Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Gaussian splatting is the current state-of-the-art for dense, deformable 3D anatomy reconstruction in robot-assisted minimally invasive surgery (RAMIS); however, most pipelines are offline and depend on accurate camera trajectory priors (often from robotic kinematics), limiting applicability when priors are missing or noisy. To address these limitations, we propose Track2Map, an online 3D Gaussian Splatting pipeline that jointly optimizes camera trajectory and 3D deformable scene representation directly from surgical video. Track2Map is therefore capable of robust 3D reconstructions when camera trajectory priors are either absent or noisy, and due to its online nature it effectively works as a Simultaneous Localisation and Mapping (SLAM) method. To stabilize optimization in the presence of tissue motion and ambiguous visual cues, we introduce a track-anchored deformation initialization using dense 2D point tracks. Track statistics are further utilized to disentangle camera motion from scene deformation by detecting static camera periods and reducing drift during incremental mapping. Experiments on StereoMIS show improved reconstruction quality and camera trajectory against competing SLAM methods, as well as compared to non-SLAM methods that utilize camera trajectory priors. The code is available at https://track2map.github.io/.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2834_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: https://papers.miccai.org/miccai-2026/supp/2834_supp.zip

Link to the Code Repository

https://track2map.github.io/

Link to the Dataset(s)

https://zenodo.org/records/10867949

BibTex

@InProceedings{SonTia_Track2Map_MICCAI2026,
        author = { Song, Tianyi AND Bonilla, Sierra AND Ju, Xinwei AND Mazomenos, Evangelos B. AND Stoyanov, Danail AND Schmidt, Adam AND Mohareri, Omid AND Bano, Sophia AND Vasconcelos, Francisco},
        title = { { Track2Map: Online Deformable SLAM with Motion-Aware Pose Optimization in Robotic Surgery } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16893},
        month = {September},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    This paper presents an online 3D Gaussian Splatting pipeline for the simultaneous estimation of camera trajectory and the reconstruction of a non-rigid 3D scene representation directly from video data. The authors introduce a track-anchored deformation initialization strategy that leverages dense 2D point tracks, which constitute a strong prior in endoscopic scenarios, particularly in the presence of large deformations. The manuscript includes experimental results and relevant comparative evaluations, which demonstrate promising performance.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    Challenging problem and relevant for the community. Promising results and good comparison with other techniques. Simple yet effective model.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    Lack of related works. Lack of important details and values to guarantee reproducibility.

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html

    The authors state that their work is motivated by the need for deformable reconstruction systems that can exploit pose priors when available, remain robust when such priors are corrupted, and operate in their absence. However, this concept was previously introduced in Sequential Non-Rigid Structure from Motion Using Physical Priors. Similarly, the claim that “recent work has begun to address this gap by jointly optimizing camera motion and deformable scene structure in online settings” is not entirely accurate, as the aforementioned work (and many others) already explored joint estimation in an online framework. As a result, the manuscript does not adequately position itself within the existing literature, which weakens the motivation for the proposed approach. While the contribution itself may still be valuable and technically distinct, the overarching idea is not entirely novel, and this should be clarified in the introduction. It is also worth noting that the proposed method leverages stereo RGB frames and depth maps, rather than relying solely on monocular input. This additional information and stronger priors simplify the problem setting compared to earlier works.

    In general, the paper is clearly written and presents the problem in a structured manner. However, the discussion of related work is incomplete and, at times, misleading, as it overlooks key contributions that address similar claims. The authors also refer to the distinction between rigid and non-rigid motion, which is particularly relevant, given that a similar principle is employed in the previous reference, to update the camera parameters.

    Regarding the presentation, some elements in Figure 2A are unclear; including explicit references would improve readability. Additionally, Figures 2 and 3 contain a substantial amount of information but are presented at a scale that limits their interpretability. Enlarging these figures would enhance clarity.

    The qualitative comparison in Figure 3 demonstrates competitive performance relative to methods [6,24], while the quantitative evaluation provides stronger evidence of accuracy.

    Finally, the visual comparison presented in Figure 4 is difficult to interpret. Providing additional viewpoints to better illustrate the reconstructed 3D geometry, rather than relying solely on the camera perspective, would significantly improve the evaluation. Ideally, both camera-centered and external viewpoints should be included.

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The authors propose a method for reconstructing non-rigid bodies, arguing that the joint estimation of camera pose, non-rigid deformation, and geometry is required. However, the manuscript does not sufficiently consider the relevant body of literature to support this claim. While this statement is presented as a novel contribution, several prior works have already addressed related challenges but are not discussed. Notably, Total Estimation from RGB Video: On-line Camera Self-Calibration, Non-Rigid Shape and Motion and Active SLAM in 3D Deformable Environments are absent from the discussion, to name a few.

    Furthermore, the authors predominantly cite methods developed for rigid scenarios, such as [19] for Structure-from-Motion (SfM) and [15] for Neural Radiance Fields (NeRF). To provide a more comprehensive context, it would be beneficial to include non-rigid approaches, particularly those applied to endoscopic data. Relevant examples include Neural Rendering for Stereo 3D Reconstruction of Deformable Tissues in Robotic Surgery, EndoSurf: Neural Surface Reconstruction of Deformable Tissues with Stereo Endoscope Videos, and NeRFscopy: Neural Radiance Fields for In Vivo Time-Varying Tissues from Endoscopy.

    The manuscript states that “when the camera moves, tracked points exhibit a coherent global displacement field; when the camera is stationary, observed motion is dominated by heterogeneous tissue deformation and tool interaction.” However, the more general and realistic scenario, i.e., where both camera motion and non-rigid deformation occur simultaneously, is not discussed and should be addressed.

    Initialization remains a critical challenge, particularly in non-rigid reconstruction settings, where many approaches struggle to obtain a reliable initial solution. This aspect is not sufficiently clarified in the paper. Nevertheless, the use of stereo data and depth maps provides a significant advantage, as it increases the amount of available information for initialization.

    The reproducibility of the work is limited, as key implementation details, such as the weights in the loss function and certain coefficients described in Section 2, were not reported.

    The relation between the input data modalities requires further clarification. Initially, the manuscript suggests that the method operates solely on video input; however, subsequent sections refer to the use of stereo RGB images and even depth maps. This inconsistency may lead to confusion and should be explicitly addressed.

  • Reviewer confidence

    Very confident (4)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A



Review #2

  • Please describe the contribution of the paper

    This paper presents Track2Map, an online deformable SLAM framework for robot-assisted minimally invasive surgery that jointly estimates camera pose and deformable 3D scene structure using 3D Gaussian Splatting (3DGS). Unlike most recent surgical reconstruction pipelines, the method does not require accurate camera pose priors and remains robust under noisy or absent pose initialization. The core idea is to leverage dense 2D point tracking to (1) gate pose optimization during camera-static periods, avoiding drift caused by tissue deformation, and (2) initialize and constrain a sparse deformation model, which is used to compute the 3DGS representation. Stereo depth is used to provide metric grounding when available. The method is evaluated on StereoMIS across clean, noisy, and no-pose groups, showing substantial improvements over SLAM baselines and prior deformable 3DGS methods in both reconstruction quality and camera trajectory accuracy.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    The positive aspects of the paper are:

    • Paper is in general well written and simple to follow
    • Addresses a relevant subject for the MICCAI community and a critical limitation of current surgical reconstruction pipelines, which is the reliance on accurate camera pose priors
    • The system-level design composed of motion-aware pose gating, track-guided deformation initialization, etc. make sense and is not straightforward
    • the individual components of the system are not new, but their integration and procedural use in this context seems to be novel
    • The experimental results and comparison with SOTA approaches are convincing, showing improvements in reconstruction quality and pose accuracy
  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    The negative aspects of the paper are:

    • Limited dataset diversity: all reconstruction and SLAM evaluations are performed only on the StereoMIS dataset
    • High dependence on 2D tracking quality (pose gating, pose init, deformation init), and robustness under tracking degradation is not explicitly evaluated
    • motion gating thresholds is empirically selected and fixed
    • the current implementation runs at approximately 6 seconds per frame on an NVIDIA V100 GPU, which is far from meeting the latency requirements of a realistic OR scenario. Although this limitation is explicitly acknowledged by the authors, it significantly weakens the practical interpretation of the method as “online”, especially when compared to existing real-time or near–real-time SLAM systems in surgical environments. As such, in its current form, the system appears more suited for offline analysis, and the paper’s positioning as a surgical-targeted solution should be interpreted accordingly.
  • Please rate the clarity and organization of this paper

    Satisfactory

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission has provided an anonymized link to the source code, dataset, or any other dependencies.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
    • In Section 1, the first paragraph introduces RAMIS and its existing challenges, but does so without citing any prior work. Providing a few references at this stage would help ground these statements, contextualize the difficulties within the existing literature, and better orient the reader with respect to the current SOTA.
    • I would appreciate clarification regarding the statement on page 2: “In RAMIS, camera pose is often calculated from robot joint kinematics, which can be inaccurate due to calibration errors and measurement noise, and may also not always be available in a clinical setting.” From my understanding, medical robotic systems can be calibrated with very high accuracy (often reported below 0.01 mm), and kinematic errors are not always a dominant source of pose uncertainty. It would be helpful if the authors could better justify this claim, either by citing relevant literature or clarifying the specific scenarios in which kinematic pose becomes unreliable (e.g., hand-eye calibration).
    • A relevant SOTA paper regarding motion disentanglement should be cited: “Easi3R: Estimating Disentangled Motion from DUSt3R Without Training”, ICCV, 2025
    • In Section 2.1, the motion-gating strategy appears to rely exclusively on flow direction, without considering the flow magnitude. As a result, small-magnitude flow, potentially dominated by noise, contribute equally to large, reliable motions. This could cause problems in the optimization strategy. Could the authors comment on this design choice, and whether incorporating magnitude-based weighting or confidence-driven filtering was considered?
  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    This paper presents a well-motivated and well written paper to online deformable reconstruction in RAMIS, addressing an important limitation of recent surgical reconstruction pipelines, which is their reliance on accurate camera pose priors. The proposed system combines motion-aware pose gating, track-guided deformation initialization, and joint optimization within a 3DGS framework. While the individual components are not novel, their integration is non-trivial. The experimental results on StereoMIS are convincing, with clear improvements over SOTA baselines in both reconstruction quality and pose accuracy. However, despite these strengths, I lean toward a weak reject recommendation due to: (1) all reconstruction and SLAM evaluations are conducted on a single dataset (StereoMIS), raising concerns about robustness/generalization; (2) the method relies heavily on dense 2D tracking for pose gating, pose initialization, and deformation initialization, yet robustness under degraded tracking conditions is not explicitly evaluated, (3) motion-gating strategy depends on empirically, fixed thresholds, and only flow direction is considered, without accounting for flow magnitude, which may make the approach sensitive to noise-dominated motion, and (4) from a practical standpoint, the current implementation runs at approximately 6 seconds per frame on an NVIDIA V100 GPU, which is far from meeting the latency requirements of OR scenarios (although this limitation is acknowledged by the authors). Nevertheless, if the authors are able to address my concerns (1) demonstrating or commenting on robustness across additional datasets or surgical scenarios, (2) comment and/or providing an additional analysis of sensitivity to tracking quality and motion‑gating parameters, (3) clarifying and better supporting claims regarding the limitations of kinematic camera pose in RAMIS, and (4) more carefully positioning the method with respect to practical OR constraints, I would be willing to reconsider my rating.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    The authors addressed several of my main concerns. The clarification of the paper’s contribution (integration of deformable SLAM with 3DGS in a surgical context) resolves my concern about novelty. The discussion on dataset limitations is reasonable, and the explanation of why StereoMIS is uniquely suited for this evaluation is acceptable. The clarifications regarding kinematic pose accuracy, motion gating, and the distinction between “online” and “real-time” improve the positioning of the paper. Some limitations remain, including reliance on tracking quality, heuristic motion gating, single-dataset evaluation, and runtime constraints. However, the authors confirm that these will now be better acknowledged in the paper. Overall, I believe the work represents a relevant system contribution to the MICCAI community, and I am upgrading my score to accept.



Review #3

  • Please describe the contribution of the paper

    This paper presents Track2Map, an online deformable SLAM and 3D reconstruction method for robotic surgery videos. It jointly estimates camera pose and deformable anatomy, and is designed to work when pose priors are clean, noisy, or missing. The main idea is to use dense 2D tracks both to initialize deformation and to decide when pose updates should be made. Experiments on StereoMIS show improved reconstruction and pose estimation over the compared baselines.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    The paper addresses an important problem in robotic surgery, where accurate pose information is often unavailable. The idea of motion-aware pose updates is interesting and practically relevant. The experimental results on StereoMIS show better 3D reconstruction and pose estimation over baselines.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    The main weakness is clarity:

    • We don’t know in the abstract whether the method is based on monocular or stereo endoscope.
    • Same comment in the first paragraphs of the introduction
    • “Second, we use dense 2D point tracking to disentangle camera motion from tissue deformation.” in the introduction: the key idea on how this is achieved is not introduced.
    • In the method: notations not defined for the scene represented as a set of Gaussains (their centers appear to be x in Section 2.2), and for the sparse set of control points (?). This makes the subsequent section hard to follow.
    • The Section 2.2 is particularly hard to follow: in “which treats control-point offsets as free variables optimized per frame, we explicitly drive the deformation field using tracked anchor motion.”, what is the conceptual difference between “control-point offsets” and “tracked anchor motion”? In “its N nearest anchors”: are these anchors the control points?

    Additional justifications of the design of the “Motion-Aware Camera Update” module would be welcome:

    • Why not using standard tools of epipolar geometry to detect dominant rigid motion (eg RANSAC + F or H estimation)?
    • Why not using depth of the 2D correspondences at hand that provide additional information?
    • If the endoscope is moving along its optical axis in a cavity, the optical flow should in general present a high circular standard deviation, and therefore associated to the “scene motion” mode which is opposed to the expected behavior.

    Missing references:

    • EndoGaussian paper, and its follow up TMI paper “Foundation Model-Guided Gaussian Splatting for 4D Reconstruction of Deformable Tissues”.
    • EndoNerf not cited (Neural Rendering for Stereo 3D Reconstruction of Deformable Tissues in Robotic Surgery)
  • Please rate the clarity and organization of this paper

    Poor

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission has provided an anonymized link to the source code, dataset, or any other dependencies.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html

    Typos:

    • in the abstract: “optimsation”
    • “on to a valid”
  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (2) Reject — should be rejected, independent of rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    I gave a reject because, although the paper addresses an important problem and reports promising results, the presentation is too unclear to me in its current form. The method section is difficult to follow, key concepts and notations are not defined clearly enough.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Reject

  • [Post rebuttal] Please justify your final decision from above.

    The rebuttal addresses some of my concerns, in particular the clarification regarding input modalities and the distinction between control points and anchors. The missing references can also be addressed.

    However, the assumption that camera motion and tissue deformation can be separated based on the da Vinci camera/tool locking mechanism appears to rely on strong prior knowledge of a specific robotic platform. If the method depends on this coupling between endoscope motion and tool motion, this should be made explicit and substantially discussed, as it limits the generality of the approach.

    Finally, the specific case I raised regarding camera motion along the optical axis is acknowledged as a limitation but not addressed.

    Overall, despite the clarifications provided in the rebuttal, I maintain my recommendation to reject the paper.



Author Feedback

[R1/R2/R3] Novelty/related work We agree deformable SLAM is not new. Our contribution is adopting it for 3DGS dense reconstruction of surgery scenes using priors specific to this context. We will add NRSfM/SLAM references and clarify the contribution. [R1/R3] Input modalities The proposed method uses stereo RGB. Depth is estimated from FoundationStereo. Pose input is optional. We will consistently refer to this as “stereo RGB video with optional pose prior”. In Table 1, we test a monocular version using the left RGB and depth from DepthAnything-V2.[R1] Reproducibility Implementation details were provided in the linked GitHub repo. We will add key values(e.g. w_depth=10, w_color=5, w_pose=0.25) to the paper [R2] Dataset diversity We agree that more datasets would be welcome. To our knowledge, StereoMIS is the only public robotic-surgery benchmark supporting Track2Map full evaluation: calibrated stereo, deformable tissue and tool interaction, metric depth, and camera trajectories for ATE/RPE evaluation. Other benchmarks lack requirements: SCARED is rigid, EndoNeRF/STIR/SurgT lacks poses. We will clarify this limitation.

[R2] Robustness to tracking We tested perturbing 2D tracks with pixel-level Gaussian noise, leading to following changes (σ=1,2,4px): σ px PSNR SSIM LPIPS ATE m RPEt m RPEr deg 1 27.4 0.729 0.319 0.0276 0.0029 0.327 2 27.3 0.726 0.319 0.0288 0.0031 0.338 4 27.2 0.720 0.327 0.0290 0.0033 0.352 This suggests moderate robustness to track noise. If permitted, we will include this in the paper.

[R3] Control-point vs. anchor Control-point offsets are learnable 3D deformation variables, while anchors provide track-derived 3D displacement measurements from 2D tracks lifted with stereo depth. Previous work[8] estimated them from photometric loss; in our work, control points are optimised with additional supervision of anchors. “N nearest anchors” refers to the spatially closest anchors used to interpolate displacement at each control point location. [R2] Kinematic pose reliability Camera kinematics have low relative pose errors, but large absolute pose errors. This is relevant in SLAM where camera and anatomy are in the same coordinate system. Hand-eye calibration error is a large contributor. This is documented e.g. in Ferguson, Comparing the Accuracy of the da Vinci Xi and da Vinci Si for Image Guidance and Automation, 2017 [R2] Online We use online in an algorithmic sense: frames are processed sequentially without future observations. It reduces memory/computation vs. existing offline pipelines that optimise over a full sequence simultaneously. We agree ⅙ fps is not real-time. We state this clearly and discuss how to address it in future work. [R2/R3] Camera/tissue motion Camera and tissue motion can happen at the same time. However, large tissue deformations are caused by tools while the endoscope is stationary (the da Vinci camera locks during tool motion and vice versa). We assume that during camera motion, tissue motion is much smaller (e.g. breathing) compared to direct tissue manipulation and thus constraining tissue deformation in these cases provides a fair approximation leading to better motion estimation. [R2/R3] Optical flow gating [R3] We agree using epipolar constraints with RANSAC would be a more principled approach for motion gating. Optical flow statistics is a simple solution that doesn’t require extensive sampling in high outlier environments. We verified experimentally that there is very good separation between the 2 motion modes (shown in the supplementary). We agree that optical-axis motion is a limitation of our simplification, which didn’t significantly affect performance as lateral motions are significantly more frequent in MIS after the initial camera insertion. [R2] Flow magnitude is not used during gating, but it is used during scene-flow and pose/deformation initialization, so small noisy motions do not contribute equally to large motions during optimisation. We will clarify this.



Meta-Review

Meta-review #1

  • Your recommendation

    Invite for Rebuttal

  • Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.

    The submission received mixed reviews (2xWR, WA). The authors should prepare a rebuttal addressing the raised issues and in particular the issues raised by R1 and R2.

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation.

    The rebuttal clarified several questions and persuaded R2 to recommend acceptance. Nevertheless the paper still has mixed scores 2xA + 1xR and many outstanding concerns remain namely on clarity and missing references. I will lean towards majority and support acceptance trusting that authors will fix many of the clarity issues in an eventual camera ready version: e.g. clarify input modality, state assumption to disambiguate camera from tissue motion and how it affect generality of method, discuss preference for optical flow statistics as opposed to epipolar geometry + RANSAC and inherent limitation when camera moves along optical axis. I also have no objections to the inclusion of the results that demonstrate moderate resilience to errors in tracking as this does not requrie a new dataset/experiment.



Meta-review #2

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation.

    Note on additional experiments; Additional experiments were requested by reviewers, which should not have been done. Results from a noise sensitivity analysis were presented in the rebuttal, which I have excluded form my decision-making.

    This submission addresses an important problem in online deformable reconstruction for surgical SLAM. All reviewers agree that the method is relevant and shows promising empirical results on the StereoMIS dataset, with improvements over baseline approaches.

    R1 recommends weak accept, citing solid experimental results but noting incomplete related work coverage and missing reproducibility details. R2 initially recommended weak reject due to concerns about single-dataset evaluation, reliance on 2D tracking quality, heuristic design choices, and runtime limitations. After rebuttal, this reviewer switches to accept, acknowledging improved clarification of contributions, though several practical and methodological limitations remain. The additional noise sensitivity results presented in the rebuttal, were not acknowledged in the post-rebuttal review, where the reviewer maintains concerns about robustness to tracking noise.

    R3 maintains a reject after rebuttal, citing persistent issues with method clarity and concerns about generalization regarding motion constraints and robotic platform. I do not believe, however, these limitations, while valid, are sufficient to sway the reviewer’ majority opinion of accept,

    Overall, there is agreement on the strength of the empirical results, but concerns remain regarding clarity, generalization, reliance on tracking quality, and limited evaluation scope. I recommend that this is a borderline-accept submission. If accepted, I highly recommend that the authors address as best possible the remaining clarity concerns about method clarity, assumptions, and generalization limits.



Meta-review #3

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation.

    Authors have addressed most of the reviewers concerns. I strongly suggest that they address all concerns of the reviewers and additional directions provided by R3 in their revised paper.



back to top