pith. sign in

arxiv: 2607.01578 · v1 · pith:HEHCFQ3Cnew · submitted 2026-07-02 · 💻 cs.CV

MVFusion-GS: Motion-Variance Guided Temporal Attention for High-Quality Dynamic Gaussian Splatting

Jianwei Hu , Tingxuan Huang , Hengyu Zhou , Ningna Wang , Xiaohu Guo Jinshan Lai , Bin Wang This is my paper

Pith reviewed 2026-07-03 17:09 UTC · model grok-4.3

classification 💻 cs.CV
keywords dynamic gaussian splattingmotion variancetemporal attentiondeformation fieldsnovel view synthesis3D scene reconstructiondistractor removal
0
0 comments X

The pith

Adding explicit motion awareness to deformation networks improves accuracy in dynamic 3D Gaussian Splatting for both moving objects and static backgrounds.

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

The paper argues that current deformation networks for dynamic 3D Gaussian Splatting miss explicit signals about motion intensity over time and short-term coherence between frames. This causes poor modeling of moving foregrounds and leftover static artifacts in backgrounds. MVFusion-GS introduces motion-variance guidance to separate dynamic from static elements and a temporal attention module to enforce consistency across nearby timesteps. If correct, this would allow higher quality novel view synthesis in videos with moving objects. Experiments on standard benchmarks show these additions deliver better results than prior methods.

Core claim

MVFusion-GS enhances deformation networks by aggregating per-Gaussian deformation statistics across time to compute motion variance for guiding dynamic-static separation, and by applying Transformer self-attention over neighboring timesteps to capture local motion dependencies, resulting in more accurate foreground deformation and reduced pseudo-static residuals in the background.

What carries the argument

Motion-Variance Guided Refinement, which uses time-aggregated deformation statistics to estimate motion variance and direct separation, and MotionFormer Temporal Attention, which models dependencies via self-attention across timesteps.

If this is right

  • More accurate deformation for moving foreground objects.
  • Cleaner reconstruction of static background regions without pseudo-residuals.
  • Improved temporal consistency in rendered dynamic scenes.
  • State-of-the-art results on dynamic scene reconstruction benchmarks.
  • Effective distractor-free reconstruction by better handling motion.

Where Pith is reading between the lines

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

  • Similar motion variance signals could be tested in other neural rendering frameworks beyond Gaussian Splatting.
  • The approach might extend to longer-range temporal modeling if attention is scaled up.
  • Real-time performance could be evaluated by measuring the added computational cost of the attention module.
  • Application to scenes with multiple independent moving objects would test the robustness of the variance estimation.

Load-bearing premise

The assumption that per-Gaussian deformation statistics aggregated over time provide a reliable indicator of motion variance that correctly separates dynamic and static content.

What would settle it

A controlled experiment where the motion variance signal is replaced with random values or noise and performance does not degrade would falsify the claim that the guidance is effective.

Figures

Figures reproduced from arXiv: 2607.01578 by Bin Wang, Hengyu Zhou, Jianwei Hu, Ningna Wang, Tingxuan Huang, Xiaohu Guo Jinshan Lai.

Figure 1
Figure 1. Figure 1: Standard dynamic-static Gaussian decomposition leaves pseudo-static fore￾ground residuals in the static branch (top), causing background contamination. Our method enhances the motion awareness of the deformation network, correctly re￾attributing pseudo-static Gaussians to the dynamic branch (bottom). This yields dual benefits: cleaner backgrounds for distractor-free reconstruction and finer foreground deta… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed motion-aware refinement framework. The base￾line deformation network first predicts a coarse deformation for dynamic Gaussians. Two motion-aware modules are then applied: MVG extracts global trajectory signa￾tures and local variance cues, while MFTA aggregates short-term temporal context. The fused motion-aware feature produces refined Gaussian updates, followed by fore￾ground–back… view at source ↗
Figure 3
Figure 3. Figure 3: Motion descriptor extraction in MVG. Given sampled timestamps {tk} K k=1, we evaluate each Gaussian’s deformation trajectory and compute cross-frame variations in position, rotation, and scale. (A) The full trajectory is aggregated into a 13D global trajectory signature that captures long-term motion statistics. (B) Local motion vari￾ance is cached as a time-indexed dictionary: given a query timestamp t, t… view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of per-Gaussian motion variance heatmaps for individual at￾tributes and the fused motion intensity score. tk tk-w tk tk+w zk-w zk zk+w e (tk) v · Δpost Δrt Δst ΔSHm t e(t) Φ hbase(tk) ΔG d MR hMVG(tk) Linear Projection (Q, K, V) Add & Norm Feed-Forward Add & Norm Head (Residual) Multi-Head Self-Attention · transformer hMFTA(tk) hbase(tk) + (C) MV-Fusion (A) MVG Injection (B) MFTA Injection Po… view at source ↗
Figure 5
Figure 5. Figure 5: Feature-space motion injection and fusion. (A) MVG encodes the cached global signature vi and local intensity ei(t) into a residual feature hMVG. (B) MFTA forms temporal tokens zi,t over a short window and aggregates them via cross-attention to obtain hMFTA. (C) The fused feature is decoded by the unchanged deformation heads Φ to predict refined deformation ∆G MR d . where ¯· and σ· denote temporal mean an… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison of static background reconstruction on held-out test views from the NeRF On-the-go benchmark. Red boxes highlight regions where De￾Gauss retains visible pseudo-static foreground residuals (e.g., ghostly pedestrian silhou￾ettes in Patio-High), while our method produces substantially cleaner backgrounds. the improved dynamic-static separation preserves overall rendering quality, while … view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison of background-only and full composed renderings on NeRF On-the-go benchmark. Red boxes reveal that DeGauss backgrounds retain ghostly foreground residuals, while our method achieves cleaner backgrounds and sharper full-scene compositions. GT（with Distractor） DeGauss Background Our Background [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparison on RobustNeRF. MVFusion-GS produces cleaner back￾grounds than DeGauss [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative comparison on Neu3D. MVFusion-GS reconstructs sharper motion details with fewer artifacts [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Visualization of motion variance maps under different fusion weight set￾tings. From left to right: position-only (1, 0, 0), uniform weighting (1/3, 1/3, 1/3), posi￾tion+rotation (0.8, 0.2, 0), position+scale (0.8, 0, 0.2), and the proposed (0.7, 0.2, 0.1). Highlighted regions indicate areas with noticeable motion differences. B.2 Sensitivity to Motion Statistics Update Interval The motion statistics are p… view at source ↗
Figure 12
Figure 12. Figure 12: A promising direction for future work is to leverage these variance cues [PITH_FULL_IMAGE:figures/full_fig_p024_12.png] view at source ↗
Figure 11
Figure 11. Figure 11: Branch-level decomposition visualization. [PITH_FULL_IMAGE:figures/full_fig_p025_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Failure case. In challenging cases where foreground motion is complex, the estimated deformation variance may become less discriminative, especially when the baseline deformation field underfits foreground motion. MVFusion-GS can substantially reduce pseudo-static foreground residuals, but it may not completely remove artifacts when the underlying deformation model fails to provide reliable motion cues [… view at source ↗
read the original abstract

3D Gaussian Splatting (3DGS) enables real-time novel view synthesis for static scenes. Extending it to dynamic scenes via deformation fields has recently attracted significant attention, particularly for dynamic scene reconstructionband distractor-free. However, existing deformation networks lack explicit motion awareness: they neither capture long-term motion intensity nor exploit short-term temporal coherence, leading to inaccurate foreground deformation and pseudo-static residuals in the background. We present MVFusion-GS, a method that enhances deformation networks with two complementary motion-aware mechanisms. The Motion-Variance Guided Refinement aggregates per-Gaussian deformation statistics across time to estimate motion variance and uses it to guide dynamic-static separation during deformation prediction. The MotionFormer Temporal Attention module applies Transformer self-attention over neighboring timesteps to model local motion dependencies and improve temporal consistency. Extensive experiments on both dynamic scene reconstruction and distractor-free reconstruction benchmarks demonstrate state-of-the-art performance, showing that explicit motion awareness improves both foreground motion modeling and static background reconstruction.

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 paper proposes MVFusion-GS to extend 3D Gaussian Splatting to dynamic scenes by augmenting deformation networks with two motion-aware components: Motion-Variance Guided Refinement, which aggregates per-Gaussian deformation statistics over time to estimate motion variance and guide dynamic-static separation, and MotionFormer Temporal Attention, which applies Transformer self-attention across neighboring timesteps to capture local motion dependencies. The central claim is that these explicit mechanisms yield more accurate foreground deformation and cleaner static backgrounds, supported by state-of-the-art results on dynamic scene reconstruction and distractor-free benchmarks.

Significance. If the motion-variance signal proves robust and the temporal attention demonstrably reduces artifacts without introducing new ones, the approach could meaningfully advance dynamic novel-view synthesis by providing a principled way to separate motion intensity from deformation fitting. The explicit separation of long-term variance and short-term coherence is a clear conceptual contribution over prior deformation-field methods.

major comments (2)
  1. [Method (Motion-Variance Guided Refinement)] Method section (Motion-Variance Guided Refinement): the description of aggregating deformation statistics to produce the variance signal does not specify whether the statistics are taken from a fixed initial deformation field, an iterative bootstrap, or the jointly optimized network; without this, the separation signal and the deformation parameters can co-adapt, undermining the claim that explicit motion awareness improves foreground/background modeling.
  2. [Experiments] Experiments (dynamic scene reconstruction benchmarks): the reported gains are presented without ablation isolating the contribution of the variance-guided separation versus the temporal attention module, and without error bars or multiple random seeds, making it impossible to assess whether the SOTA claim is load-bearing or sensitive to initialization.
minor comments (2)
  1. [Abstract / §2] The abstract and introduction use the phrase 'distractor-free reconstruction' without a precise definition or citation to the benchmark protocol; this should be clarified in §2 or the experimental setup.
  2. [Method] Notation for the per-Gaussian deformation statistics (e.g., how variance is normalized across time) is introduced without an equation; adding a compact definition would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments highlight important points on methodological clarity and experimental rigor that we will address in the revision.

read point-by-point responses

  1. Referee: [Method (Motion-Variance Guided Refinement)] Method section (Motion-Variance Guided Refinement): the description of aggregating deformation statistics to produce the variance signal does not specify whether the statistics are taken from a fixed initial deformation field, an iterative bootstrap, or the jointly optimized network; without this, the separation signal and the deformation parameters can co-adapt, undermining the claim that explicit motion awareness improves foreground/background modeling.

    Authors: We agree that the current description leaves the source of the deformation statistics unspecified, which is a valid concern regarding potential co-adaptation. In the revised manuscript we will update Section 3.2 to explicitly state that the per-Gaussian deformation statistics are collected from a fixed initial deformation field obtained via a short bootstrap optimization phase; this field is then frozen when computing the motion-variance signal that guides the subsequent joint optimization. This change directly addresses the co-adaptation issue while preserving the overall architecture. revision: yes

  2. Referee: [Experiments] Experiments (dynamic scene reconstruction benchmarks): the reported gains are presented without ablation isolating the contribution of the variance-guided separation versus the temporal attention module, and without error bars or multiple random seeds, making it impossible to assess whether the SOTA claim is load-bearing or sensitive to initialization.

    Authors: We acknowledge that the experiments do not isolate the individual contributions of Motion-Variance Guided Refinement and MotionFormer Temporal Attention, nor do they report error bars or multi-seed statistics. In the revised version we will add a dedicated ablation table that evaluates each module independently on the dynamic reconstruction and distractor-free benchmarks. We will also rerun the primary quantitative results across three random seeds and report mean and standard deviation to substantiate the robustness of the SOTA claims. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper introduces explicit motion-aware mechanisms (Motion-Variance Guided Refinement and MotionFormer Temporal Attention) as architectural additions to deformation networks, with performance claims supported by benchmark experiments rather than any reduction of outputs to inputs by definition, fitted parameters renamed as predictions, or load-bearing self-citations. No equations or sections in the provided text exhibit self-definitional loops, ansatzes smuggled via prior work, or renaming of known results. The derivation remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only; no free parameters, axioms, or invented entities are described.

pith-pipeline@v0.9.1-grok · 5718 in / 1081 out tokens · 23072 ms · 2026-07-03T17:09:05.266231+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

35 extracted references · 4 canonical work pages

  1. [1]

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

    Attal, B., Huang, J.B., Richardt, C., Zollhoefer, M., Kopf, J., O’Toole, M., Kim, C.: Hyperreel: High-fidelity 6-dof video with ray-conditioned sampling. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 16610–16620 (2023)

    2023

  2. [2]

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

    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 (CVPR). pp. 130–141 (June 2023)

    2023

  3. [3]

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

    Chen, Z., Funkhouser, T., Hedman, P., Tagliasacchi, A.: MobileNeRF: Exploiting the polygon rasterization pipeline for efficient neural field rendering on mobile architectures. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 16569–16578 (2023)

    2023

  4. [4]

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

    Fridovich-Keil, S., Meanti, G., Warburg, F.R., Recht, B., Kanazawa, A.: K-planes: Explicit radiance fields in space, time, and appearance. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 12479– 12488 (2023)

    2023

  5. [5]

    In: ACM SIGGRAPH 2024 Conference Papers

    Huang,B.,Yu,Z.,Chen,A.,Geiger,A.,Gao,S.:2dgaussiansplattingforgeometri- cally accurate radiance fields. In: ACM SIGGRAPH 2024 Conference Papers. SIG- GRAPH ’24, Association for Computing Machinery, New York, NY, USA (2024). https://doi.org/10.1145/3641519.3657428

    work page doi:10.1145/3641519.3657428 2024

  6. [6]

    In: The Fourteenth International Conference on Learning Representations

    Huang, T., Zhu, H., Yong, J.H., Pan, H., Wang, B.: Mango-GS: Enhancing spatio- temporal consistency in dynamic scenes reconstruction using multi-frame node- guided 4d gaussian splatting. In: The Fourteenth International Conference on Learning Representations

  7. [7]

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

    Huang, Y.H., Sun, Y.T., Yang, Z., Lyu, X., Cao, Y.P., Qi, X.: SC-GS: Sparse- controlled gaussian splatting for editable dynamic scenes. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 4220–4230 (2024)

    2024

  8. [8]

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

    Jiang, D., Hou, Z., Ke, Z., Yang, X., Zhou, X., Qiu, T.: Timeformer: Capturing temporal relationships of deformable 3d gaussians for robust reconstruction. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 8721–8732 (2025)

    2025

  9. [9]

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

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

    2023

  10. [10]

    In: Proceedings of the 38th International Conference on Neural Information Processing Systems (2024)

    Kulhanek, J., Peng, S., Kukelova, Z., Pollefeys, M., Sattler, T.: WildGaussians: 3D gaussian splatting in the wild. In: Proceedings of the 38th International Conference on Neural Information Processing Systems (2024)

    2024

  11. [11]

    In: ACM SIGGRAPH2024ConferencePapers(2024).https://doi.org/10.1145/3641519

    Li, D., Huang, S.S., Lu, Z., Duan, X., Huang, H.: ST-4DGS: Spatial-temporally consistent 4d gaussian splatting for efficient dynamic scene rendering. In: ACM SIGGRAPH2024ConferencePapers(2024).https://doi.org/10.1145/3641519. 3657520

    work page doi:10.1145/3641519 2024

  12. [12]

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

    Li, T., Slavcheva, M., Zollhöfer, M., Green, S., Lassner, C., Kim, C., Schmidt, T., Lovegrove, S., Goesele, M., Newcombe, R., Lv, Z.: Neural 3d video synthesis from multi-view video. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 5521–5531 (June 2022)

    2022

  13. [13]

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

    Li, Z., Chen, Z., Li, Z., Xu, Y.: Spacetime gaussian feature splatting for real- time dynamic view synthesis. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 8508–8520 (June 2024) MVFusion-GS for High-Quality Dynamic Gaussian Splatting 17

    2024

  14. [14]

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

    Lu, T., Yu, M., Xu, L., Xiangli, Y., Wang, L., Lin, D., Dai, B.: Scaffold-gs: Struc- tured 3d gaussians for view-adaptive rendering. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 20654–20664 (2024)

    2024

  15. [15]

    In: CVPR (2021)

    Martin-Brualla, R., Radwan, N., Sajjadi, M.S.M., Barron, J.T., Dosovitskiy, A., Duckworth, D.: NeRF in the Wild: Neural Radiance Fields for Unconstrained Photo Collections. In: CVPR (2021)

    2021

  16. [16]

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: representing scenes as neural radiance fields for view synthesis. Commun. ACM65(1), 99–106 (Dec 2021).https://doi.org/10.1145/3503250

    work page doi:10.1145/3503250 2021

  17. [17]

    ACM Transactions on Graphics41(4), 102:1– 102:15 (2022).https://doi.org/10.1145/3528223.3530127

    Müller,T.,Evans,A.,Schied,C.,Keller,A.:Instantneuralgraphicsprimitiveswith a multiresolution hash encoding. ACM Transactions on Graphics41(4), 102:1– 102:15 (2022).https://doi.org/10.1145/3528223.3530127

    work page doi:10.1145/3528223.3530127 2022

  18. [18]

    In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Park, J.J., Florence, P., Straub, J., Newcombe, R., Lovegrove, S.: DeepSDF: Learn- ing continuous signed distance functions for shape representation. In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 165–174 (2019)

    2019

  19. [19]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)

    Park, J., Bui, M.Q.V., Bello, J.L.G., Moon, J., Oh, J., Kim, M.: SplineGS: Ro- bust motion-adaptive spline for real-time dynamic 3d gaussians from monocular video. In: Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR). pp. 26866–26875 (June 2025)

    2025

  20. [20]

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

    Ren, W., Zhu, Z., Sun, B., Chen, J., Pollefeys, M., Peng, S.: NeRF On-the-go: Exploiting uncertainty for distractor-free nerfs in the wild. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 8931–8940 (June 2024)

    2024

  21. [21]

    ACM Trans

    Sabour, S., Goli, L., Kopanas, G., Matthews, M., Lagun, D., Guibas, L., Jacobson, A., Fleet, D., Tagliasacchi, A.: Spotlesssplats: Ignoring distractors in 3d gaussian splatting. ACM Trans. Graph.44(2) (Apr 2025).https://doi.org/10.1145/ 3727143

    2025

  22. [22]

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

    Sabour, S., Vora, S., Duckworth, D., Krasin, I., Fleet, D.J., Tagliasacchi, A.: RobustNeRF: Ignoring distractors with robust losses. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 20626–20636 (June 2023)

    2023

  23. [23]

    In: European Conference on Computer Vision

    Shaw, R., Nazarczuk, M., Song, J., Moreau, A., Catley-Chandar, S., Dhamo, H., Pérez-Pellitero, E.: Swings: sliding windows for dynamic 3d gaussian splatting. In: European Conference on Computer Vision. pp. 37–54. Springer (2024)

    2024

  24. [24]

    IEEE Transactions on Visualization and Computer Graphics29(5), 2732–2742 (2023)

    Song, L., Chen, A., Li, Z., Chen, Z., Chen, L., Yuan, J., Xu, Y., Geiger, A.: Nerf- player: A streamable dynamic scene representation with decomposed neural radi- ance fields. IEEE Transactions on Visualization and Computer Graphics29(5), 2732–2742 (2023)

    2023

  25. [25]

    In: SIGGRAPH Asia 2024 Conference Papers

    Stearns, C., Harley, A.W., Uy, M., Dubost, F., Tombari, F., Wetzstein, G., Guibas, L.: Dynamic gaussian marbles for novel view synthesis of casual monocular videos. In: SIGGRAPH Asia 2024 Conference Papers. pp. 1–11 (2024)

    2024

  26. [26]

    Advances in neural information pro- cessing systems30(2017)

    Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,A.N.,Kaiser, Ł., Polosukhin, I.: Attention is all you need. Advances in neural information pro- cessing systems30(2017)

    2017

  27. [27]

    19706–19716 (2023)

    Wang, F., Tan, S., Li, X., Tian, Z., Song, Y., Liu, H.: Mixed neural voxels for fast multi-view video synthesis pp. 19706–19716 (2023)

    2023

  28. [28]

    In: International Conference on Computer Vision (ICCV) (2025) 18 Jianwei Hu*, Tingxuan Huang*, et al

    Wang, Q., Ye, V., Gao, H., Zeng, W., Austin, J., Li, Z., Kanazawa, A.: Shape of motion: 4d reconstruction from a single video. In: International Conference on Computer Vision (ICCV) (2025) 18 Jianwei Hu*, Tingxuan Huang*, et al

    2025

  29. [29]

    In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    Wang, R., Lohmeyer, Q., Meboldt, M., Tang, S.: DeGauss: Dynamic-static decom- position with gaussian splatting for distractor-free 3d reconstruction. In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 6294–6303 (October 2025)

    2025

  30. [30]

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

    Wang, Y., Yang, P., Xu, Z., Sun, J., Zhang, Z., Chen, Y., Bao, H., Peng, S., Zhou, X.: FreeTimeGS: Free gaussian primitives at anytime anywhere for dynamic scene reconstruction. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 21750–21760 (2025)

    2025

  31. [31]

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

    Wang, Y., Klasson, M., Turkulainen, M., Wang, S., Kannala, J., Solin, A.: DeSplat: Decomposed gaussian splatting for distractor-free rendering. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 722–732 (June 2025)

    2025

  32. [32]

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

    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 (CVPR). pp. 20310–20320 (June 2024)

    2024

  33. [33]

    In: Proceedings of the 36th International Conference on Neural Information Processing Systems

    Wu, T., Zhong, F., Tagliasacchi, A., Cole, F., Oztireli, C.: D2NeRF: self-supervised decoupling of dynamic and static objects from a monocular video. In: Proceedings of the 36th International Conference on Neural Information Processing Systems. NIPS ’22, Curran Associates Inc., Red Hook, NY, USA (2022)

    2022

  34. [34]

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

    Yang, Z., Gao, X., Zhou, W., Jiao, S., Zhang, Y., Jin, X.: Deformable 3d gaussians for high-fidelity monocular dynamic scene reconstruction. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 20331–20341 (2024)

    2024

  35. [35]

    Yariv, L., Gu, J., Kasten, Y., Lipman, Y.: Volume rendering of neural implicit surfaces. In: Advances in Neural Information Processing Systems (NeurIPS) (2021) MVFusion-GS for High-Quality Dynamic Gaussian Splatting 19 Supplementary Material In this supplementary material, we provide additional implementation details, experiments and analyses to complemen...

    2021