pith. sign in

arxiv: 2606.20531 · v1 · pith:J65LNSA4new · submitted 2026-06-18 · 💻 cs.CV

VisDom: Sparse Novel View Synthesis with Visible Domain Constraint

Mariia Gladkova* , Tarun Yenamandra* , Edmond Boyer , Robert Maier , Tony Tung , Daniel Cremers This is my paper

Pith reviewed 2026-06-26 17:44 UTC · model grok-4.3

classification 💻 cs.CV
keywords sparse novel view synthesisvisual hullvisibility constraintNeRFGaussian splattingmulti-view geometrysilhouette carving
0
0 comments X

The pith

A minimum multi-view visibility requirement strengthens the visual hull prior for sparse novel view synthesis.

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

The paper introduces VisDom to address overfitting and floating artifacts that arise when NeRF or Gaussian splatting methods receive only a handful of input images. It augments classical silhouette carving by defining a visible domain consisting of those 3D points observed by at least K of the input cameras and then restricts volumetric sampling or Gaussian placement to lie inside that domain. This extra geometric filter is learning-free, requires only silhouettes, and is shown to produce cleaner object-centric reconstructions on three datasets even when as few as four views are available. The same constraint can be dropped on top of existing pipelines such as GaussianObject to raise their accuracy while cutting training cost by more than an order of magnitude.

Core claim

VisDom defines the visible domain as the subset of 3D space that lies inside the viewing frustum of at least K input cameras. This domain is used as an additional hard filter on top of standard visual-hull carving: any point outside the domain is excluded from volumetric sampling in NeRF-style optimization and from candidate Gaussian placement in explicit splatting pipelines. The resulting spatial prior reduces the feasible set of geometries that remain consistent with the input silhouettes, thereby limiting the solutions that can produce floating artifacts in sparse-view settings.

What carries the argument

The visible domain, the 3D region observed by at least K input views, used as a hard filter on sampling and placement.

If this is right

  • High-quality object-centric novel-view synthesis becomes possible from only four input images on three challenging datasets.
  • The same constraint can be added to both implicit volumetric and explicit point-based pipelines without introducing learned parameters.
  • Training cost can be reduced by more than twenty times while matching or exceeding the accuracy of a stronger baseline.
  • Only binary silhouettes are needed, making the method domain-agnostic.

Where Pith is reading between the lines

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

  • The same visibility filter could be applied during test-time optimization or to other reconstruction algorithms that already use silhouette consistency.
  • Varying K per scene or per region might trade off completeness against artifact suppression in a controllable way.
  • Because the domain is computed from camera poses and silhouettes alone, it could serve as a fast pre-filter before more expensive learned regularizers are applied.

Load-bearing premise

Requiring visibility in at least K views removes more incorrect geometry than it discards valid surface points that happen to be seen in fewer views.

What would settle it

A quantitative comparison, on a dataset with known ground-truth geometry, showing that the method either produces artifacts inside the true object or removes correct surface regions when K is set to the value used in the experiments.

Figures

Figures reproduced from arXiv: 2606.20531 by Daniel Cremers, Edmond Boyer, Mariia Gladkova*, Robert Maier, Tarun Yenamandra*, Tony Tung.

Figure 1
Figure 1. Figure 1: Our learning-free geometric constraint, derived purely from silhouettes, enables reconstruction from as few as 4 images. Given 4 inputs (left), ZipNeRF [2] and 3DGS￾GO [29] (cols 1-2) struggle without our constraint; adding VisDom (cols 3-4) recovers high-quality reconstructions with zero additionally learned parameters. Abstract. Sparse novel view synthesis (NVS) remains challenging due to the ambiguity o… view at source ↗
Figure 2
Figure 2. Figure 2: fig. 2. In other words, silhouettes alone cannot fully resolve the depth uncertainty [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 2
Figure 2. Figure 2: Visual hull reconstruction in 2D (left part) and top-view in 3D (right part). Left part: Given 2 silhouettes of an object (left), the visual hull is reconstructed (middle) by carving out space (white) as per the silhouettes. Our visual domain constraint removes additional space (right). Right part: Given 2 posed silhouettes (left), the visual hull is reconstructed with unwanted regions (middle left). Some … view at source ↗
Figure 3
Figure 3. Figure 3: A 2D example of the visible domain (purple) for different values of K in a 3-camera setup. Larger K values narrow the covered re￾gion, leading to more precise shapes. Volumetric rendering Ray bounds Rendered mask Rendered color vhull with VisDom Constrained volumetric rendering [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison of methods trained with our constraint and other NVS approaches in sparse settings on three diverse, challenging datasets, MipNeRF360, Omni3D, and ActorsHQ. Our constraint improves state-of-the-art sparse reconstruc￾tion methods such as CoR-GS and 3DGS-GO, and enables existing NeRF methods to perform comparably to sparse-view baselines. Best seen zoomed-in. 5.3 VisDom on Existing NeR… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative evaluation of our constraints on NeRF and 3DGS methods. Our constraints show dramatic improvement in rendering quality for NeRF-based methods. Due to our visible domain regularization, we observe fewer artifacts, especially at object boundaries in 3DGS-based reconstructions. Best seen zoomed-in. which visually demonstrate the insufficiency of silhouettes for NeRF methods in sparse settings and … view at source ↗
Figure 7
Figure 7. Figure 7: Effect of mask dilation on ZipNeRF + VisDom for MipNeRF360 dataset, mea￾sured in mean PSNR (dB) averaged over bonsai, garden, and kitchen. The r=0 point corresponds to the results reported in the main paper. Dilation improves the perfor￾mance over the no-dilation baseline, suggesting the benefit of applying small padding to the mask boundary as mitigation for their multi-view inconsistencies. Although the … view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative results of the visual hull reconstructed using different algorithms, from left to right: traditional, unit sphere heuristic, and with our VisDom constraint [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative comparisons with VaxNeRF and VaxNeRF∗ (visual hull used for inference) models on 12, 9, and 9 cameras respectively for the three datasets [PITH_FULL_IMAGE:figures/full_fig_p023_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Novel view synthesis from ZipNeRF and Instant-NGP models when trained with our constraint based on different visual hull algorithms. When trained with the visual hull reconstructed with our constraint, the models perform the best in most settings [PITH_FULL_IMAGE:figures/full_fig_p024_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Additional qualitative results on novel view synthesis from sparse views (4, 6, and 9) from diverse, challenging datasets — MipNeRF360, Omni3D, and ActorsHQ. As can be seen, our geometric constraint helps ZipNeRF and Instant-NGP achieve high-quality reconstructions from a sparse set of images. Further, our constraint also helps improve the quality of reconstructions of SOTA 3DGS-based methods, such as a r… view at source ↗
read the original abstract

Sparse novel view synthesis (NVS) remains challenging due to the ambiguity of recovering 3D geometry from few input views. While NeRF- and Gaussian Splatting (GS)-based methods perform well with dense supervision, they often overfit in sparse settings, producing floating artifacts and inconsistent geometry. Silhouette consistency is commonly used as a regularizer, but it remains insufficient, as silhouette-consistent regions can extend beyond the true object geometry. We introduce VisDom, a learning-free geometric constraint that augments classical carving-based visual hull reconstruction by enforcing a minimum multi-view visibility requirement. Specifically, we define a visible domain as the subset of 3D space observed by at least $K$ views and use it as an additional filtering criterion on top of standard silhouette-based reconstruction. This provides a stronger spatial prior in sparse-view settings. We integrate VisDom into both implicit (NeRF) and explicit (GS) pipelines by restricting volumetric sampling and guiding Gaussian placement during optimization. Experiments on three challenging datasets show consistent improvements in sparse-view NVS, enabling high-quality object-centric reconstruction from as few as four input images. Our method is domain-agnostic, requires only silhouettes, and introduces no learned parameters, making it a simple complement to existing approaches. Applying VisDom on top of GaussianObject further improves performance on Omni3D and MipNeRF360, while matching or surpassing it at 22 $\times$ lower training cost.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces VisDom, a learning-free geometric constraint for sparse novel view synthesis. It augments classical visual-hull carving by defining a visible domain as the subset of 3D space observed by at least K input views, then uses this domain to restrict volumetric sampling in NeRF and guide Gaussian placement in Gaussian Splatting. The approach requires only silhouettes, introduces no learned parameters, and is claimed to yield consistent improvements on three datasets for object-centric reconstruction from as few as four views; it is also shown to improve upon GaussianObject on Omni3D and Mip-NeRF 360 at substantially lower training cost.

Significance. If the central empirical claim holds after verification, VisDom supplies a simple, reproducible, parameter-light (only K) geometric prior that can be plugged into existing implicit and explicit pipelines without additional learned components. The learning-free construction and explicit integration into both NeRF and GS pipelines are clear strengths that distinguish it from data-driven regularizers.

major comments (2)
  1. [Visible domain definition] Visible domain definition: the claim that the K-view filter removes only incorrect geometry while preserving all true surface points is load-bearing for the improvement-without-harm assertion, yet the manuscript provides no analysis or counter-example check for the N=4 sparse case where self-occlusion routinely produces surface regions visible in only 1–2 silhouettes; any K>2 therefore risks carving away legitimate geometry that standard visual-hull (K=N) would retain.
  2. [Integration into NeRF and GS pipelines] Integration and optimization sections: the description of how the visible domain restricts NeRF sampling and guides GS placement lacks concrete implementation details (e.g., how the domain is discretized or queried during each optimization step), making it impossible to assess whether the reported gains are attributable to the geometric prior or to incidental changes in sampling density.
minor comments (2)
  1. [Method] The choice of K is listed as the sole free parameter but no guidance or ablation is given on its selection across the three datasets.
  2. [Experiments] Figure captions and table headers should explicitly state the value of K used for each reported result.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and the recommendation for major revision. We address each major comment below with clarifications and commitments to strengthen the manuscript where the points identify gaps in analysis or detail.

read point-by-point responses

  1. Referee: [Visible domain definition] Visible domain definition: the claim that the K-view filter removes only incorrect geometry while preserving all true surface points is load-bearing for the improvement-without-harm assertion, yet the manuscript provides no analysis or counter-example check for the N=4 sparse case where self-occlusion routinely produces surface regions visible in only 1–2 silhouettes; any K>2 therefore risks carving away legitimate geometry that standard visual-hull (K=N) would retain.

    Authors: We acknowledge that the manuscript asserts the K-view filter primarily removes incorrect geometry without harming true surfaces, yet provides no dedicated analysis or counter-examples for the N=4 case under self-occlusion. In practice we select K=2 for four-view experiments to limit this risk, and results show gains without new artifacts, but this does not fully address the theoretical concern. We will add a discussion section with visibility-count analysis on true surfaces (using ground-truth meshes where available) and counter-example checks in the revision. revision: yes

  2. Referee: [Integration into NeRF and GS pipelines] Integration and optimization sections: the description of how the visible domain restricts NeRF sampling and guides GS placement lacks concrete implementation details (e.g., how the domain is discretized or queried during each optimization step), making it impossible to assess whether the reported gains are attributable to the geometric prior or to incidental changes in sampling density.

    Authors: We agree the current description lacks sufficient implementation specifics on discretization and per-step querying. The visible domain is precomputed once as a binary 3D mask from the silhouettes and then applied to filter samples or guide placement, but details such as grid resolution, data structure, and exact lookup method during optimization are not fully specified. In the revised manuscript we will expand these sections with pseudocode, explicit discretization parameters, and querying procedure to enable readers to isolate the contribution of the prior. revision: yes

Circularity Check

0 steps flagged

No significant circularity: VisDom is an explicit geometric construction from input silhouettes

full rationale

The paper defines the visible domain directly as the 3D subset observed by at least K input views (a deterministic carving operation on the given silhouettes) and applies it as a hard filter on sampling/placement. This construction is independent of any optimization parameters or learned quantities inside NeRF/GS; the claimed improvement is an empirical outcome on held-out views rather than a quantity forced by the definition itself. No self-citation chains, fitted-input-as-prediction, or ansatz smuggling appear in the derivation. The method is explicitly learning-free with zero learned parameters, making the central prior self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

One tunable threshold K and the standard domain assumption of accurate input silhouettes; the visible domain itself is an invented filtering entity without external falsifiable evidence supplied in the abstract.

free parameters (1)
  • K
    Minimum number of views required for a point to belong to the visible domain; chosen as a hyperparameter.
axioms (1)
  • domain assumption Input silhouettes accurately delineate object boundaries from each camera viewpoint.
    Required for the visual-hull carving step that VisDom augments.
invented entities (1)
  • Visible domain no independent evidence
    purpose: Additional spatial filter that retains only 3D regions observed by at least K views.
    New geometric construct introduced to strengthen the prior beyond silhouette carving.

pith-pipeline@v0.9.1-grok · 5799 in / 1336 out tokens · 20580 ms · 2026-06-26T17:44:09.100160+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

38 extracted references · 3 canonical work pages

  1. [1]

    CVPR (2022)

    Barron, J.T., Mildenhall, B., Verbin, D., Srinivasan, P.P., Hedman, P.: Mip-nerf 360: Unbounded anti-aliased neural radiance fields. CVPR (2022)

    2022

  2. [2]

    ICCV (2023)

    Barron, J.T., Mildenhall, B., Verbin, D., Srinivasan, P.P., Hedman, P.: Zip-nerf: Anti-aliased grid-based neural radiance fields. ICCV (2023)

    2023

  3. [3]

    Stanford University (1974)

    Baumgart, B.G.: Geometric modeling for computer vision. Stanford University (1974)

    1974

  4. [4]

    In: CVPR

    Brazil, G., Kumar, A., Straub, J., Ravi, N., Johnson, J., Gkioxari, G.: Omni3D: A large benchmark and model for 3D object detection in the wild. In: CVPR. IEEE, Vancouver, Canada (June 2023)

    2023

  5. [5]

    In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    Chibane, J., Bansal, A., Lazova, V., Pons-Moll, G.: Stereo radiance fields (srf): Learning view synthesis from sparse views of novel scenes. In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE (jun 2021)

    2021

  6. [6]

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

    Deng, K., Liu, A., Zhu, J.Y., Ramanan, D.: Depth-supervised nerf: Fewer views and faster training for free. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 12882–12891 (2022)

    2022

  7. [7]

    In: British Machine Vision Conference (BMVC’03)

    Franco, J.S., Boyer, E.: Exact polyhedral visual hulls. In: British Machine Vision Conference (BMVC’03). vol. 1, pp. 329–338 (2003)

    2003

  8. [8]

    ACM Transactions on Graphics (TOG)42(4), 1–12 (2023)

    I¸ sık, M., R¨ unz, M., Georgopoulos, M., Khakhulin, T., Starck, J., Agapito, L., Nießner, M.: Humanrf: High-fidelity neural radiance fields for humans in motion. ACM Transactions on Graphics (TOG)42(4), 1–12 (2023). https://doi.org/10. 1145/3592415, https://doi.org/10.1145/3592415

    work page doi:10.1145/3592415 2023

  9. [9]

    ACM Transactions on Graphics42(4) (July 2023), https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

    Kerbl, B., Kopanas, G., Leimk¨ uhler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics42(4) (July 2023), https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

    2023

  10. [10]

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

    Kim, M., Seo, S., Han, B.: Infonerf: Ray entropy minimization for few-shot neural volume rendering. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 12912–12921 (2022)

    2022

  11. [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

  12. [12]

    arXiv preprint arXiv:2111.13112 (2021)

    Kondo, N., Ikeda, Y., Tagliasacchi, A., Matsuo, Y., Ochiai, Y., Gu, S.S.: Vaxnerf: Revisiting the classic for voxel-accelerated neural radiance field. arXiv preprint arXiv:2111.13112 (2021)

    arXiv 2021

  13. [13]

    International journal of computer vision38, 199–218 (2000)

    Kutulakos, K.N., Seitz, S.M.: A theory of shape by space carving. International journal of computer vision38, 199–218 (2000)

    2000

  14. [14]

    IEEE Transactions on pattern analysis and machine intelligence16(2), 150–162 (1994)

    Laurentini, A.: The visual hull concept for silhouette-based image understanding. IEEE Transactions on pattern analysis and machine intelligence16(2), 150–162 (1994)

    1994

  15. [15]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Li, J., Zhang, J., Bai, X., Zheng, J., Ning, X., Zhou, J., Gu, L.: Dngaussian: Opti- mizing sparse-view 3d gaussian radiance fields with global-local depth normaliza- tion. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 20775–20785 (2024)

    2024

  16. [16]

    In: Proceedings of the 14th Annual Conference on Computer Graphics and Interactive Techniques

    Lorensen, W.E., Cline, H.E.: Marching cubes: A high resolution 3d surface con- struction algorithm. In: Proceedings of the 14th Annual Conference on Computer Graphics and Interactive Techniques. p. 163–169. SIGGRAPH ’87, Association for Computing Machinery, New York, NY, USA (1987). https://doi.org/10.1145/ 37401.37422, https://doi.org/10.1145/37401.37422...

    work page doi:10.1145/37401.37422 1987

  17. [17]

    In: European Conference on Computer Vision (ECCV)

    Mihajlovic, M., Prokudin, S., Tang, S., Maier, R., Bogo, F., Tung, T., Boyer, E.: Splatfields: Neural gaussian splats for sparse 3d and 4d reconstruction. In: European Conference on Computer Vision (ECCV). Springer (2024)

    2024

  18. [18]

    In: ECCV (2020)

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. In: ECCV (2020)

    2020

  19. [19]

    ACM Trans

    M¨ uller, T., Evans, A., Schied, C., Keller, A.: Instant neural graphics primi- tives with a multiresolution hash encoding. ACM Trans. Graph.41(4), 102:1– 102:15 (Jul 2022). https://doi.org/10.1145/3528223.3530127, https://doi.org/10. 1145/3528223.3530127

    work page doi:10.1145/3528223.3530127 2022

  20. [20]

    In: Proc

    Niemeyer, M., Barron, J.T., Mildenhall, B., Sajjadi, M.S.M., Geiger, A., Radwan, N.: Regnerf: Regularizing neural radiance fields for view synthesis from sparse in- puts. In: Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR) (2022)

    2022

  21. [21]

    In: Pro- ceedings IEEE Conf

    Niemeyer, M., Mescheder, L., Oechsle, M., Geiger, A.: Differentiable volumetric rendering: Learning implicit 3d representations without 3d supervision. In: Pro- ceedings IEEE Conf. on Computer Vision and Pattern Recognition (CVPR) (2020)

    2020

  22. [22]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Schonberger, J.L., Frahm, J.M.: Structure-from-motion revisited. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 4104–4113 (2016)

    2016

  23. [23]

    Shi, R., Wei, X., Wang, C., Su, H.: Zerorf: Fast sparse view 360deg reconstruction with zero pretraining (2023)

    2023

  24. [24]

    In: Tenth IEEE Inter- national Conference on Computer Vision (ICCV’05) Volume 1

    Sinha, S.N., Pollefeys, M.: Multi-view reconstruction using photo-consistency and exact silhouette constraints: A maximum-flow formulation. In: Tenth IEEE Inter- national Conference on Computer Vision (ICCV’05) Volume 1. vol. 1, pp. 349–356. IEEE (2005)

    2005

  25. [25]

    In: Eurographics Symposium on Rendering (2021)

    Sun, T., Lin, K.E., Bi, S., Xu, Z., Ramamoorthi, R.: Nelf: Neural light-transport field for portrait view synthesis and relighting. In: Eurographics Symposium on Rendering (2021)

    2021

  26. [26]

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

    Wang, G., Chen, Z., Loy, C.C., Liu, Z.: Sparsenerf: Distilling depth ranking for few-shot novel view synthesis. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 9065–9076 (2023)

    2023

  27. [27]

    In: Pro- ceedings of the Computer Vision and Pattern Recognition Conference

    Wu, J., Li, R., Zhu, Y., Guo, R., Sun, J., Zhang, Y.: Sparse2dgs: Geometry- prioritized gaussian splatting for surface reconstruction from sparse views. In: Pro- ceedings of the Computer Vision and Pattern Recognition Conference. pp. 11307– 11316 (2025)

    2025

  28. [28]

    In: CVPR (2023)

    Wynn, J., Turmukhambetov, D.: DiffusioNeRF: Regularizing Neural Radiance Fields with Denoising Diffusion Models. In: CVPR (2023)

    2023

  29. [29]

    arXiv preprint arXiv:2402.10259 (2024)

    Yang, C., Li, S., Fang, J., Liang, R., Xie, L., Zhang, X., Shen, W., Tian, Q.: Gaus- sianobject: Just taking four images to get a high-quality 3d object with gaussian splatting. arXiv preprint arXiv:2402.10259 (2024)

    arXiv 2024

  30. [30]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Yang, J., Pavone, M., Wang, Y.: Freenerf: Improving few-shot neural rendering with free frequency regularization. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 8254–8263 (2023)

    2023

  31. [31]

    Advances in Neural Information Processing Systems33, 2492–2502 (2020)

    Yariv, L., Kasten, Y., Moran, D., Galun, M., Atzmon, M., Ronen, B., Lipman, Y.: Multiview neural surface reconstruction by disentangling geometry and appear- ance. Advances in Neural Information Processing Systems33, 2492–2502 (2020)

    2020

  32. [32]

    In: CVPR (2021) VisDom: Sparse Novel View Synthesis with Visible Domain Constraint 15

    Yu, A., Ye, V., Tancik, M., Kanazawa, A.: pixelNeRF: Neural radiance fields from one or few images. In: CVPR (2021) VisDom: Sparse Novel View Synthesis with Visible Domain Constraint 15

    2021

  33. [33]

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

    Yu, Z., Chen, A., Huang, B., Sattler, T., Geiger, A.: Mip-splatting: Alias-free 3d gaussian splatting. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 19447–19456 (2024)

    2024

  34. [34]

    Advances in neural information processing systems35, 25018–25032 (2022)

    Yu, Z., Peng, S., Niemeyer, M., Sattler, T., Geiger, A.: Monosdf: Exploring monoc- ular geometric cues for neural implicit surface reconstruction. Advances in neural information processing systems35, 25018–25032 (2022)

    2022

  35. [35]

    arXiv preprint arXiv:2405.12110 (2024)

    Zhang, J., Li, J., Yu, X., Huang, L., Gu, L., Zheng, J., Bai, X.: Cor-gs: Sparse- view 3d gaussian splatting via co-regularization. arXiv preprint arXiv:2405.12110 (2024)

    arXiv 2024

  36. [36]

    Zhang, L., Rao, A., Agrawala, M.: Adding conditional control to text-to-image diffusion models (2023)

    2023

  37. [37]

    In: European conference on computer vision

    Zhu, Z., Fan, Z., Jiang, Y., Wang, Z.: Fsgs: Real-time few-shot view synthesis using gaussian splatting. In: European conference on computer vision. pp. 145–

  38. [38]

    Gladkova et al

    Springer (2024) 16 M. Gladkova et al. Supplementary Material — VisDom: Sparse Novel View Synthesis with Visible Domain Constraint In this supplementary, we first provide background on NeRF and 3DGS in sec- tion A and dataset details in section B.In section C we show the impact of our constraint on reconstructing the visual hull and study the performance o...

    2024