

  19. [27]

    Co-slam: Joint coordinate and sparse parametric encodings for neural real-time slam,

    H. Wang, J. Wang, and L. Agapito, “Co-slam: Joint coordinate and sparse parametric encodings for neural real-time slam,” in Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Conference Proceedings, pp. 13 293–13 302

  20. [28]

    Elasticfusion: Dense slam without a pose graph

    T. Whelan, S. Leutenegger, R. F. Salas-Moreno, B. Glocker, and A. J. Davison, “Elasticfusion: Dense slam without a pose graph.” in Robotics: science and systems , vol. 11. Rome, Italy, 2015, p. 3

    2015

  21. [29]

    Hero-slam: Hybrid enhanced robust optimization of neural slam,

    Z. Xin, Y . Yue, L. Zhang, and C. Wu, “Hero-slam: Hybrid enhanced robust optimization of neural slam,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2024, pp. 8610–8616

    2024

  22. [30]

    Gs- slam: Dense visual slam with 3d gaussian splatting,

    C. Yan, D. Qu, D. Xu, B. Zhao, Z. Wang, D. Wang, and X. Li, “Gs- slam: Dense visual slam with 3d gaussian splatting,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition, Conference Proceedings, pp. 19 595–19 604

  23. [31]

    Gaussian opacity fields: Efficient and compact surface reconstruction in unbounded scenes,

    Z. Yu, T. Sattler, and A. Geiger, “Gaussian opacity fields: Efficient and compact surface reconstruction in unbounded scenes,” arXiv preprint arXiv:2404.10772, 2024

    2024 arXiv

  24. [32]

    Rapid-mapping: Lidar- visual implicit neural representations for real-time dense mapping,

    H. Zhang, Y . Zou, Z. Yan, and H. Cheng, “Rapid-mapping: Lidar- visual implicit neural representations for real-time dense mapping,” IEEE Robotics and Automation Letters , 2024

    2024

  25. [33]

    Nice-slam: Neural implicit scalable encoding for slam,

    Z. Zhu, S. Peng, V . Larsson, W. Xu, H. Bao, Z. Cui, M. R. Oswald, and M. Pollefeys, “Nice-slam: Neural implicit scalable encoding for slam,” in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Conference Proceedings, pp. 12 776–12 786

    2022

  26. [34]

    State of the art on 3d reconstruction with rgb-d cameras,

    M. Zollh ¨ofer, P. Stotko, A. G¨orlitz, C. Theobalt, M. Niessner, R. Klein, and A. Kolb, “State of the art on 3d reconstruction with rgb-d cameras,” Computer Graphics F orum , vol. 37, no. 2, pp. 625–652, 2018

    2018

Pith tools

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