Pith. sign in

REVIEW 3 major objections 7 minor 35 references

SurgRIPE challenge: Benchmark of Surgical Robot Instrument Pose Estimation

T0 review · 3 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The SurgRIPE challenge provides a realistic surgical video dataset with ground-truth 6DoF instrument poses and a benchmark for markerless pose estimation, with accuracy validated in simulation at 0.253 mm translation and 0.302 degrees…

desk verdict A genuinely useful benchmark for markerless 6DoF surgical instrument pose, but the ground-truth validation leaves out the inpainting step and the rankings lack uncertainty. read the letter →

arxiv 2501.02990 v2 pith:YINQUYFA submitted 2025-01-06 cs.CV cs.RO

Haozheng Xu , Alistair Weld , Chi Xu , Alfie Roddan , Joao Cartucho , Mert Asim Karaoglu , Alexander Ladikos , Yangke Li
show 11 more authors
Yiping Li Daiyun Shen Geonhee Lee Seyeon Park Jongho Shin Young-Gon Kim Lucy Fothergill Dominic Jones Pietro Valdastri Duygu Sarikaya Stamatia Giannarou
This is my paper · ORCID
classification cs.CVcs.RO
keywords Surgicalinstrumentposeestimation6DoFmarkerlesstrackingbenchmarkdatasetimageinpaintinggroundtruthannotationroboticsurgerymetrics
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

Accurate 6DoF poses of surgical instruments are hard to obtain in real endoscopic video, and existing surgical datasets mostly offer 2D labels or kinematics rather than true 3D poses. This paper introduces the SurgRIPE challenge dataset, which pairs realistic monocular endoscopic frames with ground-truth 6DoF poses of the instrument wrist, generated by tracking a removable keydot marker and then inpainting the marker out of the image. The authors argue this is the first pipeline that can produce a large markerless surgical pose dataset with high-accuracy annotations, and they support the accuracy with a simulation-based check reporting 0.253 mm translation error and 0.302 degrees rotation error. The challenge then evaluates six submitted markerless methods and a baseline using ADD, Average Accuracy (0–5 mm), translation, rotation, proj2d, and mmd5 metrics. If the ground truth is accurate in real scenes, the released dataset and toolkit become a reusable evaluation standard for markerless surgical pose estimation, with direct implications for autonomous robotic surgery.

What carries the argument

The load-bearing mechanism is the marker-to-instrument pose chain. A PnP solver estimates the keydot marker pose $T^C_M$; a hand-adjusted transformation $T^M_I$, tuned in a GUI until the projected 3D model aligns with the instrument, converts this into the instrument wrist pose $P_I$; and the same projection generates segmentation masks. Deep-learning inpainting then removes the marker, so the released frames look markerless and contain no consistent visual shortcut for a network to exploit. The evaluation protocol—ADD, Average Accuracy over 0–5 mm thresholds, translation error, rotation error, proj2d, and mmd5—is what turns the dataset into a benchmark and determines the method ranking.

What would settle it

Reproject the provided 3D instrument model onto the real test frames using the published ground-truth poses and measure the mean pixel distance from the projected contour to the visible instrument contour; if that distance substantially exceeds the error implied by the simulation validation, or if an independent tracker such as an electromagnetic sensor disagrees with the poses by more than a few millimeters, the claimed real-world accuracy does not hold.

Watch

Extended reading notes

Core claim

The central claim is that markerless 6DoF surgical instrument pose estimation can be benchmarked on realistic images instead of simulation or marker-laden scenes, and that the SurgRIPE dataset delivers the ground truth needed to do so. The authors' pipeline estimates the pose of a keydot marker attached to the instrument wrist using a PnP solver, hand-tunes the fixed transform from marker to instrument in a GUI until the 3D model projection overlays the instrument, and removes the marker with deep-learning inpainting so the final images contain no marker cues. Segmentation masks are generated by projecting the 3D model using the solved pose. Accuracy is claimed on the basis of 50 simulated frames, where the pipeline's translation error is 0.253 mm and rotation error is 0.302 degrees. Using this dataset, the challenge ranks six submissions plus a PVNet baseline, with the winning method achieving the highest ADD and Average Accuracy across all four test subsets (with and without occlusion for both instrument types).

Load-bearing premise

The load-bearing premise is that the marker-to-instrument transform hand-tuned in the GUI is correct in the real endoscopic images, even though the pipeline's accuracy was only checked on 50 simulated frames.

Editorial extensions

If this is right

  • Markerless deep-learning pose estimators can now be trained and tested on realistic endoscopic frames with dense 6DoF labels rather than simulation-only or kinematic-only data.
  • The benchmark provides a public ranking protocol, so future methods can be compared directly on occluded and non-occluded sequences for two instrument types.
  • The accuracy check on simulated frames suggests the annotation pipeline can produce sub-millimeter, sub-degree ground truth, which is the precision needed for instruments only a few millimeters wide.
  • The winning method's consistent advantage across all four test subsets indicates that candidate-hypothesis pose estimation generalizes well to occluded surgical scenes.

Reading between the lines

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

  • The paper validates ground-truth accuracy only on simulated images; if the hand-tuned marker-to-instrument transform is biased in real endoscopic scenes, all pose labels and rankings inherit that bias. An independent check on real frames would settle this.
  • Inpainting removes the marker but could leave faint artifacts; a method could in principle exploit those artifacts as a shortcut, so an artifact audit would strengthen the benchmark's claim to measure genuine pose estimation.
  • The manual GUI calibration limits how quickly new instruments can be added; automating the marker-to-instrument calibration would make the pipeline scalable to the full instrument set.
  • The paper's metric choice—Average Accuracy over 0–5 mm rather than BOP's 10% of model diameter—reflects the fact that long cylindrical tools have large diameters that would make the standard threshold too forgiving; adopting this metric more broadly could change rankings on existing pose benchmarks.
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

3 major / 7 minor

Summary. The paper reports on the SurgRIPE challenge held at MICCAI 2023, whose goal is to provide a benchmark for markerless 6DoF pose estimation of da Vinci surgical instrument wrists. The authors introduce a dataset of 2,841 annotated frames (plus larger raw video sets) for two instruments, Large Needle Driver and Maryland Bipolar Forceps, under conditions with and without occlusion. Ground-truth poses are obtained by attaching a keydot marker to the instrument wrist, estimating the marker pose with PnP, and manually refining the marker-to-instrument transformation in a GUI until the projected CAD model aligns with the image; the marker is then removed with LaMa inpainting, and segmentation masks are rendered from the CAD model. The paper describes six challenge submissions and a PVNet baseline, evaluates them with ADD, average accuracy (0–5 mm), translation/rotation error, proj2d, and mmd5, and identifies IGTUM as the winner. The central claims are that the dataset provides realistic surgical video with high-accuracy ground-truth poses and that the resulting benchmark is a reusable evaluation standard for markerless surgical instrument pose estimation.

Significance. If the ground-truth accuracy claim holds, this is a useful community resource: the dataset and evaluation toolkit are publicly released, the annotation pipeline is described in sufficient detail to be reproduced, and the paper includes an independent simulation-based check (0.253 mm translation error, 0.302 deg rotation error on 50 VisionBlender images). The challenge format, with multiple independent method implementations and open-source benchmarking code, is a genuine contribution to a field that lacks a standardized 6DoF surgical instrument pose benchmark. The main value of the paper therefore hinges on whether the released images are actually consistent with the published ground-truth poses; the current validation does not fully establish this, and the ranking claims would be strengthened by uncertainty quantification. If the authors address the validation gap on the released/inpainted data, the manuscript would make a solid contribution to surgical vision benchmarking.

major comments (3)
  1. [§9.2, §2.2.1] The ground-truth accuracy validation is performed on 50 simulated VisionBlender images that still contain the keydot marker (Fig. 15), whereas the released dataset is produced by removing the marker with LaMa inpainting using random-size masks (Section 2.2.1). The simulation therefore validates the PnP marker pose and the hand-adjusted T_M_I, but it does not validate that the inpainted released images still depict the instrument at the pose recorded by the ground truth. Because the marker is attached to the wrist, the exact region whose pose is being annotated, and because random-size masks can cover instrument pixels beyond the marker, inpainting could locally reshape the instrument boundary in the released images. A shift of a few pixels at 960×540 would make the ground-truth pose inconsistent with the image content and would bias every ADD, proj2d, translation, and rotation score in Tables 4–9. Please add a validation on the released data condition, for example by (a) simulating images with markers, running the full pipeline including inpainting, and comparing the estimated poses against known simulation poses, or (b) measuring reprojection alignment of the CAD model on a sample of released inpainted images, or (c) quantifying the geometric change introduced by inpainting (e.g., boundary displacement) and converting it into an expected pose error.
  2. [§7, Tables 4–7, §3.4] The rankings are based on point estimates of ADD and Avg Acc on test sets of only 373, 209, 238, and 387 frames (Table 2), with no confidence intervals, bootstrap estimates, or significance tests. Several differences between methods are small: in Table 5, IGTUM (ADD 0.3876), ICL (0.3684), and MVL_3S (0.3541) are within 0.034; in Table 4, TUDU (0.1314) and EUT (0.1796) are close. The statement in Section 7.2 that IGTUM 'consistently outperformed' the other methods is not supported without a statistical analysis or at least an explicit statement of ranking sensitivity. Please add error bars or pairwise significance tests on ADD and Avg Acc, or otherwise clarify that the reported ordering is a point estimate that may be within the noise level of the test sets.
  3. [§2.2, §9.2] The marker-to-instrument transformation T_M_I is adjusted manually in a GUI until the 3D model projection aligns with the instrument (Fig. 6), and the only independent accuracy check is on 50 simulated images (§9.2). For the real endoscopic frames, no independent assessment is provided for the PnP marker pose or for T_M_I under realistic lighting, specularity, motion blur, and tissue contact. Since all benchmark metrics (ADD, proj2d, translation, rotation) are computed with respect to this ground truth, a systematic bias in T_M_I or a PnP failure mode in real scenes would directly corrupt every method score. Please provide an independent check on a subset of real frames, for example manual keypoint annotation, stereo triangulation from the right camera, or comparison with a second tracking modality, and report error statistics on real data rather than only on rendered simulation images.
minor comments (7)
  1. [§9.2] The validation reports only mean translation error (0.253 mm) and mean rotation error (0.302 deg); please also report standard deviations, medians, and worst-case errors, since a small mean can hide occasional large outliers in the ground-truth annotations.
  2. [§3.1] The BOP metrics are attributed to (Brachmann et al., 2014); the standard BOP challenge protocol is normally associated with the BOP challenge papers (e.g., Hodaň et al.). Please check and correct this citation.
  3. [Table 1] The dataset name 'EndiVisPose' appears to be a typo for 'EndoVisPose'; please verify consistent spelling across the paper.
  4. [Table 2] The instrument name 'Maryland Bipolar Forcep' is missing the plural 'Forceps'.
  5. [Tables 6 and 7] The table headings read 'Best 2 methods. are in bold.'; this should be 'Best two methods are in bold.'
  6. [§7.2] The text says 'It achieved 36.06% ADD on average', while Tables 4–7 report ADD as a fraction between 0 and 1. Please standardize the units and define whether ADD is reported as a fraction or a percentage.
  7. [§10] The Conclusions contain the typo 'markless' instead of 'markerless'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular steps: the GT accuracy is checked against simulator-known poses, and the benchmark evaluation uses fixed released labels independently of the annotation pipeline.

full rationale

The claimed derivation chain is: keydot-marker pose via PnP, hand-calibrated marker-to-instrument transform (T_M_I), marker removal by inpainting, and fixed-label benchmark evaluation. The PnP + GUI alignment in Section 2.2 is an annotation procedure, not a prediction: the GT pose is defined by that procedure rather than derived from the benchmark results. The only quantitative accuracy claim is validated in Section 9.2 against VisionBlender-simulated images with known instrument poses; the reported nonzero errors (0.253 mm translation, 0.302 degrees rotation) show the estimate is not forced to match the reference by construction. The benchmark evaluation itself is external: participant methods are scored on held-out test frames against released labels with predefined metrics, and no method output feeds back into label definition. The main caveats are validation gaps, not circularity: the Section 9.2 simulation still contains the keydot marker and does not validate the inpainting step described in Section 2.2.1, and the real-data T_M_I calibration relies on visual projection alignment without independent verification. These affect the trustworthiness of the labels but do not make any derivation reduce to its own inputs. The only self-citations (VisionBlender, keydot marker pattern) are used as tools with independent simulated reference poses, not as unverified authoritative premises, so the circularity score is 0.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The benchmark's validity rests on a small set of unverified assumptions about marker pose accuracy, CAD model fidelity, inpainting cleanliness, manual alignment consistency, and the representativeness of simulation for validation. The T_M_I transform and the 0-5 mm Avg Acc threshold are hand-chosen quantities that directly affect the reported rankings.

free parameters (2)
  • T_M_I marker-to-instrument transformation = manually adjusted per instrument (no numeric value reported)
    Transformation from keydot marker to instrument joint, tuned in a GUI until 3D model projection overlays the instrument in every frame (Section 2.2). All GT poses are derived from this hand-set calibration.
  • Avg Acc threshold range 0-5 mm = 0 to 5 mm (chosen by authors)
    The evaluation metric averages ADD accuracy over thresholds from 0 to 5 mm, a task-specific choice because 10% of model diameter (~40 mm) is too permissive. This threshold range affects rankings.
assumptions (5)
  • domain assumption PnP pose estimation from keydot marker is accurate enough without real-image verification
    Section 2.2 estimates the marker pose via PnP (Lepetit et al., 2009); all GT poses inherit any bias from this solver, and no independent real-image check is reported.
  • domain assumption The dVRK CAD models match the physical instruments closely enough for alignment and segmentation
    Section 2.1 uses dVRK 3D models; no quantitative fit between model and physical tool is reported.
  • domain assumption Image inpainting removes the keydot marker without leaving consistent visual cues
    Section 2.2.1 uses random mask sizes, but no experiment shows that a pose estimator cannot recover marker remnants.
  • domain assumption Manual GUI alignment of T_M_I is consistent across frames and operators
    Section 2.2 relies on manual adjustment until projections align; no inter-operator or intra-operator variability is reported.
  • domain assumption VisionBlender simulation images are representative of real endoscopic imaging for validating GT accuracy
    Section 9.2 validates the GT pipeline on 50 simulated samples using the same camera intrinsics, assuming simulation matches real image formation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SurgRIPE challenge: Benchmark of Surgical Robot Instrument Pose Estimation." pith.science (2026). https://pith.science/paper/YINQUYFA

@misc{pith2026250102990,
  author       = {Pith},
  title        = {Pith review of: SurgRIPE challenge: Benchmark of Surgical Robot Instrument Pose Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YINQUYFA}},
  note         = {Machine review of arXiv:2501.02990}
}
read the original abstract

Accurate instrument pose estimation is a crucial step towards the future of robotic surgery, enabling applications such as autonomous surgical task execution. Vision-based methods for surgical instrument pose estimation provide a practical approach to tool tracking, but they often require markers to be attached to the instruments. Recently, more research has focused on the development of marker-less methods based on deep learning. However, acquiring realistic surgical data, with ground truth instrument poses, required for deep learning training, is challenging. To address the issues in surgical instrument pose estimation, we introduce the Surgical Robot Instrument Pose Estimation (SurgRIPE) challenge, hosted at the 26th International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) in 2023. The objectives of this challenge are: (1) to provide the surgical vision community with realistic surgical video data paired with ground truth instrument poses, and (2) to establish a benchmark for evaluating markerless pose estimation methods. The challenge led to the development of several novel algorithms that showcased improved accuracy and robustness over existing methods. The performance evaluation study on the SurgRIPE dataset highlights the potential of these advanced algorithms to be integrated into robotic surgery systems, paving the way for more precise and autonomous surgical procedures. The SurgRIPE challenge has successfully established a new benchmark for the field, encouraging further research and development in surgical robot instrument pose estimation.

Figures

Figures reproduced from arXiv: 2501.02990 by the authors.

Figure 1
Figure 1. SurgRIPE data collection pipeline. The GT pose was captured with a keydot [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 3
Figure 3. Occlusion Image Sample [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. (Left) The raw RGB image. (Middle) Projection of the 3D model onto the raw [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figures from the paper (12 more)
Figure 5
Figure 5. Figure 5: Projection of the 3D model onto the 2D image. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: GUI for 3D Pose Alignment. 2.2.1. Image inpainting Since in a real surgical scenario, the keydot marker would not be present, we used image inpainting (Suvorov et al., 2021) to remove the keydot marker from the images used to train and test the compared models, as show…
Figure 1
Figure 1. Figure 1: When inpainting was applied, we used random size masks to ensure [PITH_FULL_IMAGE:figures/full_fig_p009_1.png]
Figure 7
Figure 7. Figure 7: TUDU architecture diagram. The method includes (1) The segmentation branch [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: Visualisation of Multi Losses in UOL Method. [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]
Figure 9
Figure 9. Figure 9: Accuracy–ADD threshold Curves for test data in Testsets Without Occlusion. [PITH_FULL_IMAGE:figures/full_fig_p022_9.png]
Figure 10
Figure 10. Figure 10: Accuracy–ADD threshold Curves for test data in Testsets With Occlusion. [PITH_FULL_IMAGE:figures/full_fig_p024_10.png]
Figure 11
Figure 11. Figure 11: Trajectory comparison in LND Test Without Occlusion. [PITH_FULL_IMAGE:figures/full_fig_p027_11.png]
Figure 12
Figure 12. Figure 12: Trajectory comparison in MBF Test Without Occlusion. [PITH_FULL_IMAGE:figures/full_fig_p028_12.png]
Figure 13
Figure 13. Figure 13: Trajectory comparison in LND Test With Occlusion. [PITH_FULL_IMAGE:figures/full_fig_p029_13.png]
Figure 14
Figure 14. Figure 14: Trajectory comparison in MBF Test With Occlusion. [PITH_FULL_IMAGE:figures/full_fig_p030_14.png]
Figure 15
Figure 15. Figure 15: Simulation sample images. 9.3. Background Scene Simulation The background scene was simulated with a realistic high-fidelity ab￾dominal phantom in our experiments. To create diversity of the background conditions, video data was captured from different areas of the ph…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 19 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  3. [3]

    , author Kondo, S

    author Allan, M. , author Kondo, S. , author Bodenstedt, S. , author Leger, S. , author Kadkhodamohammadi, R. , author Luengo, I. , author Fuentes - Hurtado, F. , author Flouty, E. , author Mohammed, A.K. , author Pedersen, M. , author Kori, A. , author Varghese, A. , author Krishnamurthi, G. , author Rauber, D. , author Mendel, R. , author Palm, C. , aut...

    arXiv 2020

  4. [4]

    , author Ourselin, S

    author Allan, M. , author Ourselin, S. , author Hawkes, D.J. , author Kelly, J.D. , author Stoyanov, D. , year 2018 . title 3-d pose estimation of articulated instruments in robotic minimally invasive surgery . journal IEEE Transactions on Medical Imaging volume 37 , pages 1204--1213 . :10.1109/TMI.2018.2794439

    arXiv 2018

  5. [5]

    , author Shvets, A

    author Allan, M. , author Shvets, A. , author Kurmann, T. , author Zhang, Z. , author Duggal, R. , author Su, Y. , author Rieke, N. , author Laina, I. , author Kalavakonda, N. , author Bodenstedt, S. , author Garc \' a - Peraza, L.C. , author Li, W. , author Iglovikov, V. , author Luo, H. , author Yang, J. , author Stoyanov, D. , author Maier - Hein, L. ,...

    arXiv 2019

  6. [6]

    , author Krull, A

    author Brachmann, E. , author Krull, A. , author Michel, F. , author Gumhold, S. , author Shotton, J. , author Rother, C. , year 2014 . title Learning 6d object pose estimation using 3d object coordinates , pp. pages 536--551 . :10.1007/978-3-319-10605-2_35

    doi:10.1007/978-3-319-10605-2_35 2014

  7. [7]

    , year 2000

    author Bradski, G. , year 2000 . title The OpenCV Library . journal Dr. Dobb's Journal of Software Tools

    work page 2000

  8. [8]

    , author Vetter, M

    author Bukschat, Y. , author Vetter, M. , year 2020 . title Efficientpose: An efficient, accurate and scalable end-to-end 6d multi object pose estimation approach . arXiv:2011.04307 http://arxiv.org/abs/2011.04307

    arXiv 2020

Show all 35 references
  1. [9]

    , author Tukra, S

    author Cartucho, J. , author Tukra, S. , author Li, Y. , author S. Elson, D. , author Giannarou, S. , year 2020 . title Visionblender: a tool to efficiently generate computer vision datasets for robotic surgery . journal Computer Methods in Biomechanics and Biomedical Engineer...

    2020

  2. [10]

    , author Wang, C

    author Cartucho, J. , author Wang, C. , author Huang, B. , author S. Elson, D. , author Dariz, A. , author Giannarou, S. , year 2021 . title An enhanced marker pattern that achieves improved accuracy in surgical tool tracking . journal Computer Methods in Biomechanics and Biom...

    2021

  3. [11]

    , author Zhu, Y

    author Chen, L.C. , author Zhu, Y. , author Papandreou, G. , author Schroff, F. , author Adam, H. , year 2018 . title Encoder-decoder with atrous separable convolution for semantic image segmentation , in: booktitle ECCV

    2018

  4. [12]

    , author Cartucho, J

    author Cui, Z. , author Cartucho, J. , author Giannarou, S. , author Rodriguez y Baena, F. , year 2023 . title Caveats on the first-generation da vinci research kit: Latent technical constraints and essential calibrations . journal IEEE Robotics & Automation Magazine , pages 2...

    2023

  5. [13]

    , author Kurmann, T

    author Du, X. , author Kurmann, T. , author Chang, P.L. , author Allan, M. , author Ourselin, S. , author Sznitman, R. , author Kelly, J.D. , author Stoyanov, D. , year 2018 . title Articulated multi-instrument 2-d pose estimation using fully convolutional networks . journal I...

    2018

  6. [14]

    , author Antonino, M

    author Federico, B. , author Antonino, M. , author Erfan, S.B. , author Arianna, M. , author Alberto, A. , author Anastasios, K. , author Danail, S. , author Paolo, D. , author Gastone, C. , year 2019 . title Localization strategies for robotic endoscopic capsules: a review . ...

    2019

  7. [15]

    , author Vedula, S.S

    author Gao, Y. , author Vedula, S.S. , author Reiley, C.E. , author Ahmidi, N. , author Varadarajan, B. , author Lin, H.C. , author Tao, L. , author Zappella, L. , author B 'e jar, B. , author Yuh, D.D. , et al., year 2014 . title Jhu-isi gesture and skill assessment working s...

    2014

  8. [16]

    , author Buch, A.G

    author Haugaard, R.L. , author Buch, A.G. , year 2022 . title Surfemb: Dense and continuous correspondence distributions for object pose estimation with learnt surface embeddings , in: booktitle 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pp. p...

    2022

  9. [17]

    , author Zhang, X

    author He, K. , author Zhang, X. , author Ren, S. , author Sun, J. , year 2016 . title Deep residual learning for image recognition , in: booktitle 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pp. pages 770--778 . :10.1109/CVPR.2016.90

    2016 doi

  10. [18]

    , author Chen, Z

    author Kazanzides, P. , author Chen, Z. , author Deguet, A. , author Fischer, G.S. , author Taylor, R.H. , author DiMaio, S.P. , year 2014 . title An open-source research kit for the da vinci® surgical system , in: booktitle 2014 IEEE International Conference on Robotics and A...

    2014

  11. [19]

    , author Roumeliotis, S.I

    author Ke, T. , author Roumeliotis, S.I. , year 2017 . title An efficient algebraic solution to the perspective-three-point problem , in: booktitle 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pp. pages 4618--4626 . :10.1109/CVPR.2017.491

    2017 doi

  12. [20]

    , author Ragot, M

    author Lecuyer, G. , author Ragot, M. , author Martin, N. , author Launay, L. , author Jannin, P. , year 2020 . title Assisted phase and step annotation for surgical videos . journal International Journal of Computer Assisted Radiology and Surgery volume 15 . :10.1007/s11548-0...

    2020 doi

  13. [21]

    , author Moreno-Noguer, F

    author Lepetit, V. , author Moreno-Noguer, F. , author Fua, P. , year 2009 . title Epnp: An accurate o(n) solution to the pnp problem . journal International Journal of Computer Vision volume 81

    2009

  14. [22]

    , author Richter, F

    author Li, Y. , author Richter, F. , author Lu, J. , author Funk, E.K. , author Orosco, R.K. , author Zhu, J. , author Yip, M.C. , year 2019 . title Super: A surgical perception framework for endoscopic tissue manipulation with surgical robotics . journal CoRR volume abs/1909....

    2019 arXiv

  15. [23]

    , author Maire, M

    author Lin, T. , author Maire, M. , author Belongie, S.J. , author Bourdev, L.D. , author Girshick, R.B. , author Hays, J. , author Perona, P. , author Ramanan, D. , author Doll ' a r, P. , author Zitnick, C.L. , year 2014 . title Microsoft COCO: common objects in context . jo...

    2014 arXiv

  16. [24]

    , author Mao, H

    author Liu, Z. , author Mao, H. , author Wu, C.Y. , author Feichtenhofer, C. , author Darrell, T. , author Xie, S. , year 2022 . title A convnet for the 2020s . journal Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2022

  17. [25]

    , author Reinke, A

    author Maier-Hein, L. , author Reinke, A. , author Kozubek, M. , author Martel, A.L. , author Arbel, T. , author Eisenmann, M. , author Hanbury, A. , author Jannin, P. , author Müller, H. , author Onogur, S. , author Saez-Rodriguez, J. , author van Ginneken , B. , author Kopp-...

    2020

  18. [26]

    , author Liu, Y

    author Peng, S. , author Liu, Y. , author Huang, Q. , author Zhou, X. , author Bao, H. , year 2019 . title Pvnet: Pixel-wise voting network for 6dof pose estimation , in: booktitle CVPR

    2019

  19. [27]

    , author Reinke, A

    author Roß, T. , author Reinke, A. , author Full, P.M. , author Wagner, M. , author Kenngott, H. , author Apitz, M. , author Hempe, H. , author Mindroc-Filimon, D. , author Scholz, P. , author Tran, T.N. , author Bruno, P. , author Arbeláez, P. , author Bian, G.B. , author Bod...

    2021

  20. [28]

    , author Maaz, M

    author Shaker, A. , author Maaz, M. , author Rasheed, H. , author Khan, S. , author Yang, M.H. , author Khan, F.S. , year 2023 . title Swiftformer: Efficient additive attention for transformer-based real-time mobile vision applications , in: booktitle Proceedings of the IEEE/C...

    2023

  21. [29]

    , author Martel, J.N.P

    author Sitzmann, V. , author Martel, J.N.P. , author Bergman, A.W. , author Lindell, D.B. , author Wetzstein, G. , year 2020 . title Implicit neural representations with periodic activation functions , in: booktitle Proceedings of the 34th International Conference on Neural In...

    2020

  22. [30]

    , author Logacheva, E

    author Suvorov, R. , author Logacheva, E. , author Mashikhin, A. , author Remizova, A. , author Ashukha, A. , author Silvestrov, A. , author Kong, N. , author Goka, H. , author Park, K. , author Lempitsky, V. , year 2021 . title Resolution-robust large mask inpainting with fou...

    2021 arXiv

  23. [31]

    title Yolov5

    author Ultralytics , year 2023 . title Yolov5 . howpublished https://github.com/ultralytics/yolov5 . note Accessed: January 3, 2025

    2023

  24. [32]

    , author Cartucho, J

    author Wang, C. , author Cartucho, J. , author Elson, D. , author Darzi, A. , author Giannarou, S. , year 2022 . title Towards autonomous control of surgical instruments using adaptive-fusion tracking and robot self-calibration , in: booktitle 2022 IEEE/RSJ International Confe...

    2022

  25. [33]

    , author Li, X

    author Wang, Z. , author Li, X. , author Navarro-Alarcon, D. , author Liu, Y.h. , year 2018 . title A unified controller for region-reaching and deforming of soft objects , in: booktitle 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pp. page...

    2018

  26. [34]

    , author Yang, W

    author Wen, B. , author Yang, W. , author Kautz, J. , author Birchfield, S.T. , year 2023 . title Foundationpose: Unified 6d pose estimation and tracking of novel objects . journal 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages 17868--17879 ...

    2023

  27. [35]

    , author Schmidt, T

    author Xiang, Y. , author Schmidt, T. , author Narayanan, V. , author Fox, D. , year 2018 . title PoseCNN : A convolutional neural network for 6D object pose estimation in cluttered scenes , in: booktitle Robotics: Science and Systems (RSS)

    2018

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.