Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Diff2DGS reconstructs surgical scenes by first inpainting instruments with a diffusion video model, then fitting 2D Gaussians with learned deformations and adaptive depth weighting.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 21:56 UTC pith:QNZ2RQVX

load-bearing objection Useful engineering paper, but the geometry claims don't survive contact with their own tables, and the independent SCARED evaluation is contaminated by fine-tuning on the same data. the 4 major comments →

arxiv 2602.18314 v2 pith:QNZ2RQVX submitted 2026-02-20 cs.CV cs.GRcs.RO

Diff2DGS: Reliable Reconstruction of Occluded Surgical Scenes via 2D Gaussian Splatting

Tianyi Song , Danail Stoyanov , Evangelos Mazomenos , Francisco Vasconcelos This is my paper
classification cs.CV cs.GRcs.RO
keywords surgical scene reconstruction2D Gaussian splattingdiffusion inpaintingdeformable tissueadaptive depth lossEndoNeRFStereoMISSCARED
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper claims that reliable 3D reconstruction of deformable surgical scenes becomes tractable if you first remove instruments from the video by temporally consistent diffusion inpainting, then represent the tissue with 2D Gaussian Splatting driven by a lightweight learnable deformation model. It further claims that once occluded regions are inpainted, an adaptive depth loss is needed to prevent appearance-optimized splatting from producing geometrically wrong surfaces. The method reports state-of-the-art appearance metrics on EndoNeRF and StereoMIS and better depth error on SCARED, where real ground truth exists. The deeper message is that image-quality metrics alone can mask poor geometry, so evaluation should include depth accuracy.

Core claim

The central claim is that occlusion by surgical instruments, the main failure mode of prior deformable Gaussian splatting methods, can be eliminated before reconstruction rather than masked during optimization. Diff2DGS shows that a diffusion-based video inpainting module with temporal attention produces spatiotemporally consistent tissue appearance under surgical tools, and that plugging those inpainted frames into a 2D Gaussian Splatting pipeline — extended with a learnable deformation model and an adaptive depth loss weight — yields reconstructions with higher PSNR/SSIM and lower depth error than prior methods. The paper also demonstrates the corollary that optimizing purely for image qua

What carries the argument

The key mechanism is the combination of a diffusion video inpainting model (built on Stable Diffusion v1.5 with a causal temporal attention mask) that restores instrument-occluded tissue, and a 2D Gaussian Splatting renderer where each planar Gaussian is deformed over time by a Learnable Deformation Model (LDM) that predicts scale, rotation, and position changes through Gaussian basis functions of time. An adaptive depth loss weight, driven by the ratio of RGB to depth loss, steers training toward geometric fidelity. The depth is also used to initialize the point cloud and as a supervision signal.

Load-bearing premise

The method assumes that RAFT stereo depth is a trustworthy geometric reference for instrument-occluded and inpainted tissue regions; if RAFT errors correlate with the masked areas, both initialization and depth evaluation would be biased in the same direction.

What would settle it

Take a surgical video with known ground-truth depth from a structured-light sensor (like SCARED) or CT, artificially occlude tissue with instrument masks, run Diff2DGS, and compare the recovered geometry in the masked regions against the true anatomy. If the depth RMSE in those regions is no better than Deform3DGS or SurgicalGS, or if appearance improves while geometry diverges, the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If correct, instrument-occluded regions would reconstruct with fewer artifacts, removing the need for per-pixel masking during optimization.
  • Depth estimates would be closer to true anatomy, making intraoperative navigation and autonomous assistance more reliable.
  • The finding that appearance metrics can hide geometric error would push future surgical scene benchmarks to include depth ground truth.
  • The method's rendering speed (hundreds of times faster than NeRF-based baselines) makes real-time intraoperative use feasible.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Inference: The inpainting-before-reconstruction design suggests a general principle — for any transient occlusion (blood, smoke, tools), a temporally consistent 2D inpainting preprocessor can turn a dynamic reconstruction problem into a cleaner static-with-deformation problem.
  • Inference: Because the adaptive depth loss decouples geometry from appearance, similar mechanisms could improve other medical or robotic domains where appearance objectives dominate training.
  • Inference: The reliance on RAFT depth as both initialization and evaluation creates a coupling; a testable implication is that on a dataset with independent ground truth (CT or structured light), the reported depth advantage could shrink or grow depending on whether RAFT errors correlate with the masked regions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes Diff2DGS, a two-stage pipeline for dynamic surgical scene reconstruction. First, a diffusion-based video inpainting module removes surgical instruments and fills occluded tissue with temporally consistent content. Second, 2D Gaussian Splatting (2DGS) is combined with a Learnable Deformation Model (LDM) to model tissue deformation, and an adaptive depth loss is used to improve geometric fidelity. The method is evaluated on EndoNeRF, StereoMIS, and SCARED. The authors claim state-of-the-art performance in both appearance and geometry, citing 38.02 dB PSNR on EndoNeRF and 34.40 dB on StereoMIS, and they introduce a SCARED-based quantitative depth evaluation protocol.

Significance. If the central claims were fully supported, the paper would be a useful contribution to real-time intraoperative 3D reconstruction: explicitly inpainting occluded tissue before reconstruction, adapting 2DGS to deformable scenes, and pushing geometry evaluation beyond image-quality metrics on a dataset with true depth ground truth. The manuscript also provides code, ablations, and a unified SCARED evaluation protocol. However, as written, the evidence for the geometric superiority claim is invalid: the RMSE tables contradict the claim, and the SCARED evaluation is contaminated by fine-tuning/evaluation overlap. The appearance-level contribution appears sound but needs to be separated from the unsupported geometric claims.

major comments (4)
  1. [Section III-B, Tables I and II] The central claim that Diff2DGS 'outperforms state-of-the-art methods in both appearance and geometry' is contradicted by the paper's own RMSE results. On every scene split in Tables I and II, SurgicalGS has lower RMSE than Diff2DGS: EndoNeRF-Cutting 1.779 vs 2.443, EndoNeRF-Pulling 2.193 vs 2.574, StereoMIS-P27 2.747 vs 3.874, StereoMIS-P3 2.536 vs 3.717. Since RMSE is the only geometry metric reported on these datasets, the paper's geometry superiority claim is not supported by its own data.
  2. [Section III-A, Section III-D, Eq. (14)] The depth evaluation on EndoNeRF and StereoMIS is circular. The method uses RAFT stereo depth for depth supervision (the depth loss in Eq. (14) is trained against RAFT estimates) and then reports RMSE against those same RAFT estimates. Table V validates RAFT on SCARED keyframes, not on instrument-occluded or inpainted regions, so accuracy in the masked regions is untested. Thus the RMSE numbers in Tables I and II measure consistency with a training reference, not true anatomical accuracy. Geometry claims on these datasets should be removed or replaced by evaluation against independent ground truth.
  3. [Section III-A, Table III] The SCARED geometry evaluation is contaminated by training/evaluation overlap. Section III-A states that the inpainting module is 'fine-tuned on surgical sequences of SCARED dataset,' and Table III evaluates on SCARED with synthetic instrument masks, using the original unmasked images as ground truth. No train/test split is reported. Since SCARED contains only five porcine cadaver sequences, the model may have memorized the clean tissue appearance during fine-tuning, which would inflate the masked-region PSNR of 30.53 dB and the RMSE of 8.21 mm. The authors must specify a split that excludes the evaluation sequences from fine-tuning and report per-sequence results.
  4. [Section IV and abstract] The paper's conclusion and abstract claim 'superior performance' and 'reliable reconstruction' largely on the basis of geometric accuracy. Given that the only geometry evidence with true ground truth (SCARED) is contaminated, and the other geometry evidence is circular, the current manuscript does not substantiate its main selling point. A revised version should either provide a valid held-out SCARED evaluation or substantially weaken the geometry claims.
minor comments (5)
  1. [Section I vs abstract] The abstract reports 34.40 dB PSNR on StereoMIS, while the introduction reports 33.90 dB. This inconsistency should be corrected.
  2. [Table II] The LPIPS value for SurgicalGS on StereoMIS-P3 is listed as 0.925, which is likely a typo (probably 0.0925). Please verify.
  3. [Section III-B, Eq. (19)] The TCS discussion is confusing: the 'Inpaint result' row in Table III has TCS 0.003, which is better than Diff2DGS's 0.027, yet the text says Diff2DGS attains 'suboptimal performance—yet remains superior to all other compared methods.' Clarify what baseline is being compared and whether the inpainting result is considered a competing method.
  4. [Tables I-III] No error bars, confidence intervals, or multiple-run statistics are provided. For a paper claiming state-of-the-art performance, at least a small number of runs with variance should be reported.
  5. [Section II-C, Eq. (10)] There is a typo: 'lantent space' should be 'latent space.'

Circularity Check

2 steps flagged

Geometry evaluation is partially circular: RAFT depth is both the training/depth-loss reference and the RMSE reference, and the SCARED inpainting evaluation uses SCARED images that the inpainting model was fine-tuned on, with no reported split.

specific steps
  1. fitted input called prediction [Section III-A, Section III-B, Section III-D, Tables I-II; Eq. 14]
    "we provide depth difference (RMSE) between methods' estimates and stereo reconstructions (RAFT [38]). ... Since neither EndoNeRF nor StereoMIS provides ground truth depth for occluded regions, and StereoMIS depth maps are derived from RAFT [38], we follow prior work and use RAFT to estimate depth for inpainted regions."

    The paper reports RMSE against RAFT depth on EndoNeRF and StereoMIS, while the proposed pipeline uses stereo depth information — the same RAFT-derived depth — to initialize the Gaussian point cloud and to supervise the depth term L_depth in Eq. 14. Thus the reported geometry accuracy on the two datasets with no 3D ground truth measures how well the model fit its own depth teacher, not agreement with an independent anatomical reference. The Table V validation of RAFT against SCARED ground truth does not break this loop for EndoNeRF/StereoMIS, because the inpainted regions used for evaluation are still generated by the same depth source used in training.

  2. fitted input called prediction [Section III-A 'Experiment Details' and Section III-B 'Comparison' / Table III]
    "The model is fine-tuned on surgical sequences of SCARED dataset where supervision is provided by a mask-weighted L2 loss with the learning rate of 1e-5. ... To validate the improvement in masked regions, we synthetically overlay surgical instrument masks on the SCARED dataset, using the original unmasked images as ground truth."

    The inpainting module is fine-tuned on SCARED with a mask-weighted L2 loss whose target is the clean, unmasked tissue. The masked-region evaluation then overlays synthetic masks on SCARED and compares to the original unmasked SCARED images — exactly the supervision targets. No train/test split is reported, and SCARED contains only five porcine cadaver sequences, so the Table III masked-region PSNR/RMSE gains (e.g., PSNR 30.53 vs. 21-24 for prior methods) can reflect memorization of training images rather than reconstruction of unseen occluded tissue. Since SCARED is the one dataset with independent structured-light depth ground truth, this contamination weakens the only non-RAFT geometry evidence.

full rationale

The core model contributions — the 2D Gaussian deformation model, the adaptive depth weighting scheme, and the diffusion inpainting architecture — are not themselves derived from the paper's own outputs. The circularity is confined to the evaluation loop, but that loop is load-bearing because the paper's headline claim is superiority in geometry on top of appearance. On EndoNeRF and StereoMIS, the only depth reference is RAFT, which is also the source of the stereo depth used for initialization and depth supervision, so reported RMSE values are partly a measure of self-consistency with the training signal. On SCARED, the inpainting model is fine-tuned on the same dataset whose unmasked images are later used as ground truth for synthetic-mask evaluation, and no split is described; this compromises the masked-region PSNR/RMSE results that support the occlusion-reconstruction claim. These are two distinct evaluation circularities, so the score is 6 rather than a lower score. No load-bearing self-citation chain was identified, and the splatting/deformation derivation itself is not circular.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central claims rest on pretrained or learned components and on a hand-designed loss schedule. No new physical entities are introduced. The most load-bearing external inputs are the inpainting prior, the RAFT depth reference, and the adaptive depth-loss hyperparameters.

free parameters (4)
  • LDM basis count B and basis parameters {theta_j, sigma_j, omega_j} = B not reported; theta_j, sigma_j, omega_j learned per scene
    These define the deformation model in Eqs. (8)-(9); without reported values or initialization, deformation capacity is underspecified.
  • Depth-loss schedule {w_init, w_final, alpha, beta} = alpha=0.5; beta in (0,1]; w_max=2*w_init; w_init and w_final unreported
    Eqs. (15)-(18) define the adaptive depth weight from hand-set values; no ablation or sensitivity analysis is given.
  • Diffusion fine-tuning setup (learning rate, SCARED training sequences, PCM steps) = LR=1e-5; two denoising steps via PCM; SCARED sequences
    The inpainting model is fine-tuned on SCARED, which is also used for evaluation in Table III; the exact train/test split is not specified.
  • Gaussian Splatting training hyperparameters = LR=3.2e-3; 6000 iterations; freeze first 600 iterations
    Hand-set training choices with no sensitivity analysis, though standard for Gaussian Splatting.
axioms (5)
  • domain assumption Pretrained Stable Diffusion v1.5 and Phased Consistency Model provide a suitable generative prior for surgical tissue inpainting.
    The inpainting module is fine-tuned from SD1.5 and accelerated with PCM; the paper does not justify that this prior produces anatomically correct tissue.
  • domain assumption RAFT stereo depth is an accurate proxy for 3D ground truth in surgical scenes and in occluded/inpainted regions.
    RAFT depth is used both as depth supervision and as the RMSE reference; it is validated only on SCARED keyframes (Table V), not on inpainted regions.
  • domain assumption 2D Gaussian surfaces represent deformable tissue geometry better than 3D Gaussians.
    Borrowed from 2DGS [27] and asserted as motivation in Section II-A without a dedicated study for surgical tissue.
  • domain assumption Tissue deformation under relatively static camera views can be captured by a time-conditioned per-Gaussian deformation model.
    The conclusion limits the method to relatively static camera views; this is an explicit scope assumption for the deformation modeling.
  • standard math Standard Gaussian/rasterization mathematics (Eqs. 1-7) is taken as given.
    The paper builds on 3DGS and 2DGS formulations without proving them; this is accepted background.

pith-pipeline@v1.3.0-alltime-deepseek · 11027 in / 16517 out tokens · 139365 ms · 2026-08-02T21:56:10.423942+00:00 · methodology

0 comments
read the original abstract

Real-time reconstruction of deformable surgical scenes is vital for advancing robotic surgery, improving intraoperative guidance, and enabling automation. Recent methods achieve dense reconstructions from da Vinci robotic surgery videos, with Gaussian Splatting (GS) offering real-time performance via graphics acceleration. However, reconstruction quality in occluded regions remains limited, and depth accuracy has not been fully assessed, as benchmarks like EndoNeRF and StereoMIS lack 3D ground truth. We propose Diff2DGS, a two-stage framework for reliable 3D reconstruction of occluded surgical scenes. First, a diffusion-based video module with temporal priors inpaints tissue occluded by instruments with high spatiotemporal consistency. Second, we adapt 2D Gaussian Splatting (2DGS) with a Learnable Deformation Model (LDM) to capture dynamic tissue deformation and anatomical geometry, and introduce adaptive depth weight to improve geometric fidelity. We further extend evaluation beyond image-quality metrics by performing quantitative depth analysis on the SCARED dataset. Diff2DGS outperforms state-of-the-art methods in both appearance and geometry, reaching 38.02 dB PSNR on EndoNeRF and 34.40 dB on StereoMIS. Our experiments also show that optimizing image quality alone does not necessarily ensure accurate 3D geometry.The code is available at https://diff2dgs.github.io/.

Figures

Figures reproduced from arXiv: 2602.18314 by Danail Stoyanov, Evangelos Mazomenos, Francisco Vasconcelos, Tianyi Song.

Figure 1
Figure 1. Figure 1: Traditional endoscopic scene reconstruction methods often focus [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Our reliable surgical scene reconstruction framework, Diff2DGS, consists of Surgical Instrument Inpainting, Point Cloud Initialization, Deformation [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of the 3D reconstruction results. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Depth Quality Visualization on SCARED Dataset [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Visual comparison on SCARED dataset TABLE IV ABLATION EXPERIMENT ON STEREOMIS DATASET Models PSNR ↑ SSIM(%) ↑ Diffusion+deform3dgs 34.32 90.13 w/o deformation 32.12 86.87 w/o inpainting 31.95 86.11 Diff2dgs(Ours) 34.40 90.72 [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Ablation experiment on EndoNeRF dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Track2Map: Online Deformable SLAM with Motion-Aware Pose Optimization in Robotic Surgery

    cs.CV 2026-07 conditional novelty 6.0

    Track2Map jointly optimizes camera poses and deformable 3D Gaussian maps online from surgical stereo video via track-anchored deformation and motion-gated pose updates.

Reference graph

Works this paper leans on

38 extracted references · 6 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Automatic registration with continuous pose updates for marker-less surgical navigation in spine surgery,

    F. Liebmann, M von Atzigen, D St ¨utz, J Wolf, L Zingg, D Suter, N A. Cavalcanti, L Leoty, H Esfandiari, J G. Snedekeret al., “Automatic registration with continuous pose updates for marker-less surgical navigation in spine surgery,”Medical Image Analysis, vol. 91, p. 103027, 2024

    2024

  2. [2]

    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 ¨omet al., “Evaluation of a novel navigation platform for laparoscopic liver surgery with organ deformation compensation using injected fiducials,” Medical image analysis, vol. 69, p. 101946, 2021

    2021

  3. [3]

    Phased consistency models,

    F.-Y . Wang, Z Huang, A Bergman, D Shen, P Gao, M Lingelbach, K Sun, W Bian, G Song, Y Liuet al., “Phased consistency models,” Advances in neural information processing systems, vol. 37, pp. 83 951–84 009, 2024

    2024

  4. [4]

    Revisiting stereo depth estimation from a sequence- to-sequence perspective with transformers,

    Z Li, X Liu, N Drenkow, A Ding, F X. Creighton, R H. Taylor, and M Unberath, “Revisiting stereo depth estimation from a sequence- to-sequence perspective with transformers,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 6197–6206

    2021

  5. [5]

    High- frequency stereo matching network,

    H Zhao, H Zhou, Y Zhang, J Chen, Y Yang, and Y Zhao, “High- frequency stereo matching network,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 1327–1336

    2023

  6. [6]

    Endosurf: Neural surface reconstruction of deformable tissues with stereo endoscope videos,

    R Zha, X Cheng, H Li, M Harandi, and Z Ge, “Endosurf: Neural surface reconstruction of deformable tissues with stereo endoscope videos,” inInternational conference on medical image computing and computer-assisted intervention. Springer, 2023, pp. 13–23

    2023

  7. [7]

    Reconstructing sinus anatomy from endoscopic video–towards a radiation-free approach for quantitative longitudinal assessment,

    X Liu, M Stiber, J Huang, M Ishii, G D. Hager, R H. Taylor, and M Unberath, “Reconstructing sinus anatomy from endoscopic video–towards a radiation-free approach for quantitative longitudinal assessment,” inMedical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Proceedings, Par...

    2020

  8. [8]

    Augmented reality technology for preoperative planning and intraoperative navigation during hepatobiliary surgery: A review of current methods,

    R Tang, L.-F. Ma, Z.-X. Rong, M.-D. Li, J.-P. Zeng, X.-D. Wang, H.-E. Liao, and J.-H. Dong, “Augmented reality technology for preoperative planning and intraoperative navigation during hepatobiliary surgery: A review of current methods,”Hepatobiliary & Pancreatic Diseases International, vol. 17, no. 2, pp. 101–112, 2018

    2018

  9. [9]

    Neural lerplane representations for fast 4d reconstruction of deformable tissues,

    C Yang, K Wang, Y Wang, X Yang, and W Shen, “Neural lerplane representations for fast 4d reconstruction of deformable tissues,” inIn- ternational Conference on Medical Image Computing and Computer- Assisted Intervention. Springer, 2023, pp. 46–56

    2023

  10. [10]

    Nerf: Representing scenes as neural radiance fields for view synthesis,

    B Mildenhall, P P. Srinivasan, M Tancik, J T. Barron, R Ramamoorthi, and R Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,”Communications of the ACM, vol. 65, no. 1, pp. 99– 106, 2021

    2021

  11. [11]

    Neural rendering for stereo 3d reconstruction of deformable tissues in robotic surgery,

    Y Wang, Y Long, S H. Fan, and Q Dou, “Neural rendering for stereo 3d reconstruction of deformable tissues in robotic surgery,” in International conference on medical image computing and computer- assisted intervention. Springer, 2022, pp. 431–441

    2022

  12. [12]

    Efficient endonerf reconstruction and its application for data-driven surgical simulation,

    Y Wang, B Gong, Y Long, S H. Fan, and Q Dou, “Efficient endonerf reconstruction and its application for data-driven surgical simulation,” International Journal of Computer Assisted Radiology and Surgery, vol. 19, no. 5, pp. 821–829, 2024

    2024

  13. [13]

    Endo-4dgs: Dis- tilling depth ranking for endoscopic monocular scene reconstruction with 4d gaussian splatting,

    Y Huang, B Cui, L Bai, Z Guo, M Xu, and H Ren, “Endo-4dgs: Dis- tilling depth ranking for endoscopic monocular scene reconstruction with 4d gaussian splatting,”arXiv preprint arXiv:2401.16416, 2024

    Pith/arXiv arXiv 2024

  14. [14]

    4d gaussian splatting for real-time dynamic scene rendering,

    G Wu, T Yi, J Fang, L Xie, X Zhang, W Wei, W Liu, Q Tian, and X Wang, “4d gaussian splatting for real-time dynamic scene rendering,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 20 310–20 320

    2024

  15. [15]

    3d gaussian splatting for real-time radiance field rendering

    B Kerbl, G Kopanas, T Leimk ¨uhler, and G Drettakis, “3d gaussian splatting for real-time radiance field rendering.”ACM Trans. Graph., vol. 42, no. 4, pp. 139–1, 2023

    2023

  16. [16]

    Deformable 3d gaussians for high-fidelity monocular dynamic scene reconstruction,

    Z Yang, X Gao, W Zhou, S Jiao, Y Zhang, and X Jin, “Deformable 3d gaussians for high-fidelity monocular dynamic scene reconstruction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 20 331–20 341

    2024

  17. [17]

    Spacetime gaussian feature splat- ting for real-time dynamic view synthesis,

    Z Li, Z Chen, Z Li, and Y Xu, “Spacetime gaussian feature splat- ting for real-time dynamic view synthesis,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 8508–8520

    2024

  18. [18]

    Hexplane: A fast representation for dynamic scenes,

    A Cao and J Johnson, “Hexplane: A fast representation for dynamic scenes,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 130–141

    2023

  19. [19]

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

    A Kendall, M Grimes, and R Cipolla, “Posenet: A convolutional network for real-time 6-dof camera relocalization,” inProceedings of the IEEE international conference on computer vision, 2015, pp. 2938–2946

    2015

  20. [20]

    Deform3dgs: Flex- ible deformation for fast surgical scene reconstruction with gaussian splatting,

    S Yang, Q Li, D Shen, B Gong, Q Dou, and Y Jin, “Deform3dgs: Flex- ible deformation for fast surgical scene reconstruction with gaussian splatting,” inInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2024, pp. 132–142

    2024

  21. [21]

    Endogaussian: Gaussian splatting for deformable surgical scene reconstruction,

    Y Liu, C Li, C Yang, and Y Yuan, “Endogaussian: Gaussian splatting for deformable surgical scene reconstruction,”arXiv e-prints, pp. arXiv–2401, 2024

    2024

  22. [22]

    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 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 10 371–10 381

    2024

  23. [23]

    Towards language-driven video inpainting via multimodal large language models,

    J Wu, X Li, C Si, S Zhou, J Yang, J Zhang, Y Li, K Chen, Y Tong, Z Liuet al., “Towards language-driven video inpainting via multimodal large language models,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 12 501–12 511

    2024

  24. [24]

    Avid: Any-length video inpainting with diffusion model,

    Z Zhang, B Wu, X Wang, Y Luo, L Zhang, Y Zhao, P Vajda, D Metaxas, and L Yu, “Avid: Any-length video inpainting with diffusion model,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 7162–7172

    2024

  25. [25]

    Advanced video inpaint- ing using optical flow-guided efficient diffusion,

    B Gu, H Luo, S Guo, and P Dong, “Advanced video inpaint- ing using optical flow-guided efficient diffusion,”arXiv preprint arXiv:2412.00857, 2024

    Pith/arXiv arXiv 2024

  26. [26]

    Video diffusion models are strong video inpainter,

    M Lee, S Cho, C Shin, J Lee, S Yang, and S Lee, “Video diffusion models are strong video inpainter,”arXiv preprint arXiv:2408.11402, 2024

    Pith/arXiv arXiv 2024

  27. [27]

    2d gaussian splatting for geometrically accurate radiance fields,

    B Huang, Z Yu, A Chen, A Geiger, and S Gao, “2d gaussian splatting for geometrically accurate radiance fields,” inACM SIGGRAPH 2024 conference papers, 2024, pp. 1–11

    2024

  28. [28]

    Learning how to robustly estimate camera pose in endoscopic videos,

    M Hayoz, C Hahne, M Gallardo, D Candinas, T Kurmann, M Allan, and R Sznitman, “Learning how to robustly estimate camera pose in endoscopic videos,”International journal of computer assisted radiology and surgery, vol. 18, no. 7, pp. 1185–1192, 2023

    2023

  29. [29]

    Pytorch,

    S Imambi, K B. Prakash, and G Kanagachidambaresan, “Pytorch,” Programming with TensorFlow: solution for edge computing applica- tions, pp. 87–104, 2021

    2021

  30. [30]

    Denoising diffusion implicit models,

    J Song, C Meng, and S Ermon, “Denoising diffusion implicit models,” arXiv preprint arXiv:2010.02502, 2020

    Pith/arXiv arXiv 2010

  31. [31]

    SurgicalGS: Dynamic 3D Gaussian Splatting for Accurate Robotic-Assisted Surgical Scene Reconstruction ,

    J Chen, X Zhang, M I. Hoque, F Vasconcelos, D Stoyanov, D S. Elson, and B Huang, “ SurgicalGS: Dynamic 3D Gaussian Splatting for Accurate Robotic-Assisted Surgical Scene Reconstruction ,” in proceedings of Medical Image Computing and Computer Assisted Intervention – MICCAI 2025, vol. LNCS 15970. Springer Nature Switzerland, September 2025

    2025

  32. [32]

    Stereo correspondence and reconstruction of endoscopic data challenge,

    M Allan, J Mcleod, C Wang, J C. Rosenthal, Z Hu, N Gard, P Eisert, K X. Fu, T Zeffiro, W Xiaet al., “Stereo correspondence and reconstruction of endoscopic data challenge,”arXiv preprint arXiv:2101.01133, 2021

    Pith/arXiv arXiv 2021

  33. [33]

    Dynamic reconstruction of deformable soft-tissue with stereo scope in minimal invasive surgery,

    J Song, J Wang, L Zhao, S Huang, and G Dissanayake, “Dynamic reconstruction of deformable soft-tissue with stereo scope in minimal invasive surgery,”IEEE Robotics and Automation Letters, vol. 3, no. 1, pp. 155–162, 2017

    2017

  34. [34]

    Causal-story: Local causal attention utilizing parameter-efficient tuning for visual story synthesis,

    T Song, J Cao, K Wang, B Liu, and X Zhang, “Causal-story: Local causal attention utilizing parameter-efficient tuning for visual story synthesis,” inICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2024, pp. 3350–3354

    2024

  35. [35]

    Real-time dense reconstruction of tissue surface from stereo optical video,

    H Zhou and J Jagadeesan, “Real-time dense reconstruction of tissue surface from stereo optical video,”IEEE transactions on medical imaging, vol. 39, no. 2, pp. 400–412, 2019

    2019

  36. [36]

    V olumefusion: Deep depth fusion for 3d scene reconstruction,

    J Choe, S Im, F Rameau, M Kang, and I S. Kweon, “V olumefusion: Deep depth fusion for 3d scene reconstruction,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 16 086–16 095

    2021

  37. [37]

    Diffueraser: A diffusion model for video inpainting,

    X Li, H Xue, P Ren, and L Bo, “Diffueraser: A diffusion model for video inpainting,”arXiv preprint arXiv:2501.10018, 2025

    Pith/arXiv arXiv 2025

  38. [38]

    Raft: Recurrent all-pairs field transforms for optical flow,

    Z Teed and J Deng, “Raft: Recurrent all-pairs field transforms for optical flow,” inEuropean conference on computer vision. Springer, 2020, pp. 402–419

    2020