Pith. sign in

REVIEW 4 major objections 6 minor 30 references

EndoFlow-SLAM: Real-Time Endoscopic SLAM with Flow-Constrained Gaussian Splatting

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Adding optical flow as a geometric constraint to 3D Gaussian Splatting SLAM beats prior methods on static and dynamic endoscopic benchmarks.

desk verdict A plausible endoscopic 3DGS SLAM integration, but a table-number conflict in the reported RMSE undermines the headline depth claim until explained. read the letter →

arxiv 2506.21420 v2 pith:54VTXPCV submitted 2025-06-26 cs.CV cs.RO

Taoyu Wu , Yiyi Miao , Zhuoxiao Li , Haocheng Zhao , Kang Dang , Jionglong Su , Limin Yu , Haoang Li This is my paper
classification cs.CVcs.RO
keywords endoscopicSLAM3DGaussiansplattingopticalflowlossbundleadjustmentnovelviewsynthesisdepthregularizationdynamicscenescameraposeestimation
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

EndoFlow-SLAM is built on a single proposal: introduce optical flow as an explicit geometric constraint inside a 3D Gaussian Splatting SLAM loop, alongside the usual photometric and depth terms. The paper argues that in endoscopy, where surfaces are specular and tissue moves with breathing, appearance-only constraints are insufficient, and that a displacement field derived from re-projecting the Gaussian splats gives the optimizer a signal that ties camera pose to scene geometry. If the claim holds, real-time surgical SLAM can produce both accurate camera trajectories and high-quality novel views without relying on strong photometric consistency. The reported evaluations on static and dynamic surgical datasets show the flow-constrained system ahead of previous SLAM methods in rendering quality and pose accuracy.

What carries the argument

The carrying mechanism is the per-Gaussian flow. For each pixel covered by K overlapping Gaussians, the method projects each Gaussian's mean through the current pose at frame t and the sought pose at frame t+1, computing a displacement per Gaussian. Because the scene uses isotropic Gaussians, the covariance-dependent part of the flow expression reduces to the identity matrix, so the composite flow at a pixel is simply the alpha-weighted sum of the per-Gaussian center displacements. This composite flow is compared against an off-the-shelf optical-flow estimate in a squared-error loss, and that loss is folded into keyframe bundle adjustment together with photometric, scale-invariant depth, and depth-gradient terms.

What would settle it

Track a rigid textured phantom with controlled specular highlights where true optical flow is known from a fiducial grid; if the off-the-shelf flow errors on specular regions, the flow-constrained system should show higher ATE and lower PSNR than the same system with flow loss disabled.

Watch

Extended reading notes

Core claim

The paper's central claim is that a flow loss computed from the 3D Gaussian Splatting representation can serve as a geometric constraint that jointly improves camera tracking and scene reconstruction in endoscopic SLAM. Concretely, for each pixel the method renders the K overlapping Gaussians at frame t and re-projects them at frame t+1; because the Gaussians are isotropic, the resulting per-Gaussian flow simplifies to the displacement of Gaussian centers, and alpha-blending gives a composite flow field. The loss is the squared difference between this field and an off-the-shelf optical-flow estimate, and it is added to photometric, scale-invariant depth, and depth-gradient terms in keyframe bundle adjustment. On the C3VD static dataset the method records the best PSNR, SSIM, depth RMSE, and ATE among compared systems; on the StereoMIS dynamic dataset it records the best PSNR and ATE. The message is that geometry from flow, not appearance alone, is what makes splat-based SLAM robust in surgical conditions.

Load-bearing premise

The off-the-shelf optical flow is trusted as ground truth in the flow loss; if that flow is wrong on shiny, texture-poor endoscopic tissue, the geometric constraint would pull pose and structure optimization in the wrong direction.

Editorial extensions

If this is right

  • Real-time endoscopic SLAM does not have to choose between rendering quality and tracking accuracy: the flow constraint improves both in the reported tests.
  • The flow loss makes the system tolerate non-Lambertian surfaces and breathing motion better than appearance-only baselines, extending 3DGS SLAM to dynamic surgical scenes.
  • Focusing the refinement stage on keyframes with suboptimal rendering quality gives a better global scene without extra tracking cost.
  • Combining scale-invariant depth loss with depth-gradient regularization removes the scale ambiguity from monocular depth in splat-based endoscopic SLAM.

Reading between the lines

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

  • The paper leaves untested the consequence that the method's ceiling is set by the off-the-shelf flow estimator; a flow network trained on endoscopic tissue with specular highlights might push the same architecture further, but this is my inference, not the paper's result.
  • The flow-constrained bundle adjustment could transfer to other slowly deforming scenes, such as laparoscopy with tool-tissue contact, although the paper only demonstrates breathing motion.
  • A direct stress test would replace the flow loss with stereo-derived flow supervision, checking whether the geometric signal itself, rather than the specific estimator, is the real cause of the gains.
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

4 major / 6 minor

Summary. The paper presents EndoFlow-SLAM, a real-time 3D Gaussian Splatting SLAM system for endoscopic scenes. The method adds an optical-flow loss as a geometric constraint on camera pose and scene structure, combines depth-map normalization with a scale-invariant depth loss and gradient regularization to address monocular scale ambiguity, and introduces a two-stage global refinement strategy that prioritizes keyframes with suboptimal rendering quality. Experiments on the C3VD static dataset and the StereoMIS dynamic dataset compare against NICE-SLAM, Endo-Depth, ESLAM, and EndoGSLAM, reporting improved PSNR, SSIM, RMSE, and ATE. An ablation study examines the contributions of the depth loss, refinement strategy, and flow constraint.

Significance. If the reported results hold, the work is practically relevant: real-time dense reconstruction and tracking in endoscopic scenes with an explicit flow-based geometric constraint is a plausible step beyond photometric-only 3DGS SLAM. The paper is clearly organized, the chosen datasets match the static/dynamic claims, and the ablation isolates the proposed modules. However, the quantitative evidence as presented is not yet sufficient: the conflicting RMSE values for the same configuration, the lack of error bars or per-sequence results, and the unvalidated dependence on an external optical-flow estimator all undermine the headline performance claim. The formal derivation of the flow loss also contains a mathematical imprecision. These issues are fixable, but they must be addressed before the central claims can be accepted.

major comments (4)
  1. [Section 3.2, Table 1 vs Section 3.3, Table 3] Tables 1 and 3 report the same 'Ours' configuration on C3VD with identical PSNR (25.18), SSIM (0.82), LPIPS (0.27), and ATE (0.23), but the RMSE differs: 1.54 mm in Table 1 and 2.04 mm in Table 3. The paper gives no explanation for this discrepancy. If the Table 3 value is the correct one, EndoFlow-SLAM's C3VD RMSE (2.04 mm) is worse than NICE-SLAM's 1.88 mm reported in Table 1, which directly contradicts the abstract's claim of state-of-the-art depth performance on static scenes. The authors must clarify which number is correct, report per-sequence results for the ten C3VD sequences, and specify the evaluation protocol (depth range, masks, and alignment procedure) so the discrepancy can be resolved.
  2. [Section 2.3, Eq. (3)] The flow loss is defined against optical flow obtained from an unnamed off-the-shelf estimator used as ground truth. On shiny, texture-poor endoscopic tissue, optical-flow estimates can be unreliable, and since this flow constraint is the paper's main claimed advantage over photometric-only SLAM, the assumption is load-bearing and untested. The authors should name the estimator, report its accuracy on C3VD and StereoMIS, and provide a sensitivity analysis (for example, replacing or perturbing the flow estimates) to show that the method is robust to flow errors.
  3. [Section 2.3, Gaussian flow derivation] The claim that isotropic Gaussians make the Cholesky factorization of the covariance matrices simplify to the identity is mathematically imprecise: for an isotropic covariance Σ = σ²I, the Cholesky factor is σI, not I. The conclusion f^G_i = μ_{i,t+1} − μ_{i,t} therefore does not follow as stated. The authors need to provide a correct derivation of the per-Gaussian displacement under the pose change, or explicitly state the additional assumptions (for example, unit scale or a purely translational projected motion) that justify the simplification.
  4. [Section 3.2 and Section 3.3] All quantitative claims rest on a single run with no error bars and no per-sequence breakdown across the ten C3VD scenes or the StereoMIS segments. This makes it impossible to judge whether the reported differences over EndoGSLAM (ATE 0.34 vs 0.23 mm; RMSE 2.17 vs 1.54 mm) are statistically meaningful. Please report per-sequence numbers, multiple seeds, and standard deviations, or otherwise justify the stability of the comparisons.
minor comments (6)
  1. [Section 3.2, Table 1] The text says the comparison includes ESLAM, but Table 1 does not list ESLAM; ESLAM appears only in Table 2 on StereoMIS. Add the missing row on C3VD or amend the text.
  2. [Section 2.2, Eq. (2)] The loss weights λ1, λ2, λ3, λ4 and the depth-gradient weights w_h, w_v are introduced but no values are given. A hyperparameter table and the fixed iteration counts would substantially improve reproducibility.
  3. [Section 2.3, notation] The off-the-shelf optical flow is written as f^G_t(x), which is the same symbol family used for Gaussian flow. Use a distinct symbol, e.g., f^opt_t, to avoid notational confusion.
  4. [Figure 1 and Figure 2 captions] The 'Gaussian Prune' box in Fig. 1 is not explained in the text, and the Fig. 2 caption says 'GaussianFlow estimation' even though the paper proposes a flow-constrained SLAM module rather than GaussianFlow itself.
  5. [Throughout] There are typos and inconsistent naming: 'T able 1' and 'T able 3' in the table captions, 'EndoFLow-SLAM' vs 'EndoFlow-SLAM' in the conclusion, and 'the the pixel position' in Section 2.3.
  6. [Section 3.1, datasets] The paper says it follows [26] and evaluates on 10 C3VD sequences but does not list which sequences are used. Please specify the sequence names for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: flow and depth losses are external supervisions, not re-labeled predictions; self-citations are contextual.

full rationale

EndoFlow-SLAM's derivation chain is supervised-learning style rather than self-referential. The optical flow loss (Eq. 3) compares rendered Gaussian flow to an external off-the-shelf flow estimate; that estimate is not produced by the paper's own model and is not identical to any reported evaluation metric (PSNR, SSIM, LPIPS, RMSE, or ATE). The depth regularization (Eq. 2, Lreg_depth) uses ground-truth depth gradients, but this is standard RGB-D SLAM supervision shared with the compared baselines, and the reported RMSE is a downstream reconstruction accuracy measure rather than the optimized loss itself. Citations to the authors' prior work (e.g., PG-SLAM [13], DynSup [16], ULSR-GS [17]) are contextual design references, not load-bearing proofs or uniqueness theorems. The Table 1 vs Table 3 C3VD RMSE discrepancy (1.54 vs 2.04 mm for 'Ours') is an internal-consistency and correctness concern, not a circularity: even if one value is wrong, the claim does not reduce to its input by construction. The main untested assumption, reliability of the off-the-shelf flow estimator on endoscopic tissue, is an empirical robustness risk rather than a circular derivation. No step equates a fitted parameter with a predicted output, and no central premise is justified only by a self-citation.

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

The central claim rests on hand-tuned loss weights and strong assumptions about optical-flow reliability, the isotropic Gaussian model, and the MonoGS/3DGS base. No new physical or mathematical entities are introduced. The count of free parameters is moderate for a systems paper, but the failure to report their values and the unverified flow target are the main burdens.

free parameters (6)
  • lambda_1 (photometric loss weight) = not reported
    Balances the RGB reprojection loss in Eqs (2)-(4); value chosen by the authors but not stated.
  • lambda_2 (depth gradient regularization weight) = not reported
    Balances Lreg_depth in Eqs (2)-(3); not stated.
  • lambda_3 (scale-invariant depth loss weight) = not reported
    Balances Lscale in Eqs (2)-(3); not stated.
  • lambda_4 (flow loss weight) = not reported
    Balances Lflow, the paper's key new constraint; not stated.
  • lambda_dssim (SSIM weight in Eq 4) = not reported
    Controls the relative weight of SSIM in the refinement loss; not stated.
  • w_h and w_v (depth gradient weights) = not reported
    Weights for horizontal and vertical depth gradients in Lreg_depth; not stated.
assumptions (4)
  • domain assumption Off-the-shelf optical flow is a reliable geometric target in endoscopic scenes.
    Section 2.3 defines Lflow as the difference between Gaussian flow and this external flow, using it to optimize poses and Gaussians. The paper does not measure flow-estimation error on its datasets.
  • ad hoc to paper For isotropic Gaussians, the Cholesky factorization of the covariance simplifies to the identity, so Gaussian flow equals mean displacement.
    Section 2.3, paragraph 'Simultaneous Optimization by Flow Constrain'. For Sigma = sigma^2 I the Cholesky factor is sigma I, not I, so the simplification as written is not mathematically accurate; it appears to assume unit-variance normalization.
  • standard math The MonoGS keyframe management, 3DGS rasterizer, and Adam-based optimization are reliable building blocks that need no re-justification.
    The paper builds directly on MonoGS [20] and 3DGS [10], citing them as established; the correctness of that base is assumed.
  • domain assumption Depth maps in C3VD are ground-truth and usable for scale-invariant and gradient losses.
    Section 2.2 uses this depth as supervision in Lscale and Lreg_depth; the paper does not analyze depth noise.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EndoFlow-SLAM: Real-Time Endoscopic SLAM with Flow-Constrained Gaussian Splatting." pith.science (2026). https://pith.science/paper/54VTXPCV

@misc{pith2026250621420,
  author       = {Pith},
  title        = {Pith review of: EndoFlow-SLAM: Real-Time Endoscopic SLAM with Flow-Constrained Gaussian Splatting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/54VTXPCV}},
  note         = {Machine review of arXiv:2506.21420}
}
read the original abstract

Efficient three-dimensional reconstruction and real-time visualization are critical in surgical scenarios such as endoscopy. In recent years, 3D Gaussian Splatting (3DGS) has demonstrated remarkable performance in efficient 3D reconstruction and rendering. Most 3DGS-based Simultaneous Localization and Mapping (SLAM) methods only rely on the appearance constraints for optimizing both 3DGS and camera poses. However, in endoscopic scenarios, the challenges include photometric inconsistencies caused by non-Lambertian surfaces and dynamic motion from breathing affects the performance of SLAM systems. To address these issues, we additionally introduce optical flow loss as a geometric constraint, which effectively constrains both the 3D structure of the scene and the camera motion. Furthermore, we propose a depth regularisation strategy to mitigate the problem of photometric inconsistencies and ensure the validity of 3DGS depth rendering in endoscopic scenes. In addition, to improve scene representation in the SLAM system, we improve the 3DGS refinement strategy by focusing on viewpoints corresponding to Keyframes with suboptimal rendering quality frames, achieving better rendering results. Extensive experiments on the C3VD static dataset and the StereoMIS dynamic dataset demonstrate that our method outperforms existing state-of-the-art methods in novel view synthesis and pose estimation, exhibiting high performance in both static and dynamic surgical scenes.

Figures

Figures reproduced from arXiv: 2506.21420 by the authors.

Figure 1
Figure 1. Method overview. Given the first RGB-D image, we initialize the 3D Gaussians and subsequently perform camera tracking and mapping iteratively. Tracking handles non-keyframes with depth regularization for scale consistency (optimizing camera pose only), while keyframes incorporate optical flow as a geometric constraint in Local BA to simultaneously optimize poses and 3DGS primitives. Mapping applies BA with optical f… view at source ↗
Figure 2
Figure 2. GaussianFlow estimation. At time t, each pixel xt results from K overlap￾ping Gaussians. At time t + 1, each K Gaussian will have a corresponding Gaussian flow (Black arrow). By accumulating these Gaussian flows, we obtain the overall Gaus￾sian flow. Our goal is to minimize the difference between GS fLow and Optical flow by optimizing both the camera pose Tˆt+1 and 3DGS primitive Gˆ. matrix Σi,t+1 for the i-th Gauss… view at source ↗
Figure 3
Figure 3. Qualitative results on C3VD and StereoMIS Dataset. 3.3 Ablation Study [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

30 extracted references · 19 canonical work pages

  1. [1]

    Medical Image Analysis p

    Bobrow, T.L., Golhar, M., Vijayan, R., Akshintala, V.S., Garcia, J.R., Durr, N.J.: Colonoscopy 3d video dataset with paired depth from 2d-3d registration. Medical Image Analysis p. 102956 (2023)

    2023

  2. [2]

    IEEE Transactions on Robotics37(6), 1874–1890 (2021)

    Campos, C., Elvira, R., Rodríguez, J.J.G., Montiel, J.M., Tardós, J.D.: Orb-slam3: An accurate open-source library for visual, visual–inertial, and multimap slam. IEEE Transactions on Robotics37(6), 1874–1890 (2021)

    2021

  3. [3]

    Computer methods and programs in biomedicine158, 135–146 (2018)

    Chen, L., Tang, W., John, N.W., Wan, T.R., Zhang, J.J.: Slam-based dense sur- face reconstruction in monocular minimally invasive surgery and its application to augmented reality. Computer methods and programs in biomedicine158, 135–146 (2018)

    work page 2018

  4. [4]

    arXiv preprint arXiv:2403.12365 (2024)

    Gao,Q.,Xu,Q.,Cao,Z.,Mildenhall,B.,Ma,W.,Chen,L.,Tang,D.,Neumann,U.: Gaussianflow: Splatting gaussian dynamics for 4d content creation. arXiv preprint arXiv:2403.12365 (2024)

    arXiv 2024

  5. [5]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Guo, J., Wang, J., Kang, D., Dong, W., Wang, W., Liu, Y.h.: Free-surgs: Sfm-free 3d gaussian splatting for surgical scene reconstruction. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 350–360. Springer (2024)

    work page 2024

  6. [6]

    International journal of computer assisted radiology and surgery18(7), 1185–1192 (2023)

    Hayoz, M., Hahne, C., Gallardo, M., Candinas, D., Kurmann, T., Allan, M., Sznit- man, R.: Learning how to robustly estimate camera pose in endoscopic videos. International journal of computer assisted radiology and surgery18(7), 1185–1192 (2023)

    2023

  7. [7]

    Wiley interdisciplinary reviews: computa- tional statistics 1(2), 251–254 (2009)

    Higham, N.J.: Cholesky factorization. Wiley interdisciplinary reviews: computa- tional statistics 1(2), 251–254 (2009)

    work page 2009

  8. [8]

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

    Johari, M.M., Carta, C., Fleuret, F.: Eslam: Efficient dense slam system based on hybrid representation of signed distance fields. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 17408–17419 (2023)

    work page 2023

Show all 30 references
  1. [9]

    Keetha, N., Karhade, J., Jatavallabhula, K.M., Yang, G., Scherer, S., Ramanan, D., Luiten, J.: Splatam: Splat track and map 3d gaussians for dense rgb-d slam. 10 T. Wu et al. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 21357–21366 (2024)

    2024

  2. [10]

    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

  3. [11]

    arXiv preprint arXiv:1412.6980 (2014)

    Kingma, D.P.: Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014)

    2014 arXiv

  4. [12]

    IEEE Robotics and Automation Letters7(4), 11450–11457 (2022)

    Lamarca, J., Rodríguez, J.J.G., Tardós, J.D., Montiel, J.M.: Direct and sparse deformable tracking. IEEE Robotics and Automation Letters7(4), 11450–11457 (2022)

    2022

  5. [13]

    arXiv preprint arXiv:2411.15800 (2024)

    Li, H., Meng, X., Zuo, X., Liu, Z., Wang, H., Cremers, D.: Pg-slam: Photo- realistic and geometry-aware rgb-d slam in dynamic environments. arXiv preprint arXiv:2411.15800 (2024)

    2024 arXiv

  6. [14]

    In: 2018 IEEE International Conference on Robotics and Automation (ICRA)

    Li, H., Yao, J., Bazin, J.C., Lu, X., Xing, Y., Liu, K.: A monocular slam system leveraging structural regularity in manhattan world. In: 2018 IEEE International Conference on Robotics and Automation (ICRA). pp. 2518–2525. IEEE (2018)

    2018

  7. [15]

    IEEE Transactions on Pattern Analysis and Machine Intelligence45(11), 13035–13053 (2023)

    Li, H., Zhao, J., Bazin, J.C., Kim, P., Joo, K., Zhao, Z., Liu, Y.H.: Hong kong world: Leveraging structural regularity for line-based slam. IEEE Transactions on Pattern Analysis and Machine Intelligence45(11), 13035–13053 (2023)

    2023

  8. [16]

    arXiv preprint arXiv:2412.00851 (2024)

    Li, W., Chen, W., Qian, S., Chen, J., Cremers, D., Li, H.: Dynsup: Dynamic gaussian splatting from an unposed image pair. arXiv preprint arXiv:2412.00851 (2024)

    2024 arXiv

  9. [17]

    arXiv preprint arXiv:2412.01402 (2024)

    Li, Z., Yao, S., Wu, T., Yue, Y., Zhao, W., Qin, R., Garcia-Fernandez, A.F., Levers, A.,Zhu,X.:Ulsr-gs:Ultralarge-scalesurfacereconstructiongaussiansplattingwith multi-view geometric consistency. arXiv preprint arXiv:2412.01402 (2024)

    2024 arXiv

  10. [18]

    In: 2022 International conference on robotics and automation (ICRA)

    Liu, X., Li, Z., Ishii, M., Hager, G.D., Taylor, R.H., Unberath, M.: Sage: slam with appearance and geometry prior for endoscopy. In: 2022 International conference on robotics and automation (ICRA). pp. 5587–5593. IEEE (2022)

    2022

  11. [19]

    British journal of neurosurgery 28(5), 606–610 (2014)

    Marcus, H.J., Cundy, T.P., Hughes-Hallett, A., Yang, G.Z., Darzi, A., Nandi, D.: Endoscopic and keyhole endoscope-assisted neurosurgical approaches: a qualita- tive survey on technical challenges and technological solutions. British journal of neurosurgery 28(5), 606–610 (2014)

    2014

  12. [20]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition

    Matsuki, H., Murai, R., Kelly, P.H., Davison, A.J.: Gaussian splatting slam. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition. pp. 18039–18048 (2024)

    2024

  13. [21]

    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

  14. [22]

    IEEE transactions on pattern analysis and machine intelligence44(3), 1623–1637 (2020)

    Ranftl, R., Lasinger, K., Hafner, D., Schindler, K., Koltun, V.: Towards robust monocular depth estimation: Mixing datasets for zero-shot cross-dataset transfer. IEEE transactions on pattern analysis and machine intelligence44(3), 1623–1637 (2020)

    2020

  15. [23]

    IEEE Robotics and Automation Letters6(4), 7225– 7232 (2021)

    Recasens, D., Lamarca, J., Fácil, J.M., Montiel, J., Civera, J.: Endo-depth-and- motion: Reconstruction and tracking in endoscopic videos using depth networks and photometric constraints. IEEE Robotics and Automation Letters6(4), 7225– 7232 (2021)

    2021

  16. [24]

    Shan, J., Li, Y., Xie, T., Wang, H.: Enerf-slam: a dense endoscopic slam with neural implicitrepresentation.IEEETransactionsonMedicalRoboticsandBionics(2024)

    2024

  17. [25]

    In: 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

    Shan, J., Li, Y., Yang, L., Feng, Q., Han, L., Wang, H.: Dds-slam: Dense semantic neural slam for deformable endoscopic scenes. In: 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). pp. 10837–10842. IEEE (2024) EndoFlow-SLAM 11

    2024

  18. [26]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Wang, K., Yang, C., Wang, Y., Li, S., Wang, Y., Dou, Q., Yang, X., Shen, W.: En- dogslam: Real-time dense reconstruction and tracking in endoscopic surgeries using gaussian splatting. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. p...

    2024

  19. [27]

    In: MICCAI

    Wang, Y., Long, Y., Fan, S.H., Dou, Q.: Neural rendering for stereo 3d reconstruc- tion of deformable tissues in robotic surgery. In: MICCAI. pp. 431–441. Springer (2022)

    2022

  20. [28]

    In: 2024 International Symposium on Power Electronics, Electrical Drives, Au- tomation and Motion (SPEEDAM)

    Wu, T., Zhang, Y., Zhao, H., Yue, Y., Yu, L., Wang, X.: Enhancing automated guided vehicle navigation with multi-sensor fusion and algorithmic optimization. In: 2024 International Symposium on Power Electronics, Electrical Drives, Au- tomation and Motion (SPEEDAM). pp. 557–562...

    2024

  21. [29]

    Advances in Neural Information Processing Systems37, 21875–21911 (2025)

    Yang, L., Kang, B., Huang, Z., Zhao, Z., Xu, X., Feng, J., Zhao, H.: Depth anything v2. Advances in Neural Information Processing Systems37, 21875–21911 (2025)

    2025

  22. [30]

    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 August 6, 2026 · model on record in the stance chip above.