Pith. sign in

REVIEW 4 major objections 6 minor 90 references

Unifying Scale-Aware Depth Prediction and Perceptual Priors for Monocular Endoscope Pose Estimation and Tissue Reconstruction

T0 review · 4 major / 6 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 A monocular endoscope video alone can be turned into scale-consistent depth and a 3D tissue mesh by fitting a single pseudo-stereo baseline between two foundation depth models.

desk verdict A serious engineering integration whose central scale-consistency claim is not supported by the evidence - sign error in Eq. (4) and a single fitted baseline cannot absorb per-frame depth-scale drift. read the letter →

arxiv 2508.11282 v1 pith:WVFCB227 submitted 2025-08-15 cs.CV

Muzammil Khan , Enzo Kerkhof , Matteo Fusaglia , Koert Kuhlmann , Theo Ruers , Françoise J. Siepel This is my paper
classification cs.CV
keywords monocularreconstructiondepthestimationendoscopeposetissuefoundationmodelsscaleconsistencypseudo-stereobaselineendoscopy
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that the metric-scale ambiguity that blocks monocular 3D reconstruction in minimally invasive surgery can be resolved by coupling two pre-trained depth models: Depth Pro, which outputs metric depth on the first frame, and Depth Anything, which outputs fast per-frame disparity. A single scalar, the pseudo-stereo baseline $B$, is fitted so the first frame's disparity-derived depth matches Depth Pro's metric depth, and that same $B$ converts every later disparity map into pseudo-metric depth. The resulting depth maps are temporally refined by warping along RAFT optical flow and blending with the per-frame estimate using LPIPS perceptual similarity, then registered by a dog-leg pose optimizer regularized with an exponential moving average, and fused through TSDF and marching cubes into a tissue mesh. On HEVD and SCARED, the framework reports trajectory length ratios close to 1.0 and lower pose errors than the compared baselines, supporting the claim of scale-consistent reconstruction.

What carries the argument

The pseudo-stereo baseline $B$: a single scalar fitted by L-BFGS-B so that Depth Anything's disparity predicts the same first-frame depth as Depth Pro's metric estimate, via $D^{\mathrm{est}}_1 = f_{\mathrm{pred}}\hat{B}/d^{\mathrm{mde}}_1$. This $B$ is the bridge that converts every subsequent disparity map into metric depth and simultaneously defines the trajectory length ratio used as the paper's scale-awareness metric.

What would settle it

Deliberately scale the first-frame Depth Pro depth by 1.2 on a SCARED sequence with ground truth and rerun the pipeline; if the final trajectory length ratio and surface scale shift by that same factor, the single baseline $B$ is the sole carrier of metric scale. Alternatively, run on a specular low-texture sequence where Depth Pro's zero-shot metric depth is suspect and check whether TLR departs from 1.000.

Watch

Extended reading notes

Core claim

The central discovery is that a single fitted number—the pseudo-stereo baseline $B$—is enough to lift a fast, scale-ambiguous disparity stream into metric scale by anchoring it to the first-frame metric depth of a slower, stronger foundation model. The framework computes $B$ by L-BFGS-B minimization of the discrepancy between Depth Pro's predicted depth $D^{\mathrm{mde}}_1$ and the disparity-based estimate $f_{\mathrm{pred}}\hat{B}/d^{\mathrm{mde}}_1$, then uses that one scalar for every subsequent frame. Temporal coherence is added by warping the previous MAPIS-Depth map along RAFT optical flow and blending it with the current estimate, with the LPIPS perceptual similarity serving as the bl

Load-bearing premise

The load-bearing premise is that Depth Pro's zero-shot first-frame metric depth and focal-length prediction are accurate on endoscopic tissue; if that first scale is wrong, all later pseudo-metric depth and the reconstructed geometry inherit the error with no independent endoscopic calibration to catch it.

Editorial extensions

If this is right

  • A monocular endoscope workflow can produce metric depth and tissue meshes without stereo hardware or patient-specific calibration.
  • Any improvement in the first-frame metric depth network directly improves the whole trajectory's scale fidelity, because all later depths inherit the single baseline.
  • LPIPS-weighted temporal blending suppresses inter-frame depth artifacts from tissue deformation and motion, giving smoother registered point clouds.
  • The WEMA-regularized dog-leg pose optimizer maintains tracking over full HEVD sequences where several compared methods terminate early.
  • A trajectory length ratio close to 1.0 offers a practical, dataset-level check that the reconstruction's global scale has not drifted.

Reading between the lines

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

  • The 'cheap network plus one metric anchor' pattern could transfer to other real-time depth settings—laparoscopy, cystoscopy, arthroscopy—wherever a slower metric model is affordable for a single reference frame.
  • Because the entire scale rests on the first frame, a corrupted or specular first frame would bias the whole run; a sliding-window or multi-frame anchor is a plausible robustness extension the paper does not explore.
  • The ablation that removes optical flow shifts TLR from 1.000 to 1.343, suggesting the perceptual blending is doing substantial work; explicit per-pixel uncertainty from the depth networks could be tested as a cheaper substitute for LPIPS.
  • The TLR metric collapses scale error into one global number; reporting per-keyframe or per-segment scale ratios would show where drift accumulates, which the paper does not include.
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 proposes a unified monocular endoscope pose estimation and tissue reconstruction framework combining a depth module (MAPIS-Depth), a pose module (WEMA-RTDL), and TSDF-based volumetric fusion. MAPIS-Depth uses Depth Pro on the first frame to estimate a pseudo-stereo baseline B and focal length, then Depth Anything per-frame disparities are converted to pseudo-metric depths via D_i = f_pred·B/d_i. The resulting depth maps are temporally refined by RAFT optical-flow warping and LPIPS-based blending. WEMA-RTDL performs coarse-to-fine decoupled SO(3)/SE(3) dog-leg optimisation with EMA regularisation. Experiments on HEVD and SCARED report depth-consistency visualisations, point-cloud correspondence displacements, pose errors against several baselines, an ablation study using the Trajectory Length Ratio, and qualitative reconstructions.

Significance. If the central claims are correct, the framework would be a useful contribution: it combines strong zero-shot depth priors with temporal and geometric refinement, and reports competitive pose accuracy on two public endoscopic benchmarks. The paper is also honest in including ablations and convergence-failure comparisons. The main value lies in the integration of existing components, not in a fundamentally new theoretical result. The significance is limited by the absence of direct metric-depth evaluation and by an internal inconsistency in the temporal blending equation, both of which affect the strength of the central 'scale-consistent depth' claim.

major comments (4)
  1. [Section 3.1, Eq. (4)] The temporal blending rule contradicts the surrounding text. The text states that a lower LPIPS score indicates greater perceptual similarity and thus higher confidence in the warped depth map Dwarp', whereas a higher LPIPS score should lead to increased reliance on the original estimate Dmde_i. Equation (4), however, computes D_hat_i = Slpips·Dwarp'_i + (1-Slpips)·Dmde_i, so a smaller Slpips gives smaller weight to the warped depth. This is exactly opposite to the stated intent, and it undermines the claimed benefit of LPIPS-guided temporal refinement. The equation or the narrative must be corrected, and the ablation conclusions in Section 4.3 that rely on the effect of optical flow / temporal blending should be revisited.
  2. [Section 3.1, Eq. (1) and Section 4.3, Table 3] The 'scale-consistent depth estimates' claim is not established. A single scalar B, fitted once on frame 1 by comparing Depth Pro's metric depth with Depth Anything's disparity, is then used for all later frames as D_i = f_pred·B/d_i. Depth Anything (v1) predicts relative/affine-invariant depth; its per-image output scale can vary across frames, and no mechanism in the paper enforces cross-frame scale consistency. The evidence presented does not test this: the ORB point-to-point displacement metric measures temporal smoothness of registered point clouds, not absolute depth accuracy; the TLR is a trajectory-length proxy; and the ablation in Table 3 replaces Depth Pro with global scaling factors (30, 100, 1000), which is precisely the single-scalar model and cannot expose per-frame scale drift. A direct quantitative depth evaluation against the SCARED ground truth (e.g., AbsRel, RMSE, scal
  3. [Section 4.2 and Table 1/Table 2] The pose evaluation protocol is under-specified. The tables report mean ± std for per-axis, global, and frame-level errors, but the manuscript does not state how the estimated trajectory is aligned with the ground truth (e.g., Umeyama/Sim3 alignment, whether scale is optimised, or whether a fixed scale from the method is used). Since the paper's contribution is precisely scale awareness, the pose results are hard to interpret without this detail. Also missing: the number of evaluated sequences/frames, the selection criterion for the four SCARED sequences, and whether all frames or only keyframes are used. This information is necessary to judge whether the pose gains come from the proposed depth/pose modules or from the evaluation setup.
  4. [Section 4.3 and Fig. 6/Fig. 7] The 'high-fidelity tissue reconstruction' claim is supported only by qualitative figures. No quantitative geometric accuracy metric (e.g., Chamfer distance, F-score, or depth error on reconstructed surfaces) is reported, despite SCARED providing millimetre-scale structured-light ground truth. The point-cloud displacement plots in Fig. 5 reward smoothness and registration consistency, but a smooth trajectory can be consistently wrong in scale. Adding a geometric reconstruction metric would directly test the pseudo-metric depth chain and would strengthen the paper substantially.
minor comments (6)
  1. [Section 3.1, Eq. (3)] The LPIPS formula has an index error: the inner summation should be over channels c, not over layers l, which is already the outer summation index. Please correct the equation.
  2. [Section 3.2.3 and Table 3] The term 'DyEMA' appears in Table 3 but is never defined in the text; the regularisation in Eq. (13) is called EMA. Please clarify the relationship between WEMA, EMA, and DyEMA.
  3. [Section 3.1] The paper uses the term 'pseudo-metric depth' without a precise definition of the unit or the assumed metric scale. Since the method relies on f_pred from Depth Pro and a baseline B, state explicitly whether the output is intended to be in millimetres (as SCARED ground truth) or in an arbitrary but consistent unit.
  4. [Section 3.2.1 and Eq. (6)] The notation in Eq. (6) is confusing: xse3_j,r is written as a 4x4 matrix but then xse3_j,rot and xse3_j,trans are called 3x3 and 3x1 components. Please make the construction of the SE(3) matrix explicit and consistent with the update rule in Eqs. (11)-(12).
  5. [Section 4.2] The definition of TLR would benefit from a comment on how the estimated trajectory length is computed (e.g., cumulative translation of estimated poses) and whether it uses the same scale as the ground-truth trajectory.
  6. [General] Figure 2's caption mentions 'registered pseudo-RGBD reconstructions' in the bottom row, but the bottom row in the text is described as point clouds. For clarity, state explicitly which visualisation is shown and how the point clouds were generated.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: scale calibration is explicit, components are external, and pose/reconstruction are evaluated against ground truth.

full rationale

The paper's central derivation is Eq. (1), which fits a scalar pseudo-stereo baseline B so that the first frame's pseudo-metric depth reproduces Depth Pro's metric depth. This is a transparent calibration step, not a hidden fit: the paper explicitly states that Depth Pro provides robust initialization and Depth Anything provides per-frame disparity, and subsequent depths are propagated through the same B. For frame 1, the computed depth equals the input by construction, but the paper does not claim to predict the first frame from Depth Anything; it uses Depth Pro as an initialization. The claimed contribution is a scale-consistent pseudo-metric sequence, and the scale content is inherited from an external zero-shot metric depth model rather than derived from the paper's own outputs. Pose estimation and trajectory-length ratio are evaluated against SCARED ground truth, providing external validation. The temporal-consistency metric (ORB point displacement) is correlated with the LPIPS-weighted blending of warped previous depth, but this is an expected property of the algorithm rather than a statistically forced prediction; comparisons are made against an unsmoothed baseline. No load-bearing self-citation, imported uniqueness theorem, or ansatz-smuggling citation is present. The skeptic's concern about Depth Anything's per-frame scale drift is a correctness risk and a validation gap (no direct per-frame metric-depth evaluation is reported), not a circularity: the paper's equations do not assume the target result from itself. Therefore no significant circularity is identified.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The framework rests on external foundation models and a fitted scalar for its metric-scale property, plus unspecified EMA and bilateral parameters. The largest unstated load is the transfer of Depth Pro's first-frame metric scale to all frames.

free parameters (4)
  • Pseudo-stereo baseline B = not reported (per sequence)
    Fitted to match Depth Pro's first-frame depth via Eq. (1); all subsequent depth maps are B * disparity / focal length.
  • EMA weights alpha, beta, gamma = not reported
    Pose regularization in Eq. (13) requires alpha + beta + gamma = 1; values are never given.
  • EMA ramp omega = not reported
    Controls s = 1 - exp(-i*omega) in Eq. (13); no value or range is specified.
  • Bilateral filter sigma_d and sigma_r = not reported
    Spatial and range smoothing parameters in Eq. (2) are not specified.
assumptions (5)
  • domain assumption Depth Pro returns metric depth and focal length for the first frame in endoscopic scenes.
    All downstream scale depends on this initial depth; no endoscopic validation is reported.
  • domain assumption Depth Anything disparity is inversely proportional to depth with a single global scale B.
    Eq. (1) and Section 3.1 assume the disparity-to-depth mapping is constant across frames despite deformation, illumination changes, and varying tissue.
  • domain assumption LPIPS distance is a suitable confidence signal for choosing between warped and current depth maps.
    LPIPS is a perceptual image metric, not a depth-confidence metric; the paper uses it directly as a blending weight.
  • domain assumption The scene is locally rigid enough for SE(3) frame alignment in WEMA-RTDL.
    Eqs. (5)-(12) optimize a rigid camera transformation, while the paper claims to handle physiological tissue deformation.
  • standard math Dog-leg trust-region and L-BFGS-B converge to correct optima for this non-convex problem.
    Optimization routines [22,73] are used without proof of global convergence; this is standard practice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Unifying Scale-Aware Depth Prediction and Perceptual Priors for Monocular Endoscope Pose Estimation and Tissue Reconstruction." pith.science (2026). https://pith.science/paper/WVFCB227

@misc{pith2026250811282,
  author       = {Pith},
  title        = {Pith review of: Unifying Scale-Aware Depth Prediction and Perceptual Priors for Monocular Endoscope Pose Estimation and Tissue Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WVFCB227}},
  note         = {Machine review of arXiv:2508.11282}
}
read the original abstract

Accurate endoscope pose estimation and 3D tissue surface reconstruction significantly enhances monocular minimally invasive surgical procedures by enabling accurate navigation and improved spatial awareness. However, monocular endoscope pose estimation and tissue reconstruction face persistent challenges, including depth ambiguity, physiological tissue deformation, inconsistent endoscope motion, limited texture fidelity, and a restricted field of view. To overcome these limitations, a unified framework for monocular endoscopic tissue reconstruction that integrates scale-aware depth prediction with temporally-constrained perceptual refinement is presented. This framework incorporates a novel MAPIS-Depth module, which leverages Depth Pro for robust initialisation and Depth Anything for efficient per-frame depth prediction, in conjunction with L-BFGS-B optimisation, to generate pseudo-metric depth estimates. These estimates are temporally refined by computing pixel correspondences using RAFT and adaptively blending flow-warped frames based on LPIPS perceptual similarity, thereby reducing artefacts arising from physiological tissue deformation and motion. To ensure accurate registration of the synthesised pseudo-RGBD frames from MAPIS-Depth, a novel WEMA-RTDL module is integrated, optimising both rotation and translation. Finally, truncated signed distance function-based volumetric fusion and marching cubes are applied to extract a comprehensive 3D surface mesh. Evaluations on HEVD and SCARED, with ablation and comparative analyses, demonstrate the framework's robustness and superiority over state-of-the-art methods.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

90 extracted references · 70 canonical work pages

  1. [1]

    Three-dimensional laparoscopy

    Desmond H Birkett. Three-dimensional laparoscopy. Journal of laparoendoscopic surgery, 5(5):327–331, 1995

    1995

  2. [2]

    Towards live monocular 3d laparoscopy using shading and specularity informa- tion

    Toby Collins and Adrien Bartoli. Towards live monocular 3d laparoscopy using shading and specularity informa- tion. In International Conference on Information Processing in Computer-Assisted Interventions, pages 11–21, 2012

    2012

  3. [3]

    The effects of physical constraints in laparoscopic surgery

    Antony J Hodgson, John G Person, Septimiu E Salcudean, and Alex G Nagy. The effects of physical constraints in laparoscopic surgery. Medical Image Analysis, 3(3):275–283, 1999

    1999

  4. [4]

    Principles of laparoscopic surgery

    Christopher Limb and Timothy Rockall. Principles of laparoscopic surgery. Surgery (Oxford), 41(2):106–116, 2023

    2023

  5. [5]

    Surface deformation tracking in monocular laparoscopic video

    Ziteng Liu, Wenpeng Gao, Jiahua Zhu, Zhi Yu, and Yili Fu. Surface deformation tracking in monocular laparoscopic video. Medical image analysis, 86:102775, 2023

    2023

  6. [6]

    Non-rigid scene reconstruction of deformable soft tissue with monocular endoscopy in minimally invasive surgery

    Enpeng Wang, Yueang Liu, Jiangchang Xu, and Xiaojun Chen. Non-rigid scene reconstruction of deformable soft tissue with monocular endoscopy in minimally invasive surgery. International Journal of Computer Assisted Radiology and Surgery, pages 1–11, 2024

    2024

  7. [7]

    Learning how to robustly estimate camera pose in endoscopic videos

    Michel Hayoz, Christopher Hahne, Mathias Gallardo, Daniel Candinas, Thomas Kurmann, Maximilian Allan, and Raphael Sznitman. Learning how to robustly estimate camera pose in endoscopic videos. International Journal of Computer Assisted Radiology and Surgery, 18(7):1185–1192, 2023

    2023

  8. [8]

    3d texture reconstruction of abdominal cavity based on monocular vision slam for minimally invasive surgery.Symmetry, 14(2):185, 2022

    Haibin Wu, Ruotong Xu, Kaiyang Xu, Jianbo Zhao, Yan Zhang, Aili Wang, and Yuji Iwahori. 3d texture reconstruction of abdominal cavity based on monocular vision slam for minimally invasive surgery.Symmetry, 14(2):185, 2022

    2022

Show all 90 references
  1. [9]

    Absolute depth estimate in stereopsis

    A Verri and V Torre. Absolute depth estimate in stereopsis. JOSA A, 3(3):297–299, 1986

    1986

  2. [10]

    Models of monocular and binocular visual perception in quality assessment of stereoscopic images

    Feng Shao, Weisi Lin, Gangyi Jiang, and Qionghai Dai. Models of monocular and binocular visual perception in quality assessment of stereoscopic images. IEEE Transactions on Computational Imaging, 2(2):123–135, 2016

    2016

  3. [11]

    Monocular depth estimation: A thorough review

    Vasileios Arampatzakis, George Pavlidis, Nikolaos Mitianoudis, and Nikos Papamarkos. Monocular depth estimation: A thorough review. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

    2023

  4. [12]

    Opa-3d: Occlusion-aware pixel-wise aggregation for monocular 3d object detection

    Yongzhi Su, Yan Di, Guangyao Zhai, Fabian Manhardt, Jason Rambach, Benjamin Busam, Didier Stricker, and Federico Tombari. Opa-3d: Occlusion-aware pixel-wise aggregation for monocular 3d object detection. IEEE Robotics and Automation Letters, 8(3):1327–1334, 2023

    2023

  5. [13]

    Monocular cues are superior to binocular cues for size perception when they are in conflict in virtual reality

    Sang-Ah Yoo, Suhyun Lee, and Sung Jun Joo. Monocular cues are superior to binocular cues for size perception when they are in conflict in virtual reality. Cortex, 166:80–90, 2023

    2023

  6. [14]

    Luminance contrast provides metric depth information

    Paul B Hibbard, Ross Goutcher, Rebecca L Hornsey, David W Hunter, and Peter Scarfe. Luminance contrast provides metric depth information. Royal Society Open Science, 10(2):220567, 2023

    2023

  7. [15]

    A monocular slam system based on sift features for gastroscope tracking

    Yifan Wang, Liang Zhao, Lun Gong, Xin Chen, and Siyang Zuo. A monocular slam system based on sift features for gastroscope tracking. Medical & Biological Engineering & Computing, 61(2):511–523, 2023

    2023

  8. [16]

    Stereo correspondence and reconstruction of endoscopic data challenge

    Max Allan, Jonathan Mcleod, Congcong Wang, Jean Claude Rosenthal, Zhenglei Hu, Niklas Gard, Peter Eisert, Ke Xue Fu, Trevor Zeffiro, Wenyao Xia, et al. Stereo correspondence and reconstruction of endoscopic data challenge. arXiv preprint arXiv:2101.01133, 2021

    2021 arXiv

  9. [17]

    Colonoscopy 3d video dataset with paired depth from 2d-3d registration

    Taylor L Bobrow, Mayank Golhar, Rohan Vijayan, Venkata S Akshintala, Juan R Garcia, and Nicholas J Durr. Colonoscopy 3d video dataset with paired depth from 2d-3d registration. Medical image analysis, 90:102956, 2023

    2023

  10. [18]

    Depth pro: Sharp monocular metric depth in less than a second

    Aleksei Bochkovskii, Amaël Delaunoy, Hugo Germain, Marcel Santos, Yichao Zhou, Stephan R Richter, and Vladlen Koltun. Depth pro: Sharp monocular metric depth in less than a second. arXiv preprint arXiv:2410.02073, 2024

    2024 arXiv

  11. [19]

    Depth anything: Unleashing the power of large-scale unlabeled data

    Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything: Unleashing the power of large-scale unlabeled data. In International Conference on Computer Vision and Pattern Recognition, pages 10371–10381, 2024. 14 A PREPRINT - SEPTEMBER 2, 2025

    2024

  12. [20]

    Raft: Recurrent all-pairs field transforms for optical flow

    Zachary Teed and Jia Deng. Raft: Recurrent all-pairs field transforms for optical flow. In European Conference on Computer Vision, pages 402–419, 2020

    2020

  13. [21]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In International Conference on Computer Vision and Pattern Recognition, pages 586–595, 2018

    2018

  14. [22]

    The software tool for comparison and configuration of nonlinear optimization techniques in orb-slam

    Boris Makaev and Ilya Afanasyev. The software tool for comparison and configuration of nonlinear optimization techniques in orb-slam. In International Conference on Developments in eSystems Engineering, pages 489–494, 2019

    2019

  15. [23]

    Exponential smoothing for predicting demand

    Robert G Brown. Exponential smoothing for predicting demand. Little, 1956

    1956

  16. [24]

    A volumetric method for building complex models from range images

    Brian Curless and Marc Levoy. A volumetric method for building complex models from range images. In Annual Conference on Computer Graphics and Interactive Techniques, pages 303–312, 1996

    1996

  17. [25]

    Marching cubes: A high resolution 3d surface construction algorithm

    William E Lorensen and Harvey E Cline. Marching cubes: A high resolution 3d surface construction algorithm. In Seminal graphics: pioneering efforts that shaped the field, pages 347–353. 1998

    1998

  18. [26]

    Endo-depth-and-motion: Reconstruction and tracking in endoscopic videos using depth networks and photometric constraints

    David Recasens, José Lamarca, José M Fácil, JMM Montiel, and Javier Civera. Endo-depth-and-motion: Reconstruction and tracking in endoscopic videos using depth networks and photometric constraints. IEEE Robotics and Automation Letters, 6(4):7225–7232, 2021

    2021

  19. [27]

    Deep learning-based 3d reconstruction: a survey

    Taha Samavati and Mohsen Soryani. Deep learning-based 3d reconstruction: a survey. Artificial Intelligence Review, 56(9):9175–9219, 2023

    2023

  20. [28]

    Deep-learning-based 3-d surface reconstruction—a survey

    Anis Farshian, Markus Götz, Gabriele Cavallaro, Charlotte Debus, Matthias Nießner, Jón Atli Benediktsson, and Achim Streit. Deep-learning-based 3-d surface reconstruction—a survey. Proceedings of the IEEE, 2023

    2023

  21. [29]

    Mast3r-slam: Real-time dense slam with 3d reconstruction priors

    Riku Murai, Eric Dexheimer, and Andrew J Davison. Mast3r-slam: Real-time dense slam with 3d reconstruction priors. arXiv preprint arXiv:2412.12392, 2024

    2024 arXiv

  22. [30]

    Gaussian splatting slam

    Hidenobu Matsuki, Riku Murai, Paul HJ Kelly, and Andrew J Davison. Gaussian splatting slam. In International Conference on Computer Vision and Pattern Recognition, pages 18039–18048, 2024

    2024

  23. [31]

    Instantsplat: Sparse-view sfm-free gaussian splatting in seconds

    Zhiwen Fan, Kairun Wen, Wenyan Cong, Kevin Wang, Jian Zhang, Xinghao Ding, Danfei Xu, Boris Ivanovic, Marco Pavone, Georgios Pavlakos, et al. Instantsplat: Sparse-view sfm-free gaussian splatting in seconds. arXiv preprint arXiv:2403.20309, 2024

    2024 arXiv

  24. [32]

    Obstacle avoidance and navigation in the real world by a seeing robot rover

    Hans Peter Moravec. Obstacle avoidance and navigation in the real world by a seeing robot rover . Stanford University, 1980

    1980

  25. [33]

    Crowd-slam: visual slam towards crowded environments using object detection

    João Carlos Virgolino Soares, Marcelo Gattass, and Marco Antonio Meggiolaro. Crowd-slam: visual slam towards crowded environments using object detection. Journal of Intelligent & Robotic Systems, 102(2):50, 2021

    2021

  26. [34]

    A study of surface reconstruction for 3d mannequins based on feature curves

    Shih Wen Hsiao and Rong Qi Chen. A study of surface reconstruction for 3d mannequins based on feature curves. Computer-Aided Design, 45(11):1426–1441, 2013

    2013

  27. [35]

    A statistical method for robust 3d surface reconstruction from sparse data

    V olker Blanz, Albert Mehl, Thomas Vetter, and H P Seidel. A statistical method for robust 3d surface reconstruction from sparse data. In International Symposium on 3D Data Processing, Visualization and Transmission, pages 293–300, 2004

    2004

  28. [36]

    A quasi-dense approach to surface reconstruction from uncalibrated images

    Maxime Lhuillier and Long Quan. A quasi-dense approach to surface reconstruction from uncalibrated images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 27(3):418–433, 2005

    2005

  29. [37]

    A comparison of semiglobal and local dense matching algorithms for surface reconstruction

    Elisa Dall’Asta and Riccardo Roncella. A comparison of semiglobal and local dense matching algorithms for surface reconstruction. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 40:187–194, 2014

    2014

  30. [38]

    A hybrid approach to smooth surface reconstruction from 2-d cross sections

    Hyungjun Park. A hybrid approach to smooth surface reconstruction from 2-d cross sections. The International Journal of Advanced Manufacturing Technology, 25:1130–1136, 2005

    2005

  31. [39]

    Visual slam algorithms: A survey from 2010 to 2016

    Takafumi Taketomi, Hideaki Uchiyama, and Sei Ikeda. Visual slam algorithms: A survey from 2010 to 2016. IPSJ Transactions on Computer Vision and Applications, 9:1–11, 2017

    2010

  32. [40]

    Visual slam with line and corner features

    Woo Yeon Jeong and Kyoung Mu Lee. Visual slam with line and corner features. In International Conference on Intelligent Robots and Systems, pages 2570–2575. IEEE, 2006

    2006

  33. [41]

    Orb-slam: a versatile and accurate monocular slam system

    Raul Mur-Artal, Jose Maria Martinez Montiel, and Juan D Tardos. Orb-slam: a versatile and accurate monocular slam system. IEEE Transactions on Robotics, 31(5):1147–1163, 2015

    2015

  34. [42]

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

    Carlos Campos, Richard Elvira, Juan J Gómez Rodríguez, José MM Montiel, and Juan D Tardós. Orb-slam3: An accurate open-source library for visual, visual–inertial, and multimap slam. IEEE Transactions on Robotics, 37(6):1874–1890, 2021. 15 A PREPRINT - SEPTEMBER 2, 2025

    2021

  35. [43]

    Parallel tracking and multiple mapping (ptamm) manual

    Robert Castle. Parallel tracking and multiple mapping (ptamm) manual. Robotics Research Group, Department of Engineering Science, University of Oxford, 2010

    2010

  36. [44]

    Object recognition from local scale-invariant features

    David G Lowe. Object recognition from local scale-invariant features. In International Conference on Computer Vision, volume 2, pages 1150–1157, 1999

    1999

  37. [45]

    Speeded-up robust features (surf)

    Herbert Bay, Andreas Ess, Tinne Tuytelaars, and Luc Van Gool. Speeded-up robust features (surf). Computer Vision and Image Understanding, 110(3):346–359, 2008

    2008

  38. [46]

    Orb: An efficient alternative to sift or surf

    Ethan Rublee, Vincent Rabaud, Kurt Konolige, and Gary Bradski. Orb: An efficient alternative to sift or surf. In International Conference on Computer Vision, pages 2564–2571, 2011

    2011

  39. [47]

    Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography

    Martin A Fischler and Robert C Bolles. Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography. Communications of the ACM, 24(6):381–395, 1981

    1981

  40. [48]

    Bundle adjustment—a modern synthesis

    Bill Triggs, Philip F McLauchlan, Richard I Hartley, and Andrew W Fitzgibbon. Bundle adjustment—a modern synthesis. In Vision Algorithms: Theory and Practice, pages 298–372, 2000

    2000

  41. [49]

    Ekf monocular slam with relocalization for laparoscopic sequences

    Oscar G Grasa, Javier Civera, and JMM Montiel. Ekf monocular slam with relocalization for laparoscopic sequences. In International Conference on Robotics and Automation, pages 4816–4821, 2011

    2011

  42. [50]

    Unscented slam for large-scale outdoor environments

    Ruben Martinez Cantin and José A Castellanos. Unscented slam for large-scale outdoor environments. In International Conference on Intelligent Robots and Systems, pages 3427–3432, 2005

    2005

  43. [51]

    A square root unscented kalman filter for visual monoslam

    Steven Holmes, Georg Klein, and David W Murray. A square root unscented kalman filter for visual monoslam. In International Conference on Robotics and Automation, pages 3710–3716, 2008

    2008

  44. [52]

    3d intra- articular dense reconstruction from arthroscopic images

    Zhang Yanhao, Cheng Chonghao, Falque Raphael, Zhao Liang, Huang Shoudong, and Chen Yongbo. 3d intra- articular dense reconstruction from arthroscopic images. InInternational Conference on Robotics and Biomimetics, pages 1–7, 2023

    2023

  45. [53]

    Super: A surgical perception framework for endoscopic tissue manipulation with surgical robotics

    Yang Li, Florian Richter, Jingpei Lu, Emily K Funk, Ryan K Orosco, Jianke Zhu, and Michael C Yip. Super: A surgical perception framework for endoscopic tissue manipulation with surgical robotics. IEEE Robotics and Automation Letters, 5(2):2294–2301, 2020

    2020

  46. [54]

    Live tracking and dense reconstruction for handheld monocular endoscopy

    Nader Mahmoud, Toby Collins, Alexandre Hostettler, Luc Soler, Christophe Doignon, and Jose Maria Martinez Montiel. Live tracking and dense reconstruction for handheld monocular endoscopy. IEEE Transactions on Medical Imaging, 38(1):79–89, 2018

    2018

  47. [55]

    Advanced algorithms of slam in medical endoscopy

    Shuai Wang, Zhizhe Yu, Yudi Zhao, and Yunpeng Zhu. Advanced algorithms of slam in medical endoscopy. In AIP Conference Proceedings, volume 3017, 2023

    2023

  48. [56]

    Spsvo: A self-supervised surgical perception stereo visual odometer for endoscopy

    Junjie Zhao, Yang Luo, Qimin Li, Natalie Baddour, and Md Sulayman Hossen. Spsvo: A self-supervised surgical perception stereo visual odometer for endoscopy. Robotica, 41(12):3724–3745, 2023

    2023

  49. [57]

    Unsupervised learning of depth and ego-motion from video

    Tinghui Zhou, Matthew Brown, Noah Snavely, and David G Lowe. Unsupervised learning of depth and ego-motion from video. In International Conference on Computer Vision and Pattern Recognition, pages 1851–1858, 2017

    2017

  50. [58]

    Ongoing evolution of visual slam from geometry to deep learning: Challenges and opportunities

    Ruihao Li, Sen Wang, and Dongbing Gu. Ongoing evolution of visual slam from geometry to deep learning: Challenges and opportunities. Cognitive Computation, 10(6):875–889, 2018

    2018

  51. [59]

    Indoor relocalization in challenging environments with dual-stream convolutional neural networks

    Ruihao Li, Qiang Liu, Jianjun Gui, Dongbing Gu, and Huosheng Hu. Indoor relocalization in challenging environments with dual-stream convolutional neural networks. IEEE Transactions on Automation Science and Engineering, 15(2):651–662, 2017

    2017

  52. [60]

    Vidloc: A deep spatio-temporal model for 6-dof video-clip relocalization

    Ronald Clark, Sen Wang, Andrew Markham, Niki Trigoni, and Hongkai Wen. Vidloc: A deep spatio-temporal model for 6-dof video-clip relocalization. InInternational Conference on Computer Vision and Pattern Recognition, pages 6856–6864, 2017

    2017

  53. [61]

    Posenet: A convolutional network for real-time 6-dof camera relocalization

    Alex Kendall, Matthew Grimes, and Roberto Cipolla. Posenet: A convolutional network for real-time 6-dof camera relocalization. In International Conference on Computer Vision, pages 2938–2946, 2015

    2015

  54. [62]

    Deepvo: Towards end-to-end visual odometry with deep recurrent convolutional neural networks

    Sen Wang, Ronald Clark, Hongkai Wen, and Niki Trigoni. Deepvo: Towards end-to-end visual odometry with deep recurrent convolutional neural networks. In International Conference on Robotics and Automation, pages 2043–2050, 2017

  55. [63]

    Undeepvo: Monocular visual odometry through unsupervised deep learning

    Ruihao Li, Sen Wang, Zhiqiang Long, and Dongbing Gu. Undeepvo: Monocular visual odometry through unsupervised deep learning. In International Conference on Robotics and Automation, pages 7286–7291, 2018

    2018

  56. [64]

    Nice-slam: Neural implicit scalable encoding for slam

    Zihan Zhu, Songyou Peng, Viktor Larsson, Weiwei Xu, Hujun Bao, Zhaopeng Cui, Martin R Oswald, and Marc Pollefeys. Nice-slam: Neural implicit scalable encoding for slam. In International Conference on Computer Vision and Pattern Recognition, pages 12786–12796, 2022. 16 A PREPRI...

    2022

  57. [65]

    Unsupervised monocular depth estimation with left-right consistency

    Clement Godard, Oisin Mac Aodha, and Gabriel J Brostow. Unsupervised monocular depth estimation with left-right consistency. In International Conference on Computer Vision, pages 270–279, 2017

    2017

  58. [66]

    Artificial intelligence in surgery: Promises and perils

    Daniel A Hashimoto, Guy Rosman, Daniela Rus, and Ozanan R Meireles. Artificial intelligence in surgery: Promises and perils. Annals of surgery, 268(1):70–76, 2018

    2018

  59. [67]

    Endogslam: Real-time dense reconstruction and tracking in endoscopic surgeries using gaussian splatting

    Kailing Wang, Chen Yang, Yuehao Wang, Sikuang Li, Yan Wang, Qi Dou, Xiaokang Yang, and Wei Shen. Endogslam: Real-time dense reconstruction and tracking in endoscopic surgeries using gaussian splatting. arXiv preprint arXiv:2403.15124, 2024

    2024 arXiv

  60. [68]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4):139–1, 2023

    2023

  61. [69]

    Bodyslam: A generalized monocular visual slam framework for surgical applications

    G Manni, C Lauretti, F Prata, R Papalia, L Zollo, and P Soda. Bodyslam: A generalized monocular visual slam framework for surgical applications. arXiv preprint arXiv:2408.03078, 2024

    2024 arXiv

  62. [70]

    Zoedepth: Zero-shot transfer by combining relative and metric depth

    Shariq Farooq Bhat, Reiner Birkl, Diana Wofk, Peter Wonka, and Matthias Müller. Zoedepth: Zero-shot transfer by combining relative and metric depth. arXiv preprint arXiv:2302.12288, 2023

    2023 arXiv

  63. [71]

    Globally optimal fetoscopic mosaicking based on pose graph optimisation with affine constraints

    Liang Li, Sophia Bano, Jan Deprest, Anna L David, Danail Stoyanov, and Francisco Vasconcelos. Globally optimal fetoscopic mosaicking based on pose graph optimisation with affine constraints. IEEE Robotics and Automation Letters, 6(4):7831–7838, 2021

    2021

  64. [72]

    A survey on vision transformer

    Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao, Chunjing Xu, Yixing Xu, et al. A survey on vision transformer. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(1):87–110, 2022

    2022

  65. [73]

    Algorithm 778: L-bfgs-b: Fortran subroutines for large-scale bound-constrained optimization

    Ciyou Zhu, Richard H Byrd, Peihuang Lu, and Jorge Nocedal. Algorithm 778: L-bfgs-b: Fortran subroutines for large-scale bound-constrained optimization. ACM Transactions on Mathematical Software, 23(4):550–560, 1997

    1997

  66. [74]

    G. Bradski. The OpenCV Library. Dr. Dobb’s Journal of Software Tools, 2000

    2000

  67. [75]

    Bilateral filtering for gray and color images

    Carlo Tomasi and Roberto Manduchi. Bilateral filtering for gray and color images. In International Conference on Computer Vision, pages 839–846, 1998

    1998

  68. [76]

    Imagenet classification with deep convolutional neural networks

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. Communications of the ACM, 60(6):84–90, 2017

    2017

  69. [77]

    Multiple view geometry in computer vision

    Richard Hartley and Andrew Zisserman. Multiple view geometry in computer vision. Cambridge University Press, 2003

    2003

  70. [78]

    Theoria motus corporum coelestium in sectionibus conicis solem ambientium auctore Carolo Friderico Gauss

    Carl Friedrich Gauss. Theoria motus corporum coelestium in sectionibus conicis solem ambientium auctore Carolo Friderico Gauss. sumtibus Frid. Perthes et IH Besser, 1809

  71. [79]

    The method of steepest descent for non-linear minimization problems

    Haskell B Curry. The method of steepest descent for non-linear minimization problems. Quarterly of Applied Mathematics, 2(3):258–261, 1944

    1944

  72. [80]

    Motion segmenta- tion of truncated signed distance function based volumetric surfaces

    Samunda Perera, Nick Barnes, Xuming He, Shahram Izadi, Pushmeet Kohli, and Ben Glocker. Motion segmenta- tion of truncated signed distance function based volumetric surfaces. In Winter Conference on Applications of Computer Vision, pages 1046–1053, 2015

    2015

  73. [81]

    A benchmark for the evaluation of rgb-d slam systems

    Jürgen Sturm, Nikolas Engelhard, Felix Endres, Wolfram Burgard, and Daniel Cremers. A benchmark for the evaluation of rgb-d slam systems. In International Conference on Intelligent Robots and Systems, pages 573–580, 2012

    2012

  74. [82]

    Predicting performance of slam algorithms

    Matteo Luperto, Valerio Castelli, and Francesco Amigoni. Predicting performance of slam algorithms. arXiv preprint arXiv:2109.02329, 2021

    2021 arXiv

  75. [83]

    The influence of inaccurate gpu power measurements for machine learning workloads in industrial applications

    Marco Wagner and Devesh Vashisth. The influence of inaccurate gpu power measurements for machine learning workloads in industrial applications. In International Conference on Human-Computer Interaction, pages 233–247. Springer, 2025

    2025

  76. [84]

    Advancing dense endoscopic reconstruction with gaussian splatting-driven surface normal-aware tracking and mapping

    Yiming Huang, Beilei Cui, Long Bai, Zhen Chen, Jinlin Wu, Zhen Li, Hongbin Liu, and Hongliang Ren. Advancing dense endoscopic reconstruction with gaussian splatting-driven surface normal-aware tracking and mapping. arXiv preprint arXiv:2501.19319, 2025

    2025 arXiv

  77. [85]

    Oneslam to map them all: A generalized approach to slam for monocular endoscopic imaging based on tracking any point

    Timo Teufel, Hongchao Shu, Roger D Soberanis-Mukul, Jan Emily Mangulabnan, Manish Sahu, S Swaroop Vedula, Masaru Ishii, Gregory Hager, Russell H Taylor, and Mathias Unberath. Oneslam to map them all: A generalized approach to slam for monocular endoscopic imaging based on trac...

    2024

  78. [86]

    Bodyslam: Joint camera localisation, mapping, and human motion tracking

    Dorian F Henning, Tristan Laidlow, and Stefan Leutenegger. Bodyslam: Joint camera localisation, mapping, and human motion tracking. In European Conference on Computer Vision, pages 656–673. Springer, 2022. 17 A PREPRINT - SEPTEMBER 2, 2025

    2022

  79. [87]

    Method for registration of 3-d shapes

    Paul J Besl and Neil D McKay. Method for registration of 3-d shapes. In Sensor fusion IV: Control paradigms and data structures, volume 1611, pages 586–606. Spie, 1992

    1992

  80. [88]

    Object modelling by registration of multiple range images

    Yang Chen and Gérard Medioni. Object modelling by registration of multiple range images. Image and Vision Computing, 10(3):145–155, 1992

    1992

  81. [89]

    Real-time visual odometry from dense rgb-d images

    Frank Steinbrücker, Jürgen Sturm, and Daniel Cremers. Real-time visual odometry from dense rgb-d images. In International Conference on Computer Vision, pages 719–722, 2011

    2011

  82. [90]

    Colored point cloud registration revisited

    Jaesik Park, Qian-Yi Zhou, and Vladlen Koltun. Colored point cloud registration revisited. In International Conference on Computer Vision, pages 143–152, 2017. 18

    2017

Pith tools

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