Pith. sign in

REVIEW 3 major objections 6 minor 31 references

SurgTPGS: Semantic 3D Surgical Scene Understanding with Text Promptable Gaussian Splatting

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read SurgTPGS makes 3D surgical scenes text-queryable by embedding language features in deformable Gaussian splatting.

desk verdict Useful integration of language-embedded 3DGS for surgery, but the SOTA claim lacks a 2D oracle baseline and its own ablation table. read the letter →

arxiv 2506.23309 v2 pith:BPECTC5T submitted 2025-06-29 eess.IV cs.CV

Yiming Huang , Long Bai , Beilei Cui , Kun Yuan , Guankun Wang , Mobarak I. Hoque , Nicolas Padoy , Nassir Navab
show 1 more author
Hongliang Ren
This is my paper
classification eess.IVcs.CV
keywords 3Dsceneunderstandingtext-promptablesegmentationGaussiansplattingsurgicalreconstructionopen-vocabularydeformablemodelingroboticsurgeryvision-languagemodel
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims to be the first text-promptable Gaussian Splatting method for 3D surgical scene understanding. Its goal is to let a surgeon ask a reconstructed 3D scene a natural-language question, such as 'grasper' or 'abdominal wall,' and receive a viewpoint-consistent 3D segmentation mask in real time. Existing surgical vision-language models and segmentation networks operate on 2D frames, while 3D reconstruction methods lack language-aligned understanding; SurgTPGS is built to fill that gap. The authors report that on the CholecSeg8K and EndoVis18 datasets, their method outperforms LangSplat, OpenGaussian, and DGD in segmentation mIoU, while keeping query speed at 67.5 FPS and training time around 4 minutes on a single GPU. If true, the work would make language-driven semantic awareness practical for intraoperative navigation and robot-assisted surgery.

What carries the argument

The central object is a deformable 3D Gaussian splatting scene model in which every Gaussian carries a compressed language-semantic feature vector alongside its mean, rotation, scale, opacity, and color. The semantic features come from SAM-generated masks restricting CAT-Seg's vision-language embeddings to segmented regions, then pass through an autoencoder for dimension reduction. Semantic-aware deformation tracking uses a 1D convolutional network gated by a sigmoid map to deform the feature field together with the geometric deformation, preserving spatial consistency in moving tissue. Semantic region-aware optimization applies a smoothness loss to large semantic regions to keep features uniform within each structure. Text-promptable querying renders the feature map, decodes it, and compares it with text embeddings using canonical CLIP prompts, thresholding at e = 0.4 to produce a final mask.

What would settle it

Replace CAT-Seg with a vision-language model not fine-tuned on CholecSeg8K or EndoVis18 and measure per-class mIoU on the same sequences; if the reported gap vanishes, the superiority comes from in-domain VLM training rather than the 3D machinery.

Watch

Extended reading notes

Core claim

SurgTPGS's core claim is that semantic features extracted by combining the Segment Anything Model with the CAT-Seg vision-language model can be attached to deformable 3D Gaussians and rendered into feature maps that support open-vocabulary text queries. Semantic-aware deformation tracking extends the flexible deformation scheme of Deform3DGS so that the semantic feature field deforms consistently with the moving tissue, using a lightweight 1D convolutional network gated by a sigmoid that regulates how much the deformation model changes each feature. Semantic region-aware optimization adds a smoothness loss over large annotated semantic regions during training, which the authors say improves reconstruction quality and semantic smoothness. The final representation renders color, depth, and language features simultaneously, and text queries are answered by decoding the rendered feature and computing a relevance score against the text embedding relative to CLIP canonical prompts. This yields a single scene model that trains in minutes and answers language queries in real time while outperforming the compared 3D language-splatting baselines on both real-world surgical datasets.

Load-bearing premise

The load-bearing premise is that fine-tuning CAT-Seg on the same surgical benchmarks used for evaluation is a fair source of semantic supervision; if that is removed or considered circular, the claimed advantage over 2D methods largely disappears.

Editorial extensions

If this is right

  • During a procedure, a surgeon could query the reconstructed scene for a specific instrument or tissue and get a 3D mask that stays consistent as the viewpoint changes.
  • The same representation supports novel-view rendering, depth, and language segmentation from one trained model, removing the need for per-query optimization.
  • A 4-minute training time on one GPU makes scene-specific deployment realistic between cases rather than a long offline step.
  • Because the semantic backbone is modular, future improvements in surgical vision-language models can be plugged in to directly raise 3D query accuracy.
  • The deformation-tracking design targets soft-tissue motion, so the method is positioned to generalize beyond static or rigid scenes to other dynamic surgical settings.

Reading between the lines

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

  • The paper's own ablations show the VLM choice dominates performance: CAT-Seg versions far exceed CLIP and SurgVLP versions, so the open-vocabulary behavior is effectively inherited from the VLM rather than created by the 3D pipeline.
  • The region-aware loss is supervised by ground-truth semantic maps, meaning the method needs per-scene annotated data at training time; deployment to new surgical sites would inherit that annotation cost.
  • A per-frame 2D CAT-Seg oracle applied to the same rendered viewpoints would be a direct test of whether the 3D deformation tracking and region-aware optimization add segmentation accuracy beyond the 2D backbone.
  • The reported speed numbers apply to a single trained scene on one GPU; continuous streaming across a full surgery, with frequent new scenes, remains an open question.
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 / 6 minor

Summary. The paper proposes SurgTPGS, a deformable 3D Gaussian Splatting pipeline that attaches semantic features, obtained by applying SAM masks to CAT-Seg embeddings, to Gaussians and renders those features for text-promptable 3D queries. The method adds a semantic-aware deformation tracking network and a region-aware smoothness loss, and it is evaluated with mIoU on selected sequences and classes of CholecSeg8K and EndoVis18, reporting shorter training time than DGD and real-time query rates. The central claims are that this is the first text-promptable Gaussian Splatting approach for surgical scene understanding and that it achieves state-of-the-art results on real-world surgical datasets.

Significance. If the central claims were fully supported, the method would be a useful step toward intraoperative 3D semantic querying in deformable surgical scenes. Strengths of the manuscript include the release of code, the use of real clinical datasets, a concrete integration of SAM and a strong open-vocabulary segmenter, and reporting of query speed and training time. The main caveat is that the semantic supervision is inherited from an in-domain fine-tuned CAT-Seg, so the reported gains over 3D baselines need to be separated from the choice of VLM; this concern is testable and fixable. With the addition of a 2D oracle, corrected ablation interpretation, and statistical reporting, the contribution would be meaningful for the surgical scene understanding community.

major comments (3)
  1. [Section 3.2, Eq. (2), Tables 1-2] The reported state-of-the-art claim is not yet supported because no 2D oracle baseline is included. All semantic supervision enters through CAT-Seg features (Eq. 2), and Section 3.2 states that CAT-Seg is fine-tuned on CholecSeg8K and EndoVis18 with the evaluation sequences excluded. Tables 1 and 2 compare only 3D query methods; a direct application of the same fine-tuned CAT-Seg to the held-out frames is the natural upper bound for the proposed 3D pipeline. If that 2D oracle already matches or exceeds the reported mIoU, then the 3D reconstruction contributes no measurable semantic understanding and the SOTA result is inherited from the VLM. The paper's own CLIP and SurgVLP variants (e.g., abdominal wall mIoU of 64.96 vs. 89.82 in Table 1) show that the VLM choice dominates the numbers, so this baseline is essential. Please add the 2D CAT-Seg oracle and, if possible, a cross-dataset or zero-shot evaluation to support the open-vocabulary claim.
  2. [Table 3] The ablation table contradicts the text. The row with both components removed reports liver mIoU of 79.43, while the full model reports 79.06, yet Section 3.3 states that removing components decreases mIoU. PSNR does improve from 18.72 to 24.06, but the semantic accuracy claim is not supported for the liver class. Please correct the interpretation, report the direction of each per-class change, and include error bars so that differences such as 79.06 vs. 79.43 can be assessed.
  3. [Tables 1-3] No error bars, number of runs, or dataset-level mean mIoU are reported. Tables 1 and 2 show selected classes per sequence rather than a class-averaged mIoU, and Table 3 shows single-run values, so the claimed superiority over baselines is not statistically supported. Please report mean plus/minus standard deviation over multiple runs and the standard mean-IoU averaged over all classes and sequences for the main comparisons.
minor comments (6)
  1. [Section 3.3] The sentence 'As shown in Fig 2' should refer to Fig. 3, which contains the qualitative comparison results.
  2. [Table 1 caption] The caption contains the typo 'Quantative'; it should be 'Quantitative'.
  3. [Section 3.2] The phrase 'excluding all sequences in our experiments' is ambiguous; please specify the exact training and validation split and state how many sequences were used for CAT-Seg fine-tuning.
  4. [Eq. (5) and surrounding text] The notation F-hat is used for the feature map produced by CAT-Seg, but the text calls it 'ground truth'; this is misleading because F-hat is not a semantic label ground truth but a VLM-derived feature map.
  5. [Section 2.2] The autoencoder Phi and the deformation network g are described only by name; please provide architecture details and training procedures for reproducibility.
  6. [Table 1] The header formatting, such as '01_0008001_00240', makes the class-to-sequence association hard to read; consider grouping the class names clearly under each sequence identifier.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: CAT-Seg is trained on disjoint sequences, semantic features are distilled into 3D Gaussians, and mIoU is measured against manual annotations on held-out sequences.

full rationale

The paper's derivation chain is not circular. Section 3.2 states: "We train CAT-Seg [3] with the CholecSeg8K [5] and EndoVis18 [1], excluding all sequences in our experiments." This means the VLM that produces the semantic supervision (Eq. 2) has not seen the test sequences used for evaluation. The rendered semantic features are compared to manual ground-truth masks via mIoU (Tables 1 and 2), not to the same CAT-Seg features used in training, so the reported segmentation accuracy is not an identity or a fitted-value reproduction. The ablation study (Table 3) removes individual components and measures mIoU against the same manual ground truth; although the text's claim that removing components always decreases mIoU is contradicted by the liver row (79.43 vs. 79.06), that is an internal correctness issue, not a circularity. Self-citations appear (e.g., EndoDAC [4] for depth supervision in Eq. 6), but they are externally published, peer-reviewed tools used as fixed inputs and are not load-bearing for the semantic querying claim. The lack of a 2D oracle baseline is a meaningful evaluation gap, but it does not make any equation or conclusion equivalent to its own inputs by construction.

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

No new physical or ontological entities are proposed. The listed free parameters are manual thresholds and weights that affect mIoU and reconstruction; the main assumptions are the inherited deformation model, the reliability of pretrained depth, and the in-domain training of the VLM.

free parameters (4)
  • Query threshold e = 0.4
    Eq. 5: final mask is score >= e; no sensitivity analysis is reported, and mIoU depends on this cutoff.
  • Region size threshold for region-aware loss = 1000 pixels
    Eq. 5 applies L_RS only to semantic regions with |l| > 1000 pixels; the cutoff is set by hand with no ablation.
  • Sigmoid slope delta = 2.5
    Eq. 3: beta = 1/(1+exp(-delta*psi)); delta=2.5 is fixed without sensitivity analysis.
  • Loss weight lambda = 0.01
    Eq. 6: lambda weights TV and region-aware losses; set to 0.01 with no ablation.
assumptions (4)
  • domain assumption Deformation can be represented by the flexible deformation model of Eq. 1 (Fourier/polynomial basis with learnable centers and variances) from Deform3DGS.
    The method inherits this model for mean, rotation, scale, and semantic feature deformation; if tissue motion is discontinuous this representation may fail, and no analysis is provided.
  • domain assumption SAM masks and CAT-Seg features provide pixel-aligned semantic supervision that is consistent enough to train 3D Gaussians.
    Eq. 2 defines the semantic features from SAM masks and CAT-Seg; errors in these 2D features propagate into the 3D semantic field.
  • domain assumption Depth from the pretrained EndoDAC model is reliable enough to serve as metric depth supervision in Eq. 6.
    The final loss uses |D - D_hat|/D_hat as a geometric term; incorrect depth would distort Gaussian positions and the rendered feature field.
  • ad hoc to paper CAT-Seg is fine-tuned on CholecSeg8K and EndoVis18 and thus provides features that are well-aligned with the test-domain classes.
    Section 3.2 trains CAT-Seg on the same two datasets (excluding test sequences); this is a domain-specific assumption that makes the evaluation closer to closed-set recognition than open-vocabulary understanding.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SurgTPGS: Semantic 3D Surgical Scene Understanding with Text Promptable Gaussian Splatting." pith.science (2026). https://pith.science/paper/BPECTC5T

@misc{pith2026250623309,
  author       = {Pith},
  title        = {Pith review of: SurgTPGS: Semantic 3D Surgical Scene Understanding with Text Promptable Gaussian Splatting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BPECTC5T}},
  note         = {Machine review of arXiv:2506.23309}
}
read the original abstract

In contemporary surgical research and practice, accurately comprehending 3D surgical scenes with text-promptable capabilities is particularly crucial for surgical planning and real-time intra-operative guidance, where precisely identifying and interacting with surgical tools and anatomical structures is paramount. However, existing works focus on surgical vision-language model (VLM), 3D reconstruction, and segmentation separately, lacking support for real-time text-promptable 3D queries. In this paper, we present SurgTPGS, a novel text-promptable Gaussian Splatting method to fill this gap. We introduce a 3D semantics feature learning strategy incorporating the Segment Anything model and state-of-the-art vision-language models. We extract the segmented language features for 3D surgical scene reconstruction, enabling a more in-depth understanding of the complex surgical environment. We also propose semantic-aware deformation tracking to capture the seamless deformation of semantic features, providing a more precise reconstruction for both texture and semantic features. Furthermore, we present semantic region-aware optimization, which utilizes regional-based semantic information to supervise the training, particularly promoting the reconstruction quality and semantic smoothness. We conduct comprehensive experiments on two real-world surgical datasets to demonstrate the superiority of SurgTPGS over state-of-the-art methods, highlighting its potential to revolutionize surgical practices. SurgTPGS paves the way for developing next-generation intelligent surgical systems by enhancing surgical precision and safety. Our code is available at: https://github.com/lastbasket/SurgTPGS.

Figures

Figures reproduced from arXiv: 2506.23309 by the authors.

Figure 1
Figure 1. SurgTPGS provides 3D semantic-segmentation with text-promptable queries, complementing the drawbacks of 4DGS, VLM, and segmentation networks. 1 Introduction In modern computer-assisted interventions (CAI) procedures, accurately under￾standing 3D surgical scenes has become a cornerstone for successful surgical outcomes [16]. Previous studies [1, 5] have emphasized the significance of in￾corporating semantic segmentat… view at source ↗
Figure 2
Figure 2. Overview of SurgTPGS. We first extract semantic embedding from SAM [11] and VLM [3]; then, we train the deformable Gaussians with semantic-aware deforma￾tion tracking and semantic-region-aware optimization. SurgTPGS supports real-time semantic 3D query and novel-view rendering simultaneously. deal with deformable tissue challenges: ψ (t; Θ) = X B j=1 ωj ˜b (t; θj , σj ), | ˜b(t; θ, σ) = exp  − 1 2σ 2 (t − θ) 2  , … view at source ↗
Figure 3
Figure 3. Qualitative result on CholecSeg8K [5] and EndoVis18 [1] dataset. We demonstrate the results of LangSplat [18], OpenGaussian [24], and ours with CAT￾Seg [3] as VLM. Our method shows more consistent and precise segmentation. 3.2 Implementation Details We run all experiments on a single RTX4090 GPU. We utilize Adam optimizer with an initial learning rate of 1.6×10−3 . We train CAT-Seg [3] with the Cholec￾Seg8K [5] and … view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Visualization of the ablation results. Results from our full model present smoother boundaries and clear visualization quality [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

31 extracted references · 11 canonical work pages

  1. [1]

    arXiv preprint arXiv:2001.11190 (2020)

    Allan,M.,Kondo,S.,Bodenstedt,S.,Leger,S.,Kadkhodamohammadi,R.,Luengo, I., Fuentes, F., Flouty, E., Mohammed, A., Pedersen, M., et al.: 2018 robotic scene segmentation challenge. arXiv preprint arXiv:2001.11190 (2020)

    arXiv 2020

  2. [2]

    In: 2023 IEEE International Conference on Robotics and Automation (ICRA)

    Bai, L., Islam, M., Seenivasan, L., Ren, H.: Surgical-vqla: Transformer with gated vision-language embedding for visual question localized-answering in robotic surgery. In: 2023 IEEE International Conference on Robotics and Automation (ICRA). pp. 6859–6865. IEEE (2023)

    work page 2023

  3. [3]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Cho,S.,Shin,H.,Hong,S.,Arnab,A.,Seo,P.H.,Kim,S.:Cat-seg:Costaggregation for open-vocabulary semantic segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 4113–4123 (2024)

    2024

  4. [4]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Cui, B., Islam, M., Bai, L., Wang, A., Ren, H.: Endodac: Efficient adapting foun- dation model for self-supervised depth estimation from any endoscopic camera. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 208–218. Springer (2024)

    2024

  5. [5]

    arXiv preprint arXiv:2012.12453 (2020)

    Hong, W.Y., Kao, C.L., Kuo, Y.H., Wang, J.R., Chang, W.L., Shih, C.S.: Cholec- seg8k: a semantic segmentation dataset for laparoscopic cholecystectomy based on cholec80. arXiv preprint arXiv:2012.12453 (2020)

    arXiv 2020

  6. [6]

    arXiv preprint arXiv:2501.19319 (2025)

    Huang, Y., Cui, B., Bai, L., Chen, Z., Wu, J., Li, Z., Liu, H., Ren, H.: Advancing dense endoscopic reconstruction with gaussian splatting-driven surface normal- aware tracking and mapping. arXiv preprint arXiv:2501.19319 (2025)

    arXiv 2025

  7. [7]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Huang, Y., Cui, B., Bai, L., Guo, Z., Xu, M., Islam, M., Ren, H.: Endo-4dgs: Endo- scopic monocular scene reconstruction with 4d gaussian splatting. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 197–207. Springer (2024)

    2024

  8. [8]

    IEEE Transactions on Medical Imaging41(11), 2991–3002 (2022)

    Jin, Y., Yu, Y., Chen, C., Zhao, Z., Heng, P.A., Stoyanov, D.: Exploring intra-and inter-video relation for surgical semantic scene segmentation. IEEE Transactions on Medical Imaging41(11), 2991–3002 (2022)

    work page 2022

Show all 31 references
  1. [9]

    ACM Transactions on Graphics42(4) (2023)

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics42(4) (2023)

    2023

  2. [10]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Kerr, J., Kim, C.M., Goldberg, K., Kanazawa, A., Tancik, M.: Lerf: Language em- bedded radiance fields. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 19729–19739 (2023) 10 Y. Huang et al

    2023

  3. [11]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A.C., Lo, W.Y., et al.: Segment anything. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 4015–4026 (2023)

    2023

  4. [12]

    In: European Conference on Computer Vision

    Labe, I., Issachar, N., Lang, I., Benaim, S.: Dgd: Dynamic 3d gaussians distillation. In: European Conference on Computer Vision. pp. 361–378. Springer (2024)

    2024

  5. [13]

    arXiv preprint arXiv:2408.07981 (2024)

    Li,J.,Skinner,G.,Yang,G.,Quaranto,B.R.,Schwaitzberg,S.D.,Kim,P.C.,Xiong, J.: Llava-surg: towards multimodal surgical assistant via structured surgical video learning. arXiv preprint arXiv:2408.07981 (2024)

    2024 arXiv

  6. [14]

    IEEE Transac- tions on Medical Imaging43(4), 1308–1322 (2023)

    Liu, M., Han, Y., Wang, J., Wang, C., Wang, Y., Meijering, E.: Lskanet: Long strip kernel attention network for robotic surgical scene segmentation. IEEE Transac- tions on Medical Imaging43(4), 1308–1322 (2023)

    2023

  7. [15]

    arXiv preprint arXiv:2401.12561 (2024)

    Liu, Y., Li, C., Yang, C., Yuan, Y.: Endogaussian: Gaussian splatting for de- formable surgical scene reconstruction. arXiv preprint arXiv:2401.12561 (2024)

    2024 arXiv

  8. [16]

    McLachlan, G.: From 2d to 3d: the future of surgery? The Lancet378(9800), 1368 (2011)

    2011

  9. [17]

    Commu- nications of the ACM65(1), 99–106 (2021)

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. Commu- nications of the ACM65(1), 99–106 (2021)

    2021

  10. [18]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Qin, M., Li, W., Zhou, J., Wang, H., Pfister, H.: Langsplat: 3d language gaussian splatting. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 20051–20060 (2024)

    2024

  11. [19]

    In: International conference on machine learning

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. In: International conference on machine learning. pp. 8748–8763. PmLR (2021)

    2021

  12. [20]

    In: International Conference on Med- ical Image Computing and Computer-Assisted Intervention

    Seenivasan,L.,Islam,M.,Krishna,A.K.,Ren,H.:Surgical-vqa:Visualquestionan- swering in surgical scenes using transformer. In: International Conference on Med- ical Image Computing and Computer-Assisted Intervention. pp. 33–43. Springer (2022)

    2022

  13. [21]

    IEEE transactions on medical imaging36(1), 86–97 (2016)

    Twinanda, A.P., Shehata, S., Mutter, D., Marescaux, J., De Mathelin, M., Padoy, N.: Endonet: a deep architecture for recognition tasks on laparoscopic videos. IEEE transactions on medical imaging36(1), 86–97 (2016)

    2016

  14. [22]

    IEEE Transactions on Medical Imaging (2024)

    Wang, H., Yang, G., Zhang, S., Qin, J., Guo, Y., Xu, B., Jin, Y., Zhu, L.: Video- instrument synergistic network for referring video instrument segmentation in robotic surgery. IEEE Transactions on Medical Imaging (2024)

    2024

  15. [23]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Wang, Y., Long, Y., Fan, S.H., Dou, Q.: Neural rendering for stereo 3d recon- struction of deformable tissues in robotic surgery. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 431–441. Springer (2022)

    2022

  16. [24]

    arXiv preprint arXiv:2406.02058 (2024)

    Wu, Y., Meng, J., Li, H., Wu, C., Shi, Y., Cheng, X., Zhao, C., Feng, H., Ding, E., Wang, J., et al.: Opengaussian: Towards point-level 3d gaussian-based open vocabulary understanding. arXiv preprint arXiv:2406.02058 (2024)

    2024 arXiv

  17. [25]

    In: Interna- tional Conference on Medical Image Computing and Computer-Assisted Interven- tion

    Yang, S., Li, Q., Shen, D., Gong, B., Dou, Q., Jin, Y.: Deform3dgs: Flexible defor- mation for fast surgical scene reconstruction with gaussian splatting. In: Interna- tional Conference on Medical Image Computing and Computer-Assisted Interven- tion. pp. 132–142. Springer (2024)

    2024

  18. [26]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Yoon, J., Hong, S., Hong, S., Lee, J., Shin, S., Park, B., Sung, N., Yu, H., Kim, S., Park, S., et al.: Surgical scene segmentation using semantic image synthesis with a virtual surgery environment. In: International Conference on Medical Image Computing and Computer-Assisted ...

    2022

  19. [27]

    Yuan, K., Srivastav, V., Yu, T., Lavanchy, J., Mascagni, P., Navab, N., Padoy, N.: Learning multi-modal representations by watching hundreds of surgical video lectures (2023)

    2023

  20. [28]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Zha, R., Cheng, X., Li, H., Harandi, M., Ge, Z.: Endosurf: Neural surface re- construction of deformable tissues with stereo endoscope videos. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 13–23. Springer (2023)

    2023

  21. [29]

    Advances in Neural Infor- mation Processing Systems36, 28611–28623 (2023)

    Zhou, Z., Alabi, O., Wei, M., Vercauteren, T., Shi, M.: Text promptable surgical instrument segmentation with vision-language models. Advances in Neural Infor- mation Processing Systems36, 28611–28623 (2023)

    2023

  22. [30]

    arXiv preprint arXiv:2401.11535 (2024)

    Zhu, L., Wang, Z., Jin, Z., Lin, G., Yu, L.: Deformable endoscopic tissues recon- struction with gaussian splatting. arXiv preprint arXiv:2401.11535 (2024)

    2024 arXiv

  23. [31]

    In: Proceedings of the 28th annual conference on Computer graphics and interactive techniques

    Zwicker, M., Pfister, H., Van Baar, J., Gross, M.: Surface splatting. In: Proceedings of the 28th annual conference on Computer graphics and interactive techniques. pp. 371–378 (2001)

    2001

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.