Pith. sign in

REVIEW 2 major objections 6 minor 2 cited by

ClipGS: Clippable Gaussian Splatting for Interactive Cinematic Visualization of Volumetric Medical Data

T0 review · 2 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 A single compressed Gaussian splatting model can render cinematic-quality volumetric medical data with interactive clipping, reaching an average PSNR of 36.635 at 156 FPS with a 16.1 MB model.

desk verdict A clean, competent clipping-plane extension of 3DGS for medical volumes, but the general interactivity claim outruns the z-axis-only experiments. read the letter →

arxiv 2507.06647 v1 pith:IWYMEVWY submitted 2025-07-09 cs.CV

Chengkun Li , Yuqi Tong , Kai Chen , Zhenya Yang , Ruiyang Li , Shi Qiu , Jason Ying-Kuen Chan , Pheng-Ann Heng
show 1 more author
Qi Dou
This is my paper · ORCID
classification cs.CV
keywords volumetricmedicaldatavisualizationinteractivecinematicrenderingclippableGaussiansplattinglearnabletruncationadaptiveadjustmentmodel3Dclippingplaneinteraction
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 tries to establish that interactive clipping of cinematic medical volumes does not require a separate model per clipping layer. It introduces a learnable scalar per Gaussian that records which side of the clip plane the Gaussian lies on, so visibility is decided by a simple comparison with the plane offset, and then deforms the visible Gaussians with a small MLP as the plane moves. On five ray-traced scenes from CT and anatomical slice data, ClipGS reports an average PSNR of 36.635, 156 FPS, and 16.1 MB storage, outperforming four dynamic-scene baselines on quality, speed, and model size. If correct, this gives surgeons and students a fast way to slice into photorealistic anatomy during planning and teaching.

What carries the argument

The load-bearing object is the learnable truncation (LT) scheme on each Gaussian primitive, formalized as $m = (\mu+\delta)\cdot n$ with visibility $M = \mathbb{1}[m<z]$ trained through the straight-through estimator $\mathrm{sg}(\mathbb{1}[\sigma(m-z)<\epsilon]-\sigma(m-z))+\sigma(m-z)$. It lets the optimizer decide, for every Gaussian, whether it belongs to the visible side of the clip plane while keeping the visibility decision differentiable. The second mechanism is the adaptive adjustment model (AAM), a tiny multi-layer perceptron with positional encoding that takes Gaussian positions and the clip offset $z$ and emits deformations $\Delta\mu,\Delta r,\Delta s$, so the Gaussians touching the clip surface are reshaped to match the new interior view. Training happens in two steps: first the GS with LT alone for a coarse model, then joint optimization with the AAM to refine quality and preserve continuity across clip positions.

What would settle it

Take a trained scene, move the clipping plane to an orientation never seen in training (for example a sagittal instead of axial cut), render it, and compare with the ground-truth ray-traced image; if PSNR drops far below the reported 36.6 average, the claimed general interactive clipping support fails for rotated planes.

Watch

Extended reading notes

Core claim

The central discovery is that clipping-plane interactivity can be baked into Gaussian splatting by adding a single optimizable visibility scalar to every primitive. Writing the plane as a fixed normal direction $n$ and a scalar offset $z$, the visibility of a Gaussian is the step function $M = \mathbb{1}[m < z]$ with $m = (\mu+\delta)\cdot n$ the projected position of its effective barycenter; the hard step is trained through a straight-through estimator so the scalar $m$ is learned without perturbing the gradient of the mean position. To keep renderings consistent as the plane moves, a tiny MLP with positional encoding, conditioned on the Gaussian positions and the plane offset $z$, predicts position, rotation, and scale adjustments for the visible Gaussians. The resulting model reaches an average 36.635 PSNR, 156 FPS, and 16.1 MB size on five scenes, exceeding the compared dynamic-scene methods that treat the clip offset as a time-like coordinate.

Load-bearing premise

The method assumes that a user's clipping interaction is fully captured by a scalar offset along one fixed plane normal, and that each Gaussian is either entirely visible or entirely invisible, so clipping planes with other orientations or partial visibility are outside what was trained and tested.

Editorial extensions

If this is right

  • Because visibility is a per-Gaussian scalar, one trained ClipGS model can render any clip offset along the trained axis instead of storing separate Gaussian layers for each clip position.
  • The reported 156 FPS means clip-plane exploration stays interactive after training, which is the practical requirement that photon-mapping-based cinematic volume renderers fail to meet on consumer GPUs.
  • Compared with hard truncation based on Gaussian means, learnable truncation improves PSNR by about 0.38 dB on the authors' ablation, indicating that optimizing the effective barycenter offset $\delta$ reduces artifacts at the clip surface.
  • Treating the clip offset as a time-like coordinate lets ClipGS be compared to dynamic-scene Gaussian methods, and the reported numbers beat the strongest such baseline (4DGaussians) in PSNR, FPS, and storage.

Reading between the lines

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

  • A direct test the paper leaves open is rotating the clipping plane: since training only moves the plane along one normal, a held-out set with arbitrary plane orientations would likely expose a drop in quality unless $m$ becomes a vector or the normal is part of the conditioning; this is an extension the paper does not claim.
  • The same scalar-truncation trick could in principle handle other binary exploration tools such as clip cubes, eraser brushes, or opacity masking by replacing the scalar $z$ comparison with a signed distance to the tool volume; this is an editorial inference, not a paper claim.
  • Because each Gaussian is either fully visible or invisible, very fine vessels or membranes near the cut surface may be lost if no Gaussian center remains on the visible side; a multi-level or opacity-based visibility variant is a natural follow-up that the paper does not explore.
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

2 major / 6 minor

Summary. The paper presents ClipGS, a 3D Gaussian Splatting framework for rendering volumetric medical images with an interactive clipping plane. The central idea is to model the clipping coordinate as an additional condition analogous to time in dynamic-scene Gaussians. Visibility is controlled by a per-Gaussian learnable scalar m compared with clipping offset z through a straight-through-estimator sigmoid gate (Eq. 4), and a small MLP predicts deformations of position, rotation, and scale as a function of Gaussian position and z (Eqs. 5-6). Training uses ray-traced cinematic images with random views and random clipping offsets; the authors report an average 36.635 PSNR, 156 FPS, and 16.1 MB storage on five CT/anatomical-slice scenes, outperforming HexPlane, GauFRe, N-DG, and 4DGaussians. Ablations separate the learnable truncation and adaptive adjustment contributions.

Significance. If the reported results transfer beyond the current protocol, ClipGS is a practical contribution: a single compressed Gaussian model supports clipping-style interaction with a modest MLP overhead, and it avoids per-layer Gaussian stacks. The experimental design is largely sound: training and test frames are separate; the two proposed components are ablated; baseline comparisons report rendering quality, speed, and storage on the same dataset. I found no circularity in the learning pipeline: the visibility threshold and MLP weights are optimized from held-out-condition training data, not fitted to test outputs. The main reservation is scope: the evidence covers only clipping planes whose normal is kept fixed to the z-axis, so the claim of general interactive clipping-plane support is broader than the experiments. With the requested experiments or a narrowed claim, the work would be a solid contribution.

major comments (2)
  1. [§2.2, §2.3, §3.1] The central claim of interactive clipping support is only tested for clipping offsets along a fixed z-axis, not for general clipping-plane orientations. In Eq. 4 the visibility gate compares the scalar m with the scalar z, and Eq. 5 feeds only the scalar z into the deformation MLP; the plane normal n from Eq. 3 never appears as a network input. Consistently, Sec. 3.1 states that training renders use "random clipping plane position along the z-axis," i.e., the normal is fixed. For a rotated plane, the required visibility boundary is the signed distance (μ+δ)·n, which cannot be recovered from a per-Gaussian scalar m learned for one normal, and the deformation network has no conditioning signal carrying the new normal. The abstract and introduction claim general clipping-plane interaction rather than z-axis translation, so this is a load-bearing gap. I would like to see either (i) training and evaluation with randomly oriented clipping planes and conditioning on plane parameters, or (ii) an explicit statement that the contribution is limited to translation of a clipping plane with fixed orientation.
  2. [§3.1, Table 1] The user-facing claim of "interactive cinematic visualization" is validated only by offline FPS on synthetic ray-traced dataset frames; no evaluation includes a real-time interactive session, actual medical imaging software, or latency under interactive plane manipulation. In addition, the paper does not state whether the reported 156 FPS corresponds to the full ClipGS pipeline including the deformation MLP or only to the rasterization pass. As written, the protocol makes it difficult to know whether the reported interactivity carries over to the practical use case emphasized in the introduction. The authors should clarify the timing measurement and, ideally, add a simple real-volume experiment or an interactive-latency measurement to support the title-level claim.
minor comments (6)
  1. [§2.1] The scale s is written as a scalar (s ∈ R), although standard 3DGS uses an anisotropic scale vector in R3; please correct the notation and the corresponding covariance decomposition.
  2. [§2.3, Eq. 6] The deformation heads are listed as φμ, φγ, and φs, but Eq. 6 writes Δμ = φs(f) where the scale head is meant; this should be Δs = φs(f).
  3. [§2.3] The method is called "Continuous Clipping," but Eq. 4 is discontinuous because of the indicator operation; the claimed continuity of rendering across clipping offsets should be argued explicitly or demonstrated quantitatively.
  4. [§3.1] "Random clipping plane position along the z-axis" should be phrased as "random clipping offset along the z-axis with the plane normal fixed to (0,0,1)" to avoid ambiguity about plane orientation.
  5. [§3.3, Table 2] There is a typo: "trauncation" should be "truncation."
  6. [§2.2, §3.1] The initialization of the added attribute m is not specified, and the fixed hyperparameters ε=0.5 and λ=0.2 are not ablated; a sensitivity study for ε would strengthen the learnable-truncation section.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the method is trained on ground-truth rendered images and evaluated on held-out views and clipping offsets, and the learned visibility scalar is an ordinary optimizable parameter rather than a fitted restatement of the evaluation metric.

full rationale

The central derivation chain is self-contained against an external supervision signal. The paper renders a cinematic image sequence with a ray-tracing renderer, optimizes Gaussians with a learnable visibility attribute m and an MLP-based deformation model, and evaluates on 100 held-out frames per scene at novel views and novel clipping-plane offsets. The visibility rule in Eq. 4, M = sg(1[sigma(m - z) < eps] - sigma(m - z)) + sigma(m - z), treats m as an optimizable per-Gaussian parameter, and the deformation in Eq. 5, f = MLP(gamma(mu), gamma(z)), is also learned from the same external image loss in Eq. 7. Neither quantity is defined in terms of the reported test PSNR/SSIM/LPIPS, so the 'prediction' of held-out rendering quality does not reduce by construction to a fitted input. The hard-truncation baseline in Eq. 3 is explicitly compared against the learnable scheme, showing that the learnable scalar is not merely the geometric expression restated. The comparison methods are external baselines, and the cited references [13, 16, 19] are standard background (3DGS, STE, NeRF positional encoding) rather than load-bearing self-citations. The main weakness noted in the reader's take, that training uses random clipping planes along the z-axis only while the claim mentions general clipping-plane support, is a scope or generalization limitation, not a circularity: it concerns whether the learned representation transfers to untrained plane orientations, which is an empirical correctness risk. Per the review rules, such non-standard-consensus or scope concerns belong under correctness risk, not circularity. No self-referential limitation passage or unusual inserted statement asserts a circular step, missing support, or omitted proof. Overall, the derivation is independent of its conclusions, so the circularity score is 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on modeling assumptions: that Gaussian splatting is a good proxy for cinematic volume rendering, that a scalar threshold per Gaussian captures clipping, and that the synthetic renderer is representative. The method also has hand-set hyperparameters. No new physical entities are introduced.

free parameters (3)
  • epsilon (ST threshold) = 0.5
    Hand-set threshold in Eq. 4 for the straight-through estimator; controls smoothing of the visibility step and is not optimized.
  • lambda (loss weight) = 0.2
    Hand-set weight balancing L1 and D-SSIM in Eq. 7; inherited from 3DGS and not swept.
  • initial Gaussian count = 100,000
    Random initialization density chosen by hand; affects reconstruction capacity and is not swept.
assumptions (4)
  • domain assumption 3D Gaussian splatting with alpha blending can faithfully represent cinematic medical renderings.
    The method builds on 3DGS (Sec 2.1) and assumes the ray-traced cinematic images can be approximated by point-based alpha-blended Gaussians; no error term for this approximation is given.
  • ad hoc to paper The clipping plane can be treated as an extra dimension analogous to time in dynamic scene models, and a single scalar m per Gaussian suffices for visibility across all clipping offsets.
    This assumption underlies Eqs. 3-5 and the choice of baselines; it is plausible but only tested for one plane normal.
  • domain assumption Cinematic reference images generated by Blender/BioxelNodes are a valid proxy for clinically relevant cinematic rendering.
    All training and testing is on this synthetic data (Sec 3.1); the paper does not validate against real CT cinematic renderings.
  • standard math Straight-through estimator provides a usable gradient for the binary visibility step.
    Used in Eq. 4, widely accepted in discrete latent modeling; not proven but standard.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ClipGS: Clippable Gaussian Splatting for Interactive Cinematic Visualization of Volumetric Medical Data." pith.science (2026). https://pith.science/paper/IWYMEVWY

@misc{pith2026250706647,
  author       = {Pith},
  title        = {Pith review of: ClipGS: Clippable Gaussian Splatting for Interactive Cinematic Visualization of Volumetric Medical Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IWYMEVWY}},
  note         = {Machine review of arXiv:2507.06647}
}
read the original abstract

The visualization of volumetric medical data is crucial for enhancing diagnostic accuracy and improving surgical planning and education. Cinematic rendering techniques significantly enrich this process by providing high-quality visualizations that convey intricate anatomical details, thereby facilitating better understanding and decision-making in medical contexts. However, the high computing cost and low rendering speed limit the requirement of interactive visualization in practical applications. In this paper, we introduce ClipGS, an innovative Gaussian splatting framework with the clipping plane supported, for interactive cinematic visualization of volumetric medical data. To address the challenges posed by dynamic interactions, we propose a learnable truncation scheme that automatically adjusts the visibility of Gaussian primitives in response to the clipping plane. Besides, we also design an adaptive adjustment model to dynamically adjust the deformation of Gaussians and refine the rendering performance. We validate our method on five volumetric medical data (including CT and anatomical slice data), and reach an average 36.635 PSNR rendering quality with 156 FPS and 16.1 MB model size, outperforming state-of-the-art methods in rendering quality and efficiency.

Figures

Figures reproduced from arXiv: 2507.06647 by the authors.

Figure 1
Figure 1. Overview of our proposed framework. Our ClipGS implements a real-time interactive cinematic rendering under the query condition (view direc￾tion & clipping plane) for a given volumetric medical data. – We propose a learnable truncation (LT) scheme, which can dynamically con￾trol the visibility of each Gaussian primitive with the query clipping plane. – We design an adaptive adjustment model (AAM) to adjust the defor… view at source ↗
Figure 2
Figure 2. Qualitative results of rendered images on our cinematic medical dataset. The PSNRs in the images denote the rendering quality for the whole case. great performance in training time, reducing storage and also rendering speed, as shown in Tab. 1 and [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Hybrid Foveated Path Tracing with Peripheral Gaussians for Immersive Anatomy

    cs.GR 2026-01 conditional novelty 6.0 of 10

    A hybrid VR renderer combines foveated volumetric path tracing with a rapidly regenerated Gaussian-splatting periphery for interactive medical anatomy visualization.

  2. Gaussian Volumetric Representation for Efficient Shear-Warp Visualization

    cs.CV 2026-07 conditional novelty 5.0 of 10

    A Gaussian-blob volume field trained from sparse voxel and slice supervision reconstructs MRI and cryosection volumes well enough to render them in real time at up to ~44 FPS and ~11:1 compression.

Reference graph

Works this paper leans on

32 extracted references · 30 canonical work pages · cited by 2 Pith papers

  1. [1]

    Proceedings of the IEEE86(3), 504–511 (1998)

    Ackerman, M.: The visible human project. Proceedings of the IEEE86(3), 504–511 (1998). https://doi.org/10.1109/5.662875 6

    doi:10.1109/5.662875 1998

  2. [2]

    arXiv preprint arXiv:1308.3432 (2013) 4

    Bengio, Y., Léonard, N., Courville, A.: Estimating or propagating gradi- ents through stochastic neurons for conditional computation. arXiv preprint arXiv:1308.3432 (2013) 4

    arXiv 2013

  3. [3]

    Annals of Anatomy-Anatomischer Anzeiger222, 159– 165 (2019) 1, 2

    Binder, J., Krautz, C., Engel, K., Grützmann, R., Fellner, F.A., Burger, P.H., Scholz, M.: Leveraging medical imaging for medical education—a cinematic rendering-featured lecture. Annals of Anatomy-Anatomischer Anzeiger222, 159– 165 (2019) 1, 2

    work page 2019

  4. [4]

    In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion

    Cao, A., Johnson, J.: Hexplane: A fast representation for dynamic scenes. In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion. pp. 130–141 (2023) 6, 8

    work page 2023

  5. [5]

    Comaniciu, D., Engel, K., Georgescu, B., Mansi, T.: Shaping the future through innovations: From medical imaging to precision medicine (2016) 2

    work page 2016

  6. [6]

    Blender Foun- dation, Stichting Blender Foundation, Amsterdam (2018),http://www.blender

    Community, B.O.: Blender - a 3D modelling and rendering package. Blender Foun- dation, Stichting Blender Foundation, Amsterdam (2018),http://www.blender. org 6

    work page 2018

  7. [7]

    In: ACM SIGGRAPH 2024 Conference Papers

    Diolatzis, S., Zirr, T., Kuznetsov, A., Kopanas, G., Kaplanyan, A.: N-dimensional gaussians for fitting of high dimensional functions. In: ACM SIGGRAPH 2024 Conference Papers. pp. 1–11 (2024) 4, 6, 8

    work page 2024

  8. [8]

    American Journal of Roentgenology 209(2), 370–379 (2017) 1, 2

    Eid, M., De Cecco, C.N., Nance Jr, J.W., Caruso, D., Albrecht, M.H., Spandorfer, A.J., De Santis, D., Varga-Szemes, A., Schoepf, U.J.: Cinematic rendering in ct: a novel, lifelike 3d visualization technique. American Journal of Roentgenology 209(2), 370–379 (2017) 1, 2

    work page 2017

Show all 32 references
  1. [9]

    JAMA surgery154(8), 738–744 (2019) 2

    Elshafei, M., Binder, J., Baecker, J., Brunner, M., Uder, M., Weber, G.F., Grütz- mann, R., Krautz, C.: Comparison of cinematic rendering and computed tomog- raphy for speed and comprehension of surgical anatomy. JAMA surgery154(8), 738–744 (2019) 2

    2019

  2. [10]

    World neurosurgery114, e283–e292 (2018) 1

    Glemser, P.A., Engel, K., Simons, D., Steffens, J., Schlemmer, H.P., Orakcioglu, B.: A new approach for photorealistic visualization of rendered computed tomography images. World neurosurgery114, e283–e292 (2018) 1

    2018

  3. [11]

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

    Heinrich, F., Bornemann, K., Lawonn, K., Hansen, C.: Interacting with medical volume data in projective augmented reality. In: International Conference on Med- ical Image Computing and Computer-Assisted Intervention. pp. 429–439. Springer (2020) 2

    2020

  4. [12]

    In: EuroVis (Short Papers)

    Jabbireddy, S., Li, S., Meng, X., Terrill, J.E., Varshney, A.: Accelerated volume rendering with volume guided neural denoising. In: EuroVis (Short Papers). pp. 49–53 (2023) 1

    2023

  5. [13]

    ACM Transactions on Graphics 42(4) (July 2023), https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/ 2, 3, 6 10 C

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics 42(4) (July 2023), https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/ 2, 3, 6 10 C. Li et al

    2023

  6. [14]

    arXiv preprint arXiv:2410.16978 (2024) 2, 5

    Kleinbeck, C., Schieber, H., Engel, K., Gutjahr, R., Roth, D.: Multi-layer gaussian splatting for immersive anatomy visualization. arXiv preprint arXiv:2410.16978 (2024) 2, 5

    2024 arXiv

  7. [15]

    Lafortune, E.P., Willems, Y.D.: Bi-directional path tracing (1993) 2

    1993

  8. [16]

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

    Lee, J.C., Rho, D., Sun, X., Ko, J.H., Park, E.: Compact 3d gaussian representation for radiance field. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 21719–21728 (2024) 4

    2024

  9. [17]

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

    Li, T., Slavcheva, M., Zollhoefer, M., Green, S., Lassner, C., Kim, C., Schmidt, T., Lovegrove, S., Goesele, M., Newcombe, R., et al.: Neural 3d video synthesis from multi-view video. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 55...

    2022

  10. [18]

    In: WACV (2025) 4, 6, 8

    Liang, Y., Khan, N., Li, Z., Nguyen-Phuoc, T., Lanman, D., Tompkin, J., Xiao, L.: Gaufre: Gaussian deformation fields for real-time dynamic novel view synthesis. In: WACV (2025) 4, 6, 8

    2025

  11. [19]

    Commu- nications of the ACM65(1), 99–106 (2021) 2, 5

    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) 2, 5

    2021

  12. [20]

    Vision, Modeling, and Visualization, L

    Niedermayr, S., Neuhauser, C., Petkov, K., Engel, K., Westermann, R.: Applica- tion of 3d gaussian splatting for cinematic anatomy on consumer class devices. Vision, Modeling, and Visualization, L. Linsen and J. Thies, Eds. The Eurograph- ics Association (2024) 2, 5

    2024

  13. [21]

    OmooLab: Bioxelnodes (2025), https://github.com/OmooLab/BioxelNodes, ac- cessed: 2025-02-27 6

    2025

  14. [22]

    Advances in neural information processing sys- tems 32 (2019) 6

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al.: Pytorch: An imperative style, high- performance deep learning library. Advances in neural information processing sys- tems 32 (2019) 6

    2019

  15. [23]

    In: 2025 IEEE International Conference on Artificial Intelligence and eXtended and Virtual Reality (AIxVR)

    Qiu, S., Xie, B., Liu, Q., Heng, P.A.: Advancing extended reality with 3d gaussian splatting: Innovations and prospects. In: 2025 IEEE International Conference on Artificial Intelligence and eXtended and Virtual Reality (AIxVR). pp. 203–208 (2025) 2

    2025

  16. [24]

    In: 2025 IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW)

    Qiu, S., Xie, B., Liu, Q., Heng, P.A.: Creating virtual environments with 3d gaus- sian splatting: A comparative study. In: 2025 IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW). pp. 1332–1333 (2025) 2

    2025

  17. [25]

    Annals of Anatomy- Anatomischer Anzeiger 241, 151905 (2022) 1

    Steffen, T., Winklhofer, S., Starz, F., Wiedemeier, D., Ahmadli, U., Stadlinger, B.: Three-dimensional perception of cinematic rendering versus conventional vol- ume rendering using ct and cbct data of the facial skeleton. Annals of Anatomy- Anatomischer Anzeiger 241, 151905 (2022) 1

    2022

  18. [26]

    In: 2024 IEEE Conference Virtual Reality and 3D User Interfaces (VR)

    Taibo, J., Iglesias-Guitian, J.A.: Immersive 3d medical visualization in virtual re- ality using stereoscopic volumetric path tracing. In: 2024 IEEE Conference Virtual Reality and 3D User Interfaces (VR). pp. 1044–1053. IEEE (2024) 2

    2024

  19. [27]

    Veach,E.,Guibas,L.:Bidirectionalestimatorsforlighttransport.In:Photorealistic Rendering Techniques, pp. 145–167. Springer (1995) 2

    1995

  20. [28]

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

    Wu, G., Yi, T., Fang, J., Xie, L., Zhang, X., Wei, W., Liu, W., Tian, Q., Wang, X.: 4d gaussian splatting for real-time dynamic scene rendering. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 20310–20320 (2024) 4, 6, 8

    2024

  21. [29]

    Yuan, Y., Yang, J., Sun, Q., Huang, Y., Ma, S.: Cinematic volume rendering algo- rithmbasedonmultiplelightsphotonmapping.MultimediaToolsandApplications 83(2), 5799–5812 (2024) 2 Title Suppressed Due to Excessive Length 11

    2024

  22. [30]

    Journal of digital imaging24, 640–664 (2011) 1

    Zhang, Q., Eagleson, R., Peters, T.M.: Volume visualization: a technical overview with a focus on medical applications. Journal of digital imaging24, 640–664 (2011) 1

    2011

  23. [31]

    In: 2005 IEEE Engineering in Medicine and Biology 27th Annual Conference

    Zhang, S.X., Heng, P.A., Liu, Z.J.: Chinese visible human project: Dataset acqui- sition and its primary applications. In: 2005 IEEE Engineering in Medicine and Biology 27th Annual Conference. pp. 4168–4170 (2005) 6

    2005

  24. [32]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Zhu, R., Qiu, S., Liu, Z., Hui, K.H., Wu, Q., Heng, P.A., Fu, C.W.: Rethinking end-to-end 2d to 3d scene segmentation in gaussian splatting. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 3656–3665 (2025) 2

    2025

Pith tools

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