Pith. sign in

REVIEW 4 major objections 6 minor 52 references

Landmark-Free Preoperative-to-Intraoperative Registration in Laparoscopic Liver Resection

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A landmark-free pipeline registers preoperative 3D liver models to laparoscopic video by treating the task as 3D-to-3D point cloud alignment.

desk verdict A genuinely useful dataset and a plausible landmark-free pipeline, but the main 3D metric is the training loss and the PCA scale recovery is unvalidated; conditional at best. read the letter →

arxiv 2504.15152 v1 pith:G4DE6FCH submitted 2025-04-21 cs.CV cs.AI

Jun Zhou , Bingchen Gao , Kai Wang , Jialun Pei , Pheng-Ann Heng , Jing Qin This is my paper · ORCID
classification cs.CVcs.AI
keywords preoperative-to-intraoperativeregistrationlaparoscopicliverresection3D-3Dpointcloudself-supervisedlearningfeature-disentangledtransformerstructure-regularizedshapeadaptationaugmentedrealitysurgeryP2I-LRegdataset
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

This paper claims that preoperative-to-intraoperative registration in laparoscopic liver resection can be done without anatomical landmarks by re-framing the problem as a 3D-to-3D point cloud alignment. The pipeline reconstructs a partial intraoperative liver point cloud from a monocular laparoscopic frame using a depth estimator and a liver mask, scales it to match the preoperative CT-derived model via principal component analysis on the principal axis, then registers the two clouds in two stages: a rigid stage learned by a feature-disentangled transformer and a non-rigid stage that deforms the preoperative model with a structure-regularized shape-adaptation network. Training is self-supervised: a differentiable renderer projects the deformed model back to 2D and enforces consistency with the observed mask and point cloud. On a new in-vivo dataset of 21 patients, the method reports higher Dice, 78.89% versus 75.24%, and lower Chamfer distance, 2.97 mm versus 3.20 mm, than prior point-cloud registration baselines. If the claim holds, it removes the need for manual landmark identification and provides a fully automatic augmented-reality overlay during surgery.

What carries the argument

The machinery has four parts. (1) Scale-consistent recovery: a monocular depth map and a segmented liver mask are reprojected to a partial point cloud, then PCA on the principal axis computes an isotropic scale factor that matches the intraoperative cloud to the preoperative model, compensating for the unknown scale of monocular depth. (2) Feature Disentangled Transformer (FDT): weight-sharing point-cloud backbones extract downsampled keypoints and geometric features; rotary positional encoding injects explicit relative geometry, and bidirectional cross-attention exchanges semantic cues before a confidence matrix selects top-k inlier correspondences for a weighted $\mathrm{SE}(3)$ solution. (3) Structure-Regularized Shape Adaptation (SRSA): a deformation decomposition pyramid represents the warp as a sequence of per-point $\mathrm{SE}(3)$ updates with confidence scores, while a low-rank structure-similarity network computes a deformation weight matrix from the inner products of source and target features; this matrix is injected at the highest-frequency level to tell the network where to deform. (4) Self-supervision: a differentiable renderer projects the deformed model to 2D and penalizes mask mismatches and Chamfer distance to the reconstructed intraoperative point cloud. Together these parts turn an ill-posed 3D-2D landmark problem into two supervised 3D-3D subproblems.

What would settle it

On P2I-LReg keyframes, compare the PCA-derived scale factor with an independent scale measurement, for example from a surgical instrument of known size visible in the same frame or from a stereo reconstruction; if the independent scale differs by more than a few percent on low-overlap frames, the scale-consistency step is the bottleneck. A second direct check is to measure the angle between the principal axis of the visible patch and the principal axis of the full preoperative liver model; frames with large angles should be the ones where registration fails, and if they are not, the PCA assumption is not doing the claimed work.

Watch

Extended reading notes

Core claim

The paper's central claim is that the conventional 3D-2D landmark workflow can be replaced by a landmark-free 3D-3D registration pipeline, and that decoupling this pipeline into rigid and non-rigid subtasks yields more accurate overlays than existing rigid and deformable point-cloud registration methods. Self-P2IR first learns reliable point correspondences through a feature-disentangled transformer that combines rotary positional encoding with self- and cross-attention, solves the $\mathrm{SE}(3)$ transform from top-k correspondences, and then applies a structure-regularized deformation network that models global geometry similarity through a low-rank structure-similarity learning module and deforms the preoperative model through a deformation decomposition pyramid. Supervision comes from differentiable rendering of the deformed model against the intraoperative mask and reconstructed point cloud. On the contributed P2I-LReg dataset and on synthetic data, Self-P2IR reports the best rigid registration recall, inlier ratio, and feature matching recall, and improves the best non-rigid baseline by 3.65 Dice points and 0.23 mm Chamfer distance, with expert-surgeon ratings in its favor.

Load-bearing premise

The load-bearing premise is that the principal axis of the partially visible intraoperative point cloud—computed by PCA from monocular depth and a mask that may cover only 15–21% of the liver—gives a reliable global scale factor for the whole preoperative model; if that scale factor is wrong, the rigid initialization, the deformation field, and the final overlay all inherit the error.

Editorial extensions

If this is right

  • No anatomical landmark detection is needed: registration is driven by the liver mask, monocular depth, and the 3D models, so ambiguous landmarks such as ridges and silhouettes no longer gate the overlay.
  • The method tolerates limited visibility: on synthetic data it keeps registration recall near 93% even at 11.5–20.8% surface overlap, which matters because laparoscopy exposes only part of the liver.
  • Perception modules are swappable: replacing the segmentation mask or the monocular depth estimator with alternative models changes Dice by only a few points, so the framework does not depend on one specific component.
  • The P2I-LReg dataset, with 346 keyframes, liver masks, camera intrinsics, and reconstructed point clouds, provides an in-vivo benchmark for evaluating 3D-2D registration in laparoscopic liver surgery.
  • The self-supervised loss via differentiable rendering means no ground-truth deformation fields are needed for the non-rigid stage; the intraoperative mask and point cloud provide supervision directly.

Reading between the lines

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

  • Editorial extension: the same rigid/non-rigid decoupling could transfer to other deformable laparoscopic targets, such as kidney or uterus, whenever a monocular depth prior and a target mask are available; the paper evaluates only the liver, but the pipeline itself is not liver-specific beyond training data.
  • Editorial extension: the PCA scale step is the most fragile part of the pipeline in clinical use, so an end-to-end learned scale estimator, or a scale prior from a known-size instrument, is a natural next step that the paper leaves to future work.
  • Editorial extension: the paper fits visible liver portions precisely but treats hidden regions with a relatively static deformation; a testable follow-up would quantify how well the overlay tracks the margin between visible and hidden anatomy, since that margin is where surgeons would most need to trust the AR overlay.
  • Editorial extension: the reported gains come from the full pipeline; ablating the feature-disentangled transformer against a standard cross-attention baseline, while keeping the rigid and non-rigid stages otherwise identical, would isolate how much of the improvement is due to the disentangled representation.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 Self-P2IR, a landmark-free registration framework for laparoscopic liver resection. The method converts the conventional 3D-2D registration problem into a 3D-3D partial-to-global point cloud registration task, decoupling it into a rigid registration stage using a feature-disentangled transformer and a non-rigid deformation stage using structure-regularized shape adaptation (SRSA). Training is self-supervised via 2D liver masks, a differentiable renderer, and a monocular-depth-based intraoperative point cloud target. The authors also introduce P2I-LReg, a new in-vivo dataset of 346 keyframes from 21 patients with liver masks and calibrated camera intrinsics. Experiments on synthetic and in-vivo data report improvements over several state-of-the-art rigid and non-rigid registration methods in terms of Dice, Chamfer distance, and a user study.

Significance. If the reported results hold, the paper makes a useful contribution to intraoperative AR for liver surgery: it removes the landmark-annotation bottleneck, provides a new in-vivo dataset, and demonstrates a practical monocular pipeline. The 2D Dice evaluation is independent of the training objective and consistently favors the proposed method, and the synthetic rigid-registration study is extensive, covering multiple estimators, noise levels, and overlap ratios. The code and dataset are promised publicly. However, the 3D evaluation is weakened by two load-bearing issues: the Chamfer distance metric coincides with the training loss, and the intraoperative target cloud is generated by an unvalidated monocular-depth plus PCA-scale procedure. These issues must be addressed before the claimed 3D registration superiority is established.

major comments (4)
  1. [Sec. IV-C, Eq. (2), and Tab. V] The PCA-based scale factor σ = σ_S / σ_To is never validated against a known ground-truth scale. The paper's own overlap analysis (Figs. 15 and 16) shows in-vivo surface overlap as low as 15-21%, where the principal axis of the visible patch can be a chord of the visible region rather than the organ's long axis. Since Tab. V shows that removing scale consistency degrades CD from 2.97 mm to 10.66 mm, any bias in σ propagates into the rigid initialization, the deformation field, and the target point cloud used for both training and evaluation. Please add a validation of σ on synthetic data with known ground-truth scale and on in-vivo data with stereo or other ground-truth depth, and report the error of the recovered scale factor.
  2. [Sec. IV-E, Eq. (14), and Tab. II] The Chamfer distance used as the headline 3D metric is the same L_geo loss minimized during training, and the target point cloud T_o is produced by the method's own depth-estimation and PCA-scale pipeline. Consequently, the CD comparison in Tab. II measures, at least in part, fit to a self-generated target rather than registration accuracy against an independent ground truth. Please report an independent 3D metric (e.g., against synthetic data with known deformation, phantom data, or stereo reconstructions), or clearly demote CD to a diagnostic and base the 3D accuracy claim on the independent 2D Dice result.
  3. [Fig. 12 and Tab. II] The ablation in Fig. 12 reports CD values near 1.7 mm for all pyramid levels, whereas Tab. II reports 2.97 mm for the full model. Please state whether Fig. 12 uses a different fold, a different test subset, or a different point-cloud sampling, and reconcile the numbers; as written, the ablation cannot be interpreted without this information.
  4. [Sec. V-A and Tab. I] The rigid-registration results in Tab. I are reported as point estimates without error bars or statistical significance. Since the claimed superiority over GeoTrans and Predator is modest on some metrics (e.g., FMR 96.9 vs 96.7), please provide standard deviations or confidence intervals across data splits or training seeds, and clarify whether the synthetic test set is patient-disjoint from the training set.
minor comments (6)
  1. [Fig. 1 caption] The Fig. 1 caption contains an author instruction in Chinese ('Synthetic and liver model order swap'); please remove this editorial artifact.
  2. [Tab. III] The symbols denoting the frozen and unfrozen strategies for Φ appear corrupted in the table; please use unambiguous labels such as 'frozen' and 'unfrozen'.
  3. [Sec. IV-C] The text around Eq. (2) should clarify that eigenvalues and eigenvectors are computed from the covariance matrix, and that σ_S and σ_T_o are the projected extents along the first principal eigenvector; the current wording is ambiguous.
  4. [Abstract and Sec. III] The dataset contribution is described as including 'corresponding 3D point clouds,' but these point clouds are generated by monocular depth estimation and PCA scale recovery rather than by a validated reconstruction method; please state this clearly in the dataset description.
  5. [Fig. 9 and Fig. 11] The abbreviation 'PSR' is used in figures before being defined; define PointSetReg at first use in the text.
  6. [Abstract] The statement 'The code and dataset are available at Self-P2IR' should include a working URL or repository identifier.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; CD/loss overlap is a standard task metric, and independent Dice and user-study evaluations support the central claim.

full rationale

The claimed derivation is not circular. The rigid registration network is trained and evaluated on synthetic data with known ground-truth poses (Sec. V-C.1, Tab. I), providing an external benchmark that does not depend on the paper's own reconstruction. For the in-vivo non-rigid evaluation, the two headline metrics are not merely the training objective: Dice compares the projected deformed model against manually annotated liver masks (an independent 2D label), and the user study (Fig. 11) provides six expert surgeons' ratings. The Chamfer distance in Tab. II is indeed the same bidirectional distance as Lgeo in Eq. (14), which is minimized during training, but this is the standard choice of a task metric rather than a circular prediction: five-fold cross-validation at the patient level keeps test cases out of training, so the reported CD measures generalization to held-out data. The PCA scale-consistency step (Sec. IV-C) fixes the unknown monocular depth scale using the preoperative model's principal-axis extent; this is a geometric normalization, not a fitted prediction, and the Discussion explicitly flags it as an offline assumption. Self-citations appear only in related-work contexts (e.g., refs. [10], [14], [36]) and do not carry the load of any derivation or uniqueness argument. No equation in the paper reduces a reported result to an input by construction.

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

The framework is an empirical learning system, so its central claim rests less on mathematical axioms than on assumptions about the reliability of the video-derived point cloud, the transferability of pretrained monocular depth, and the sufficiency of mask and chamfer self-supervision. The free parameters listed above are the hand-set hyperparameters the reported results depend on. No new physical entities are introduced; the proposed modules are algorithmic.

free parameters (5)
  • Correspondence confidence threshold eta_H = 0.15
    Selected by hand in Sec. V-A. Determines which point correspondences are used in the SVD pose solver and therefore controls rigid registration quality.
  • Loss weights lambda_dice, lambda_mask, lambda_geo, lambda_tran = All set to 1; lambda_tran = 1
    Chosen by hand in Sec. V-A and ablated in Tab. IV. Balances mask, geometric consistency, and transformation losses, and the non-rigid result depends on this balance.
  • Pyramid levels L = 10
    Set to 10 following NDP in Sec. IV-D.2. Deformation field capacity depends on the number of frequency levels.
  • Voxel size for downsampling = 0.001 m
    Fixed preprocessing choice in Sec. V-A. Affects point density and feature extraction.
  • Input point count = 8192
    Fixed in Sec. V-A and ablated in Fig. 13. Performance saturates above 2k points, but all reported numbers use 8192.
assumptions (5)
  • domain assumption Monocular depth estimation from Depth Anything V1 transfers to laparoscopic liver images and provides correct relative depth within the masked liver region.
    Invoked in Sec. IV-C Eq. (1) to reconstruct the intraoperative point cloud To. Never validated against ground-truth depth or independent 3D geometry.
  • domain assumption PCA-based scale consistency computes a globally correct scale factor from the partial visible liver point cloud.
    Sec. IV-C Eq. (2) assumes the principal axis of a partially visible liver surface matches that of the full preoperative model. Tab. V shows large performance drops when scale handling changes, so the result hinges on this assumption.
  • domain assumption The deformation pyramid in Eq. (10) can represent real physiological liver deformation from pneumoperitoneum and tool manipulation.
    Borrowed from NDP [27]. The paper does not verify against physical deformation ground truth, and the discussion admits only visible portions are fitted accurately.
  • domain assumption Self-supervision through rendered 2D masks and chamfer distance to the reconstructed point cloud (Eqs. 13-14) is a sufficient learning signal for non-rigid registration.
    The network is trained to match To, which is also the evaluation target for the CD metric, creating partial circularity.
  • domain assumption Manual liver masks and calibrated camera intrinsics are accurate ground truth for training and evaluation.
    Sec. III: four surgeons labeled and two verified, but no inter-observer variability or annotation error analysis is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Landmark-Free Preoperative-to-Intraoperative Registration in Laparoscopic Liver Resection." pith.science (2026). https://pith.science/paper/G4DE6FCH

@misc{pith2026250415152,
  author       = {Pith},
  title        = {Pith review of: Landmark-Free Preoperative-to-Intraoperative Registration in Laparoscopic Liver Resection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G4DE6FCH}},
  note         = {Machine review of arXiv:2504.15152}
}
read the original abstract

Liver registration by overlaying preoperative 3D models onto intraoperative 2D frames can assist surgeons in perceiving the spatial anatomy of the liver clearly for a higher surgical success rate. Existing registration methods rely heavily on anatomical landmark-based workflows, which encounter two major limitations: 1) ambiguous landmark definitions fail to provide efficient markers for registration; 2) insufficient integration of intraoperative liver visual information in shape deformation modeling. To address these challenges, in this paper, we propose a landmark-free preoperative-to-intraoperative registration framework utilizing effective self-supervised learning, termed \ourmodel. This framework transforms the conventional 3D-2D workflow into a 3D-3D registration pipeline, which is then decoupled into rigid and non-rigid registration subtasks. \ourmodel~first introduces a feature-disentangled transformer to learn robust correspondences for recovering rigid transformations. Further, a structure-regularized deformation network is designed to adjust the preoperative model to align with the intraoperative liver surface. This network captures structural correlations through geometry similarity modeling in a low-rank transformer network. To facilitate the validation of the registration performance, we also construct an in-vivo registration dataset containing liver resection videos of 21 patients, called \emph{P2I-LReg}, which contains 346 keyframes that provide a global view of the liver together with liver mask annotations and calibrated camera intrinsic parameters. Extensive experiments and user studies on both synthetic and in-vivo datasets demonstrate the superiority and potential clinical applicability of our method.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 44 canonical work pages

  1. [1]

    The value of augmented reality in surgery—a usability study on laparoscopic liver surgery,

    J. Ramalhinho, S. Yoo, T. Dowrick, B. Koo, M. Somasundaram, K. Gurusamy, D. J. Hawkes, B. Davidson, A. Blandford, and M. J. Clarkson, “The value of augmented reality in surgery—a usability study on laparoscopic liver surgery,” MedIA, vol. 90, p. 102943, 2023

    work page 2023

  2. [2]

    Surgical workflow recognition and blocking effectiveness detection in laparoscopic liver resection with pringle maneuver,

    D. Guo, W. Si, Z. Li, J. Pei, and P.-A. Heng, “Surgical workflow recognition and blocking effectiveness detection in laparoscopic liver resection with pringle maneuver,” in AAAI, vol. 39, no. 3, 2025, pp. 3220–3228

    work page 2025

  3. [3]

    Automatic, global registration in laparoscopic liver surgery,

    B. Koo, M. R. Robu, M. Allam, M. Pfeiffer, S. Thompson, K. Gu- rusamy, B. Davidson, S. Speidel, D. Hawkes et al., “Automatic, global registration in laparoscopic liver surgery,” IJCARS, pp. 1–10, 2022

    work page 2022

  4. [4]

    Neural patient-specific 3d–2d registration in laparoscopic liver resection,

    I. Mhiri, D. Pizarro, and A. Bartoli, “Neural patient-specific 3d–2d registration in laparoscopic liver resection,” IJCARS, pp. 1–8, 2024

    work page 2024

  5. [5]

    Evaluation of a novel navigation platform for laparoscopic liver surgery with organ deformation compensation using injected fiducials,

    E. Pelanis, A. Teatini, B. Eigl, A. Regensburger, A. Alzaga, R. P. Kumar, T. Rudolph, D. L. Aghayan, C. Riediger, N. Kvarnstr ¨om et al., “Evaluation of a novel navigation platform for laparoscopic liver surgery with organ deformation compensation using injected fiducials,” MedIA, vol. 69, p. 101946, 2021

    work page 2021

  6. [6]

    Preoperative liver registration for augmented monocular laparoscopy using backward–forward biomechan- ical simulation,

    E. ¨Ozg¨ur, B. Koo, B. Le Roy et al., “Preoperative liver registration for augmented monocular laparoscopy using backward–forward biomechan- ical simulation,” IJCARS, vol. 13, pp. 1629–1640, 2018

    work page 2018

  7. [7]

    Intraoperative laparoscope augmentation for port placement and resection planning in minimally invasive liver resection,

    M. Feuerstein, T. Mussack, S. M. Heining, and N. Navab, “Intraoperative laparoscope augmentation for port placement and resection planning in minimally invasive liver resection,” IEEE Transactions on Medical Imaging, vol. 27, no. 3, pp. 355–369, 2008

    2008

  8. [8]

    Deformable registration of a preoperative 3d liver volume to a laparoscopy image using contour and shading cues,

    B. Koo, E. ¨Ozg¨ur, B. Le Roy, E. Buc, and A. Bartoli, “Deformable registration of a preoperative 3d liver volume to a laparoscopy image using contour and shading cues,” in MICCAI, 2017, pp. 326–334

    work page 2017

Show all 52 references
  1. [9]

    Using multiple images and contours for deformable 3d–2d registration of a preoperative ct in laparoscopic liver surgery,

    Y . Espinel, L. Calvet, K. Botros, E. Buc, C. Tilmant, and A. Bartoli, “Using multiple images and contours for deformable 3d–2d registration of a preoperative ct in laparoscopic liver surgery,” IJCARS, vol. 17, no. 12, pp. 2211–2219, 2022

    2022

  2. [10]

    Depth-driven geometric prompt learning for laparoscopic liver landmark detection,

    J. Pei, R. Cui, Y . Li, W. Si, J. Qin, and P.-A. Heng, “Depth-driven geometric prompt learning for laparoscopic liver landmark detection,” in MICCAI, 2024

    2024

  3. [11]

    Automatic 3d/2d deformable registration in minimally invasive liver resection using a mesh recovery network

    M. Labrunie, D. Pizarro, C. Tilmant, and A. Bartoli, “Automatic 3d/2d deformable registration in minimally invasive liver resection using a mesh recovery network.” in MIDL, 2023, pp. 1104–1123

    2023

  4. [12]

    Toward precise osteotomies: a coarse-to-fine 3d cut plane planning method for image-guided pelvis tumor resection surgery,

    Y . Zhang, F. Li, L. Qiu, L. Xu, X. Niu, Y . Sui, S. Zhang, Q. Zhang, and L. Zhang, “Toward precise osteotomies: a coarse-to-fine 3d cut plane planning method for image-guided pelvis tumor resection surgery,”IEEE Transactions on Medical Imaging , vol. 39, no. 5, pp. 1511–1523, 2019

    2019

  5. [13]

    Keyhole-aware laparoscopic augmented reality,

    Y . Espinel, N. Rabbani, T. B. Bui, M. Ribeiro, E. Buc, and A. Bartoli, “Keyhole-aware laparoscopic augmented reality,” MedIA, vol. 94, p. 103161, 2024

    2024

  6. [14]

    Epicardium prompt-guided real-time cardiac ultrasound frame- to-volume registration,

    L. Lei, J. Zhou, J. Pei, B. Zhao, Y . Jin, Y .-C. J. Teoh, J. Qin, and P.-A. Heng, “Epicardium prompt-guided real-time cardiac ultrasound frame- to-volume registration,” in MICCAI, 2024

    2024

  7. [15]

    Non-rigid volume to surface registration using a data-driven biomechanical model,

    M. Pfeiffer, C. Riediger, S. Leger, J.-P. K ¨uhn, D. Seppelt, R.-T. Hoff- mann, J. Weitz et al., “Non-rigid volume to surface registration using a data-driven biomechanical model,” in MICCAI, 2020, pp. 724–734

    2020

  8. [16]

    Augmented reality guided laparoscopic surgery of the uterus,

    T. Collins, D. Pizarro, S. Gasparini, N. Bourdel, P. Chauvet, M. Canis, L. Calvet, and A. Bartoli, “Augmented reality guided laparoscopic surgery of the uterus,” IEEE Transactions on Medical Imaging , vol. 40, no. 1, pp. 371–380, 2020

    2020

  9. [17]

    Video-based soft tissue de- formation tracking for laparoscopic augmented reality-based navigation in kidney surgery,

    E. Wang, Y . Liu, P. Tu, Z. A. Taylor et al., “Video-based soft tissue de- formation tracking for laparoscopic augmented reality-based navigation in kidney surgery,” IEEE Transactions on Medical Imaging , 2024

    2024

  10. [18]

    Point cloud registration in laparoscopic liver surgery using keypoint correspondence registration network,

    Y . Zhang, Y . Zou, and P. X. Liu, “Point cloud registration in laparoscopic liver surgery using keypoint correspondence registration network,” IEEE Transactions on Medical Imaging , 2024

    2024

  11. [19]

    Rotational pro- jection statistics for 3d local surface description and object recognition,

    Y . Guo, F. Sohel, M. Bennamoun, M. Lu, and J. Wan, “Rotational pro- jection statistics for 3d local surface description and object recognition,” IJCV, vol. 105, pp. 63–86, 2013

    2013

  12. [20]

    A symmetric objective function for icp,

    S. Rusinkiewicz, “A symmetric objective function for icp,” ACM TOG, vol. 38, no. 4, pp. 1–7, 2019

    2019

  13. [21]

    Regtr: End-to-end point cloud correspon- dences with transformers,

    Z. J. Yew and G. H. Lee, “Regtr: End-to-end point cloud correspon- dences with transformers,” in IEEE CVPR, 2022, pp. 6677–6686

    2022

  14. [22]

    Predator: Registration of 3d point clouds with low overlap,

    S. Huang, Z. Gojcic, M. Usvyatsov, A. Wieser, and K. Schindler, “Predator: Registration of 3d point clouds with low overlap,” in IEEE CVPR, 2021, pp. 4267–4276

    2021

  15. [23]

    Lepard: Learning partial point cloud matching in rigid and deformable scenes,

    Y . Li and T. Harada, “Lepard: Learning partial point cloud matching in rigid and deformable scenes,” in IEEE CVPR, 2022, pp. 5554–5564

    2022

  16. [24]

    Geotransformer: Fast and robust point cloud registration with geometric transformer,

    Z. Qin, H. Yu, C. Wang, Y . Guo, Y . Peng, S. Ilic, D. Hu, and K. Xu, “Geotransformer: Fast and robust point cloud registration with geometric transformer,” IEEE TPAMI, vol. 45, no. 8, pp. 9806–9821, 2023

    2023

  17. [25]

    Global correspondence optimiza- tion for non-rigid registration of depth scans,

    H. Li, R. W. Sumner, and M. Pauly, “Global correspondence optimiza- tion for non-rigid registration of depth scans,” in Computer Graphics Forum, vol. 27, no. 5, 2008, pp. 1421–1430

    2008

  18. [26]

    Neural deformation graphs for globally-consistent non-rigid reconstruc- tion,

    A. Bozic, P. Palafox, M. Zollhofer, J. Thies, A. Dai, and M. Nießner, “Neural deformation graphs for globally-consistent non-rigid reconstruc- tion,” in IEEE CVPR, 2021, pp. 1450–1459

    2021

  19. [27]

    Non-rigid point cloud registration with neural deformation pyramid,

    Y . Li and T. Harada, “Non-rigid point cloud registration with neural deformation pyramid,” NeurIPS, vol. 35, pp. 27 757–27 768, 2022

    2022

  20. [28]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser et al., “Attention is all you need,” NeurIPS, vol. 30, 2017

    2017

  21. [29]

    An objective comparison of methods for augmented reality in laparoscopic liver resection by preoperative-to-intraoperative image fusion from the miccai2022 chal- lenge,

    S. Ali, Y . Espinel, Y . Jin, P. Liu, B. G ¨uttner, X. Zhang, L. Zhang, T. Dowrick, M. J. Clarkson, S. Xiao et al. , “An objective comparison of methods for augmented reality in laparoscopic liver resection by preoperative-to-intraoperative image fusion from the miccai2022 cha...

    2024

  22. [30]

    A flexible new technique for camera calibration,

    Z. Zhang, “A flexible new technique for camera calibration,” IEEE TPAMI, vol. 22, no. 11, pp. 1330–1334, 2000

    2000

  23. [31]

    Generating large labeled data sets for laparoscopic image processing tasks using unpaired image-to-image translation,

    M. Pfeiffer, I. Funke, M. R. Robu, S. Bodenstedt, L. Strenger, S. En- gelhardt, T. Roß, M. J. Clarkson, K. Gurusamy et al., “Generating large labeled data sets for laparoscopic image processing tasks using unpaired image-to-image translation,” in MICCAI, 2019, pp. 119–127

    2019

  24. [32]

    Method for registration of 3-d shapes,

    P. J. Besl and N. D. McKay, “Method for registration of 3-d shapes,” in Sensor Fusion IV: control paradigms and data structures , vol. 1611, 1992, pp. 586–606

    1992

  25. [33]

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

    L. Yang, B. Kang, Z. Huang, X. Xu, J. Feng, and H. Zhao, “Depth anything: Unleashing the power of large-scale unlabeled data,” in IEEE CVPR, 2024, pp. 10 371–10 381

    2024

  26. [34]

    Principal component analysis,

    S. Wold, K. Esbensen, and P. Geladi, “Principal component analysis,” Chemometrics and Intelligent Laboratory Systems , vol. 2, no. 1-3, pp. 37–52, 1987

    1987

  27. [35]

    Kpconv: Flexible and deformable convolution for point clouds,

    H. Thomas, C. R. Qi, J.-E. Deschaud, B. Marcotegui, F. Goulette, and L. J. Guibas, “Kpconv: Flexible and deformable convolution for point clouds,” in IEEE ICCV, 2019, pp. 6411–6420

    2019

  28. [36]

    S2former- or: Single-stage bi-modal transformer for scene graph generation in or,

    J. Pei, D. Guo, J. Zhang, M. Lin, Y . Jin, and P.-A. Heng, “S2former- or: Single-stage bi-modal transformer for scene graph generation in or,” IEEE Transactions on Medical Imaging , 2024. AUTHOR et al.: PREP ARA TION OF P APERS FOR IEEE TRANSACTIONS ON MEDICAL IMAGING 13

    2024

  29. [37]

    Pointnet++: Deep hierarchical feature learning on point sets in a metric space,

    C. R. Qi, L. Yi, H. Su, and L. J. Guibas, “Pointnet++: Deep hierarchical feature learning on point sets in a metric space,” NeurIPS, vol. 30, 2017

    2017

  30. [38]

    Linformer: Self-attention with linear complexity,

    S. Wang, B. Z. Li, M. Khabsa, H. Fang et al., “Linformer: Self-attention with linear complexity,” arXiv preprint arXiv:2006.04768 , 2020

    2006 arXiv

  31. [39]

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

    M. A. Fischler and R. C. Bolles, “Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,” Communications of the ACM, vol. 24, no. 6, pp. 381–395, 1981

    1981

  32. [40]

    Accelerating 3d deep learning with pytorch3d,

    N. Ravi, J. Reizenstein, D. Novotny, T. Gordon, W.-Y . Lo, J. Johnson, and G. Gkioxari, “Accelerating 3d deep learning with pytorch3d,” arXiv:2007.08501, 2020

    2007 arXiv

  33. [41]

    Interpolating between optimal transport and mmd using sinkhorn divergences,

    J. Feydy, T. S ´ejourn´e, F.-X. Vialard, S.-i. Amari, A. Trouv ´e, and G. Peyr ´e, “Interpolating between optimal transport and mmd using sinkhorn divergences,” in AISTATS, 2019, pp. 2681–2690

    2019

  34. [42]

    Nerfies: Deformable neural radiance fields,

    K. Park, U. Sinha, J. T. Barron, S. Bouaziz, D. B. Goldman, S. M. Seitz, and R. Martin-Brualla, “Nerfies: Deformable neural radiance fields,” in IEEE ICCV, 2021, pp. 5865–5874

    2021

  35. [43]

    Neural scene flow prior,

    X. Li, J. Kaesemodel Pontes, and S. Lucey, “Neural scene flow prior,” NeurIPS, vol. 34, pp. 7838–7851, 2021

    2021

  36. [44]

    Dynamic point fields,

    S. Prokudin, Q. Ma, M. Raafat, J. Valentin, and S. Tang, “Dynamic point fields,” in IEEE ICCV, October 2023, pp. 7964–7976

    2023

  37. [45]

    Correspondence-free non-rigid point set registration using unsupervised clustering analysis,

    M. Zhao, J. Jiang, L. Ma, S. Xin, G. Meng, and D.-M. Yan, “Correspondence-free non-rigid point set registration using unsupervised clustering analysis,” in IEEE CVPR, 2024, pp. 21 199–21 208

    2024

  38. [46]

    Least-squares estimation of transformation parameters between two point patterns,

    S. Umeyama, “Least-squares estimation of transformation parameters between two point patterns,” IEEE TPAMI, vol. 13, no. 04, pp. 376– 380, 1991

    1991

  39. [47]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Loet al., “Segment anything,” in IEEE ICCV, 2023, pp. 4015–4026

    2023

  40. [48]

    Segment anything in high quality,

    L. Ke, M. Ye, M. Danelljan, Y .-W. Tai, C.-K. Tang, F. Yuet al., “Segment anything in high quality,” NeurIPS, vol. 36, pp. 29 914–29 934, 2023

    2023

  41. [49]

    Segment anything in medical images,

    J. Ma, Y . He, F. Li, L. Han, C. You, and B. Wang, “Segment anything in medical images,” Nature Communications, vol. 15, no. 1, p. 654, 2024

    2024

  42. [50]

    Vision transformers for dense prediction,

    R. Ranftl, A. Bochkovskiy, and V . Koltun, “Vision transformers for dense prediction,” in IEEE ICCV, 2021, pp. 12 179–12 188

    2021

  43. [51]

    Towards robust monocular depth estimation: Mixing datasets for zero-shot cross- dataset transfer,

    R. Ranftl, K. Lasinger, D. Hafner, K. Schindler, and V . Koltun, “Towards robust monocular depth estimation: Mixing datasets for zero-shot cross- dataset transfer,” IEEE TPAMI, vol. 44, no. 3, 2022

    2022

  44. [52]

    Depth anything v2,

    L. Yang, B. Kang, Z. Huang, Z. Zhao, X. Xu, J. Feng, and H. Zhao, “Depth anything v2,” NeurIPS, vol. 37, pp. 21 875–21 911, 2024

    2024

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.