Pith. sign in

REVIEW 4 major objections 3 minor 3 cited by

LocalDyGS: Multi-view Global Dynamic Scene Modeling via Adaptive Local Implicit Feature Decoupling

T0 review · 4 major / 3 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 LocalDyGS argues that a dynamic scene can be reconstructed globally by decomposing it into seed-anchored local spaces, each generating time-varying Temporal Gaussians from decoupled static and dynamic features, allowing one method to…

desk verdict A plausible architecture with real ablations, but the 'first large-scale' claim is undercut by the omitted Swift4D comparison and thin VRU evidence. read the letter →

arxiv 2507.02363 v1 pith:ICXRWP7M submitted 2025-07-03 cs.CV

Jiahao Wu , Rui Peng , Jianbo Jiao , Jiayu Yang , Luyang Tang , Kaiqiang Xiong , Jie Liang , Jinbo Yan
show 2 more authors
Runling Liu Ronggang Wang
This is my paper
classification cs.CV
keywords dynamicscenereconstruction3DGaussiansplattinglocalspacedecompositionstatic-dynamicfeaturedecouplingTemporalGaussiansadaptiveseedgrowingmulti-viewvideosynthesisnovelview
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

LocalDyGS sets out to show that a globally complex dynamic scene does not need to be tracked through time as a whole. Instead, the paper decomposes space into small local regions anchored by seed points, and within each region generates time-varying Temporal Gaussians from a feature that is fixed across time plus a sparse dynamic residual that changes with time. The claim is that this local, decoupled design handles both small hand motions and large basketball-court movements with a single framework, at real-time rendering speed and with a compact model. If this holds, free-viewpoint video of sports and other highly dynamic events could be reconstructed from synchronized multi-view footage without the blur, flicker, and storage overhead of existing dynamic Gaussian methods.

What carries the argument

The central object is the seed-anchored local space with its accompanying Temporal Gaussian. A seed is a 3D point, initialized by fusing Structure-from-Motion point clouds from several video frames, and it carries a learnable static feature and a scale that controls how far its influence reaches. For every query time, a shared dynamic residual field built from multi-resolution hash encoding over space and time and a shallow MLP produces a time-specific residual feature; a weight field predicts how much static versus dynamic content to use; and a decoder MLP outputs the parameters of k Temporal Gaussians, including their positions relative to the seed. Two mechanisms keep this efficient: Temporal Gaussians with opacity below a threshold are deactivated for a given time, and an adaptive seed-growing process adds new seeds wherever the maximum 2D projection gradient of a Temporal Gaussian exceeds a threshold. The load-bearing idea is that local, independently parameterized motion is enough to explain global dynamics, so no persistent per-Gaussian trajectory or canonical deformation field is needed.

What would settle it

Train LocalDyGS on a multi-view dynamic sequence where a fast-moving object passes through a textureless region completely absent from the Structure-from-Motion point cloud, and inspect whether any new seeds are grown inside that region and whether rendering there shows blur or ghosting; if no seeds appear and quality collapses, the coverage assumption is necessary for the claimed large-scale performance.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that long-range motion can be represented by short-range local responses: a moving object is never followed as a persistent point; it simply activates different seed-centered local spaces at different times. Each local space stores a time-invariant static feature, and a shared dynamic residual field supplies a small time-specific correction; a weight field blends the two, and a decoder turns the blend into a set of Temporal Gaussians, i.e., 3D Gaussians whose mean, rotation, scale, opacity, and color all depend on the query time. An adaptive seed-growing step monitors the 2D projection gradient of these Gaussians and inserts new seeds wherever motion is under-covered, which the paper argues makes the model robust to incomplete Structure-from-Motion initialization. The paper reports state-of-the-art or competitive PSNR and LPIPS on the N3DV and MeetRoom datasets, competitive results on a basketball-court dataset with large-scale motion, and higher speed and lower storage than prior dynamic Gaussian methods.

Load-bearing premise

The method assumes the fused Structure-from-Motion point cloud from a few frames already places seeds in every region where moving objects will appear, and that the adaptive seed-growing strategy can fill any remaining gaps; if SfM misses a region entirely, rendering quality in that region will degrade.

Editorial extensions

If this is right

  • A moving object no longer needs to be followed as a persistent point across time; different local spaces activate as it moves, so long-range motion can be decomposed into short-range pieces.
  • Because the dynamic residual field carries only time-specific corrections, the model stays compact (about 100 MB for 300 N3DV frames) and trains in under an hour on a single GPU.
  • Temporal Gaussian deactivation and adaptive seed growing trade a small amount of storage for large speedups and robustness to incomplete Structure-from-Motion point clouds.
  • The same model achieves state-of-the-art or competitive quality on fine-scale datasets and on a basketball-court dataset with large-scale motion, supporting the claim that one framework spans motion scales.
  • The static/dynamic decoupling and local-space structure may also make the representation easier to edit, stream, or extend to longer sequences, since each local space has independent parameters.

Reading between the lines

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

  • An implicit extension not tested in the paper: because motion is local and seeds are independent, the representation may be amenable to incremental or streaming updates, where only newly activated local spaces need to be refined as a long video progresses.
  • The static/dynamic split is a natural handle for editing: freezing the dynamic residual field and modifying static features could let a user change a background or object while keeping motion intact, though the paper does not demonstrate editing.
  • A testable consequence of the coverage assumption is that the required seed density should scale with the spatial extent and speed of motion; measuring PSNR against deliberately removed seed regions would directly quantify this dependence.
  • As the paper itself notes, a pretrained geometric prior could replace or augment the multi-view Structure-from-Motion point cloud, potentially extending the method to monocular input.
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

4 major / 3 minor

Summary. The paper proposes LocalDyGS, a multi-view dynamic scene reconstruction method based on 3D Gaussian Splatting. Seeds are initialized from SfM point clouds fused over N frames, each seed defines a local space with a static feature shared across time, and a global dynamic residual field supplies time-specific features. These are combined and decoded into Temporal Gaussians, with a deactivation mechanism and an adaptive seed growing strategy. The method is evaluated on N3DV, MeetRoom, and the VRU basketball court dataset, reporting competitive PSNR/LPIPS, compact model size, and fast training and rendering.

Significance. If the reported results are reliable, the local-space decomposition is a promising approach to reducing the complexity of dynamic scene modeling, and the efficiency numbers (0.58h training, 100MB model, 105 FPS on N3DV) are attractive. The ablations support the individual design choices: static/dynamic decoupling, Temporal Gaussian deactivation, and adaptive seed growing all improve quality or speed. However, the central claims of being the 'first' to model large-scale dynamic scenes and of superior performance on the VRU dataset are not supported by the presented evidence, largely because the closest baseline, Swift4D, is omitted from the comparison and the large-scale experiments lack the analyses needed to substantiate the seed-coverage mechanism.

major comments (4)
  1. [Abstract; Section 1; Section 4.2; Table 3] The abstract and contribution list claim that LocalDyGS is 'the first attempt to model larger and more complex highly dynamic scenes.' This is contradicted by the paper's own related work: Swift4D [49] is described as validating on the same VRU basketball court dataset, and Section 4.2 states that this dataset 'is used for the first time in Swift4D.' Swift4D is not included in Table 3, so the firstness claim is unsupported. The authors should remove or substantially qualify this claim and should include Swift4D as a baseline or justify its absence.
  2. [Table 3; Section 4.3] The VRU quantitative results do not show that LocalDyGS outperforms static single-frame methods on all metrics. In Table 3, 2DGS achieves higher PSNR (30.78 vs 30.58), and GOF has higher SSIM (0.949 vs 0.944) and lower LPIPS (0.141 vs 0.173). In addition, the caption says static methods are 'tested on frame 0' while dynamic methods are evaluated over the sequence; this protocol needs clarification because a static fit to a single frame is not directly comparable to a dynamic reconstruction over the full sequence. The text in Section 4.3 calling the VRU performance 'robust' overstates what the table shows.
  3. [Section 3.2; Section 3.5; Section 5; Tables 4 and 6] The claim that LocalDyGS adapts to large-scale motion depends on seed coverage of all regions where moving objects appear. The initial seeds come from SfM point clouds fused from N=6 frames, and Section 5 concedes that significant SfM failure can impact rendering quality. Adaptive Seed Growing (Section 3.5) adds a seed only at the 3D position of an existing Temporal Gaussian whose projected gradient exceeds tau_g, so it cannot create coverage in regions that have no Temporal Gaussian at all. The only large-scale dataset, VRU, is not accompanied by seed-coverage statistics, an ASG ablation, or an N-variation study; the ASG ablation in Table 4 is on MeetRoom and the N study in Table 6 is on N3DV, both fine-scale datasets. Additional evidence on VRU is needed to support the large-scale adaptivity claim.
  4. [Tables 1-3] No error bars, variance estimates, or multiple-seed statistics are reported for any quantitative comparison. The margins over some baselines on N3DV are small (e.g., PSNR 32.28 vs 32.05), and Table 3 shows mixed results. Without variance information, the state-of-the-art claim is not statistically supported.
minor comments (3)
  1. [Section 3.4, Eq. (9)] The symbol T is used both for the set of query times and for the number of iterations; please use a distinct notation to avoid ambiguity.
  2. [Table 4] The metric names 'SSIM1' and 'SSIM2' in Table 4 are inconsistent with 'DSSIM1' and 'DSSIM2' used in Table 1; please unify the naming and define the data ranges in the caption.
  3. [Figure 1] Panel (c) reports PSNR 34.10 and FPS 105 without specifying the scene or the comparison setting; please refer to the corresponding table entry so that the figure can be interpreted independently.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: LocalDyGS is an empirical reconstruction method whose components (seed initialization, static/dynamic feature decoupling, Temporal Gaussian decoding, ASG) are not constructed from the quantities they are used to predict; the only self-citation is contextual and not load-bearing.

full rationale

LocalDyGS makes no closed-form predictive claim that could collapse into its inputs. Seeds are initialized from SfM point clouds (Sec. 3.2) and used to define local spaces; Temporal Gaussians are decoded from static/dynamic features (Eqs. 6-8) and supervised by the multi-view rendering loss (Eq. 11). The 'large-scale' capability is asserted conditionally on seed coverage ('When seeds cover all regions where a moving object appears...', Sec. 1) and then tested on the VRU dataset (Tab. 3); while the paper does not provide seed-coverage statistics or an N-variation/ASG ablation on VRU, that is an evidence gap about assumption sensitivity, not a definitional reduction. The Adaptive Seed Growing rule (Sec. 3.5) adds seeds where the projection gradient of an existing Temporal Gaussian exceeds tau_g; this can densify but not invent coverage in empty regions, which the paper itself hedges in Sec. 5 ('If SfM fails significantly, it may impact rendering quality'). That is a stated limitation, not a circular step. The one self-citation, Swift4D [49] by overlapping authors, is used only as related work describing the VRU dataset and as one of several inspirations for feature decoupling (Sec. 3.3), alongside independent [48] and [56]; no load-bearing argument reduces to this citation. The Abstract's 'first attempt' claim is inconsistent with the paper's own acknowledgment that Swift4D already validated on the basketball-court dataset, but this is a novelty/soundness overclaim, not circularity. No equation in the paper is defined in terms of the result it is used to establish, and no fitted parameter is relabeled as a prediction. Therefore no significant circularity is present.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The method relies on standard learned 3DGS primitives plus a set of hand-chosen hyperparameters. No physically new entities are introduced.

free parameters (5)
  • Deactivation threshold tau_alpha = 0.01
    Hand-chosen threshold to deactivate low-opacity Temporal Gaussians (Section 3.4).
  • Seed growing threshold tau_g = 0.001
    Hand-chosen gradient threshold for adding seeds (Section 3.5).
  • Volume regularization weight lambda_v = 0.001
    Hand-chosen weight in loss function (Section 3.6).
  • SSIM loss weight lambda_SSIM = 0.2
    Hand-chosen weight in loss function (Section 3.6).
  • Temporal Gaussians per seed k = 10
    Chosen by ablation; k=10 balances quality and speed (Table 7).
assumptions (4)
  • domain assumption SfM point clouds provide sufficient seed coverage for dynamic scenes.
    Seed positions are initialized from fused SfM point clouds (Section 3.2); Section 5 admits failure risk.
  • domain assumption Multi-view synchronized cameras provide dense spatiotemporal supervision.
    Method requires multi-view inputs for training (Section 1, 5).
  • domain assumption Static/dynamic feature decoupling is sufficient to represent scenes.
    The method combines a static feature and a dynamic residual field (Section 3.3); ablation shows removal of static feature hurts quality.
  • standard math 3D Gaussian splatting and hash encoding from prior work are reliable.
    Rendering and encoding follow prior work [20, 30, 31] (Section 3.1, 3.3).

how reviews work

0 comments
Cite this review

Pith. "Pith review of LocalDyGS: Multi-view Global Dynamic Scene Modeling via Adaptive Local Implicit Feature Decoupling." pith.science (2026). https://pith.science/paper/ICXRWP7M

@misc{pith2026250702363,
  author       = {Pith},
  title        = {Pith review of: LocalDyGS: Multi-view Global Dynamic Scene Modeling via Adaptive Local Implicit Feature Decoupling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ICXRWP7M}},
  note         = {Machine review of arXiv:2507.02363}
}
read the original abstract

Due to the complex and highly dynamic motions in the real world, synthesizing dynamic videos from multi-view inputs for arbitrary viewpoints is challenging. Previous works based on neural radiance field or 3D Gaussian splatting are limited to modeling fine-scale motion, greatly restricting their application. In this paper, we introduce LocalDyGS, which consists of two parts to adapt our method to both large-scale and fine-scale motion scenes: 1) We decompose a complex dynamic scene into streamlined local spaces defined by seeds, enabling global modeling by capturing motion within each local space. 2) We decouple static and dynamic features for local space motion modeling. A static feature shared across time steps captures static information, while a dynamic residual field provides time-specific features. These are combined and decoded to generate Temporal Gaussians, modeling motion within each local space. As a result, we propose a novel dynamic scene reconstruction framework to model highly dynamic real-world scenes more realistically. Our method not only demonstrates competitive performance on various fine-scale datasets compared to state-of-the-art (SOTA) methods, but also represents the first attempt to model larger and more complex highly dynamic scenes. Project page: https://wujh2001.github.io/LocalDyGS/.

Figures

Figures reproduced from arXiv: 2507.02363 by the authors.

Figure 1
Figure 1. (a) shows our foundational idea: Decomposing a globally complex dynamic scene into a series of streamlined local spaces. The [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of existing dynamic methods. to use multi-view synchronized videos to provide dense spatiotemporal supervision [4, 24, 28, 42, 63]. Many researchers have explored multi-view dynamic scene reconstruction from different perspectives to enhance visual quality. For example, 3DGStream [39] utilizes a Neural Transformation Cache (NTC) to model each frame individually, enabling streaming dynamic scene reconstructi… view at source ↗
Figure 3
Figure 3. Overview of LocalDyGS. We sample N frames across the time domain to extract the SfM [37] point cloud, using it to initialize seeds and local spaces, with each seed assigned two learnable parameters: a static feature fs shared across all time steps, and a scale v defining the local space range. Additionally, we construct a global dynamic residual field and a weighting field to provide temporal information for the loc… view at source ↗
Figures from the paper (8 more)
Figure 5
Figure 5. Figure 5: We add seeds where the 2D projection gradient [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 4
Figure 4. Figure 4: (a) and (b) show the results decoded with static and [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 6
Figure 6. Figure 6: Qualitative results of coffee martini and sear steak from the N3DV dataset [24] (a dataset featuring fine-scale motion). We compare our method with SOTA approaches, including STGS [26] and 3DGStream [39]. Our method produces fewer floaters and preserves more details in…
Figure 7
Figure 7. Figure 7: Qualitative result on the discussion of Meetroom dataset [23] (a dataset featuring sparse views and large textureless regions). the N3DV dataset and the Meet Room dataset with current mainstream SOTA methods, including the streaming method 3DGStream [39] and non-stream…
Figure 8
Figure 8. Figure 8: Qualitative results on VRU GZ [40] (a dataset featuring large-scale, complex motion). Compared to current SOTA dynamic methods, our approach is particularly effective at adapting to large-scale, complex motion scenes. More results can be seen in our videos [PITH_FULL_…
Figure 9
Figure 9. Figure 9: A comparison of (a) to (b) shows that training using only [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Ablation study conducted on the discussion scene. in [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: Our deactivation strategy effectively reduces a [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 3 Pith papers

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

  1. DLGStream: Dynamic Language-embedded Guassian Splatting for Open-vocabulary Enabled Free-viewpoint Video Streaming

    cs.CV 2026-06 unverdicted novelty 7.0 of 10

    DLGStream introduces dual-opacity dynamic language Gaussians and an interpolation deformation field to enable language-embedded FVV streaming at 43 KB per frame with improved open-vocabulary segmentation and reconstru...

  2. ClipGStream: Clip-Stream Gaussian Splatting for Any Length and Any Motion Multi-View Dynamic Scene Reconstruction

    cs.CV 2026-04 unverdicted novelty 7.0 of 10

    ClipGStream enables scalable flicker-free reconstruction of long dynamic multi-view videos by performing stream optimization at the clip level with clip-independent spatio-temporal fields, residual anchor compensation...

  3. LiveStre4m: Feed-Forward Live Streaming of Novel Views from Unposed Multi-View Video

    cs.CV 2026-04 unverdicted novelty 6.0 of 10

    LiveStre4m delivers real-time novel-view video streaming from unposed multi-view inputs via a multi-view vision transformer, diffusion-transformer interpolation, and a learned camera pose predictor.

Reference graph

Works this paper leans on

64 extracted references · 40 canonical work pages · cited by 3 Pith papers

  1. [49]

    Swift4d: Adaptive divide-and-conquer gaussian splatting for compact and efficient reconstruction of dynamic scene

    Jiahao Wu, Rui Peng, Zhiyan Wang, Lu Xiao, Luyang Tang, Jinbo Yan, Kaiqiang Xiong, and Ronggang Wang. Swift4d: Adaptive divide-and-conquer gaussian splatting for compact and efficient reconstruction of dynamic scene. In The Thirteenth International Conference on Learning Representations. 2, 4, 6

    work page

  2. [1]

    HyperReel: High-fidelity 6-DoF video with ray-conditioned sampling

    Benjamin Attal, Jia-Bin Huang, Christian Richardt, Michael Zollhoefer, Johannes Kopf, Matthew O’Toole, and Changil Kim. HyperReel: High-fidelity 6-DoF video with ray-conditioned sampling. arXiv preprint arXiv:2301.02238,

    arXiv

  3. [2]

    https://www.avs.org.cn/

    A VS. https://www.avs.org.cn/. 2024. 6

    2024

  4. [3]

    Per-gaussian embedding-based deformation for deformable 3d gaussian splatting

    Jeongmin Bae, Seoha Kim, Youngsik Yun, Hahyun Lee, Gun Bang, and Youngjung Uh. Per-gaussian embedding-based deformation for deformable 3d gaussian splatting. arXiv preprint arXiv:2404.03613, 2024. 3

    arXiv 2024

  5. [4]

    4d visualization of dynamic events from unconstrained multi-view videos

    Aayush Bansal, Minh V o, Yaser Sheikh, Deva Ramanan, and Srinivasa Narasimhan. 4d visualization of dynamic events from unconstrained multi-view videos. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5366–5375, 2020. 2

    2020

  6. [5]

    Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields

    Jonathan T Barron, Ben Mildenhall, Matthew Tancik, Peter Hedman, Ricardo Martin-Brualla, and Pratul P Srinivasan. Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields. In Proceedings of the IEEE/CVF international conference on computer vision, pages 5855–5864, 2021. 2

    2021

  7. [6]

    Mip-nerf 360: Unbounded anti-aliased neural radiance fields

    Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P Srinivasan, and Peter Hedman. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5470–5479, 2022

    2022

  8. [7]

    Zip-nerf: Anti-aliased grid-based neural radiance fields

    Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P Srinivasan, and Peter Hedman. Zip-nerf: Anti-aliased grid-based neural radiance fields. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 19697–19705, 2023. 2

    2023

Show all 64 references
  1. [8]

    6dgs: 6d pose estimation from a single image and a 3d gaussian splatting model

    Matteo Bortolon, Theodore Tsesmelis, Stuart James, Fabio Poiesi, and Alessio Del Bue. 6dgs: 6d pose estimation from a single image and a 3d gaussian splatting model. arXiv preprint arXiv:2407.15484, 2024. 2

    2024 arXiv

  2. [9]

    Hexplane: A fast representation for dynamic scenes

    Ang Cao and Justin Johnson. Hexplane: A fast representation for dynamic scenes. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 130–141, 2023. 2, 6

    2023

  3. [10]

    Tensorf: Tensorial radiance fields

    Anpei Chen, Zexiang Xu, Andreas Geiger, Jingyi Yu, and Hao Su. Tensorf: Tensorial radiance fields. In European conference on computer vision , pages 333–350. Springer,

  4. [11]

    4d-rotor gaussian splatting: towards efficient novel view synthesis for dynamic scenes

    Yuanxing Duan, Fangyin Wei, Qiyu Dai, Yuhang He, Wenzheng Chen, and Baoquan Chen. 4d-rotor gaussian splatting: towards efficient novel view synthesis for dynamic scenes. In ACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024. 3

    2024

  5. [12]

    Instantsplat: Unbounded sparse-view pose-free gaussian splatting in 40 seconds

    Zhiwen Fan, Wenyan Cong, Kairun Wen, Kevin Wang, Jian Zhang, Xinghao Ding, Danfei Xu, Boris Ivanovic, Marco Pavone, Georgios Pavlakos, et al. Instantsplat: Unbounded sparse-view pose-free gaussian splatting in 40 seconds. arXiv preprint arXiv:2403.20309, 2024. 2

    2024 arXiv

  6. [13]

    Plenoxels: Radiance fields without neural networks

    Sara Fridovich-Keil, Alex Yu, Matthew Tancik, Qinhong Chen, Benjamin Recht, and Angjoo Kanazawa. Plenoxels: Radiance fields without neural networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5501–5510, 2022. 2, 6

    2022

  7. [14]

    K-planes: Explicit radiance fields in space, time, and appearance

    Sara Fridovich-Keil, Giacomo Meanti, Frederik Rahbæk Warburg, Benjamin Recht, and Angjoo Kanazawa. K-planes: Explicit radiance fields in space, time, and appearance. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12479–12488, 2023. 2, 6

    2023

  8. [15]

    Efros, and Xiaolong Wang

    Yang Fu, Sifei Liu, Amey Kulkarni, Jan Kautz, Alexei A. Efros, and Xiaolong Wang. Colmap-free 3d gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 20796–20805,

  9. [16]

    Dynamic view synthesis from dynamic monocular video

    Chen Gao, Ayush Saraf, Johannes Kopf, and Jia-Bin Huang. Dynamic view synthesis from dynamic monocular video. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 5712–5721, 2021. 2

    2021

  10. [17]

    Tri-miprf: Tri-mip representation for efficient anti-aliasing neural radiance fields

    Wenbo Hu, Yuling Wang, Lin Ma, Bangbang Yang, Lin Gao, Xiao Liu, and Yuewen Ma. Tri-miprf: Tri-mip representation for efficient anti-aliasing neural radiance fields. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 19774–19783, 2023. 2

    2023

  11. [18]

    2d gaussian splatting for geometrically accurate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically accurate radiance fields. arXiv preprint arXiv:2403.17888,

    arXiv

  12. [19]

    Sc-gs: Sparse-controlled gaussian splatting for editable dynamic scenes

    Yi-Hua Huang, Yang-Tian Sun, Ziyi Yang, Xiaoyang Lyu, Yan-Pei Cao, and Xiaojuan Qi. Sc-gs: Sparse-controlled gaussian splatting for editable dynamic scenes. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4220–4230, 2024. 2, 3

    2024

  13. [20]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4):139–1, 2023. 2, 3, 5, 6, 7

    2023

  14. [21]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 ,

    arXiv

  15. [22]

    Dynmf: Neural motion factorization for real-time dynamic view synthesis with 3d gaussian splatting

    Agelos Kratimenos, Jiahui Lei, and Kostas Daniilidis. Dynmf: Neural motion factorization for real-time dynamic view synthesis with 3d gaussian splatting. arXiv preprint arXiv:2312.00112, 2023. 3

    2023 arXiv

  16. [23]

    Streaming radiance fields for 3d video synthesis

    Lingzhi Li, Zhen Shen, Zhongshu Wang, Li Shen, and Ping Tan. Streaming radiance fields for 3d video synthesis. Advances in Neural Information Processing Systems , 35: 13485–13498, 2022. 6, 7, 9

    2022

  17. [24]

    Neural 3d video synthesis from multi-view video

    Tianye Li, Mira Slavcheva, Michael Zollhoefer, Simon Green, Christoph Lassner, Changil Kim, Tanner Schmidt, Steven Lovegrove, Michael Goesele, Richard Newcombe, et al. Neural 3d video synthesis from multi-view video. In Proceedings of the IEEE/CVF Conference on Computer Vision...

    2022

  18. [25]

    Neural scene flow fields for space-time view synthesis of dynamic scenes

    Zhengqi Li, Simon Niklaus, Noah Snavely, and Oliver Wang. Neural scene flow fields for space-time view synthesis of dynamic scenes. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 6498–6508, 2021. 2, 6

    2021

  19. [26]

    Spacetime gaussian feature splatting for real-time dynamic view synthesis

    Zhan Li, Zhang Chen, Zhong Li, and Yi Xu. Spacetime gaussian feature splatting for real-time dynamic view synthesis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8508–8520,

  20. [27]

    Gaussian-flow: 4d reconstruction with dynamic 3d gaussian particle

    Youtian Lin, Zuozhuo Dai, Siyu Zhu, and Yao Yao. Gaussian-flow: 4d reconstruction with dynamic 3d gaussian particle. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 21136–21145, 2024. 3

    2024

  21. [28]

    Neural volumes: Learning dynamic renderable volumes from images

    Stephen Lombardi, Tomas Simon, Jason Saragih, Gabriel Schwartz, Andreas Lehrmann, and Yaser Sheikh. Neural volumes: Learning dynamic renderable volumes from images. arXiv preprint arXiv:1906.07751, 2019. 2

    1906 arXiv

  22. [29]

    Mixture of volumetric primitives for efficient neural rendering

    Stephen Lombardi, Tomas Simon, Gabriel Schwartz, Michael Zollhoefer, Yaser Sheikh, and Jason Saragih. Mixture of volumetric primitives for efficient neural rendering. ACM Transactions on Graphics (ToG), 40(4):1–13, 2021. 5

    2021

  23. [30]

    Scaffold-gs: Structured 3d gaussians for view-adaptive rendering

    Tao Lu, Mulin Yu, Linning Xu, Yuanbo Xiangli, Limin Wang, Dahua Lin, and Bo Dai. Scaffold-gs: Structured 3d gaussians for view-adaptive rendering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 20654–20664, 2024. 2, 4, 5, 8

    2024

  24. [31]

    Instant neural graphics primitives with a multiresolution hash encoding

    Thomas M ¨uller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding. ACM transactions on graphics (TOG), 41(4):1–15, 2022. 4, 6

    2022

  25. [32]

    Nerfies: Deformable neural radiance fields

    Keunhong Park, Utkarsh Sinha, Jonathan T Barron, Sofien Bouaziz, Dan B Goldman, Steven M Seitz, and Ricardo Martin-Brualla. Nerfies: Deformable neural radiance fields. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 5865–5874, 2021. 2

    2021

  26. [33]

    Hypernerf: A higher-dimensional representation for topologically varying neural radiance fields

    Keunhong Park, Utkarsh Sinha, Peter Hedman, Jonathan T Barron, Sofien Bouaziz, Dan B Goldman, Ricardo Martin-Brualla, and Steven M Seitz. Hypernerf: A higher-dimensional representation for topologically varying neural radiance fields. arXiv preprint arXiv:2106.13228, 2021. 1, 3, 4

    2021 arXiv

  27. [34]

    D-nerf: Neural radiance fields for dynamic scenes

    Albert Pumarola, Enric Corona, Gerard Pons-Moll, and Francesc Moreno-Noguer. D-nerf: Neural radiance fields for dynamic scenes. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 10318–10327, 2021. 1, 2, 3

    2021

  28. [35]

    Stopthepop: Sorted gaussian splatting for view-consistent real-time rendering

    Lukas Radl, Michael Steiner, Mathias Parger, Alexander Weinrauch, Bernhard Kerbl, and Markus Steinberger. Stopthepop: Sorted gaussian splatting for view-consistent real-time rendering. ACM Transactions on Graphics (TOG), 43(4):1–17, 2024. 2

    2024

  29. [36]

    Octree-gs: Towards consistent real-time rendering with lod-structured 3d gaussians

    Kerui Ren, Lihan Jiang, Tao Lu, Mulin Yu, Linning Xu, Zhangkai Ni, and Bo Dai. Octree-gs: Towards consistent real-time rendering with lod-structured 3d gaussians. arXiv preprint arXiv:2403.17898, 2024. 2

    2024 arXiv

  30. [37]

    Structure-from-motion revisited

    Johannes L Schonberger and Jan-Michael Frahm. Structure-from-motion revisited. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4104–4113, 2016. 2, 3

    2016

  31. [38]

    Nerfplayer: A streamable dynamic scene representation with decomposed neural radiance fields

    Liangchen Song, Anpei Chen, Zhong Li, Zhang Chen, Lele Chen, Junsong Yuan, Yi Xu, and Andreas Geiger. Nerfplayer: A streamable dynamic scene representation with decomposed neural radiance fields. IEEE Transactions on Visualization and Computer Graphics, 29(5):2732–2742, 2023. 2, 6

    2023

  32. [39]

    3dgstream: On-the-fly training of 3d gaussians for efficient streaming of photo-realistic free-viewpoint videos

    Jiakai Sun, Han Jiao, Guangyuan Li, Zhanjie Zhang, Lei Zhao, and Wei Xing. 3dgstream: On-the-fly training of 3d gaussians for efficient streaming of photo-realistic free-viewpoint videos. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pa...

    2024

  33. [40]

    https://anonymous.4open.science/r/vru-sequence/

    VRU. https://anonymous.4open.science/r/vru-sequence/

  34. [41]

    Superpoint gaussian splatting for real-time high-fidelity dynamic scene reconstruction

    Diwen Wan, Ruijie Lu, and Gang Zeng. Superpoint gaussian splatting for real-time high-fidelity dynamic scene reconstruction. arXiv preprint arXiv:2406.03697, 2024. 2, 3

    2024 arXiv

  35. [42]

    Mixed neural voxels for fast multi-view video synthesis

    Feng Wang, Sinan Tan, Xinghang Li, Zeyue Tian, and Huaping Liu. Mixed neural voxels for fast multi-view video synthesis. arXiv preprint arXiv:2212.00190, 2022. 2, 6

    2022 arXiv

  36. [43]

    Mixed neural voxels for fast multi-view video synthesis

    Feng Wang, Sinan Tan, Xinghang Li, Zeyue Tian, Yafei Song, and Huaping Liu. Mixed neural voxels for fast multi-view video synthesis. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 19706–19716, 2023. 2, 6

    2023

  37. [44]

    Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction

    Peng Wang, Lingjie Liu, Yuan Liu, Christian Theobalt, Taku Komura, and Wenping Wang. Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction. arXiv preprint arXiv:2106.10689, 2021. 2

    2021 arXiv

  38. [45]

    Shape of motion: 4d reconstruction from a single video

    Qianqian Wang, Vickie Ye, Hang Gao, Weijia Zeng, Jake Austin, Zhengqi Li, and Angjoo Kanazawa. Shape of motion: 4d reconstruction from a single video. 2024. 3

    2024

  39. [46]

    Neus2: Fast learning of neural implicit surfaces for multi-view reconstruction

    Yiming Wang, Qin Han, Marc Habermann, Kostas Daniilidis, Christian Theobalt, and Lingjie Liu. Neus2: Fast learning of neural implicit surfaces for multi-view reconstruction. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3295–3306, 2023. 2

    2023

  40. [47]

    4d gaussian splatting for real-time dynamic scene rendering

    Guanjun Wu, Taoran Yi, Jiemin Fang, Lingxi Xie, Xiaopeng Zhang, Wei Wei, Wenyu Liu, Qi Tian, and Xinggang Wang. 4d gaussian splatting for real-time dynamic scene rendering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20310–20320,...

    2024

  41. [48]

    Dynamic lidar re-simulation using compositional neural fields

    Hanfeng Wu, Xingxing Zuo, Stefan Leutenegger, Or Litany, Konrad Schindler, and Shengyu Huang. Dynamic lidar re-simulation using compositional neural fields. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19988–19998, 2024. 2, 4

    2024

  42. [50]

    Hdrgs: High dynamic range gaussian splatting

    Jiahao Wu, Lu Xiao, Rui Peng, Kaiqiang Xiong, and Ronggang Wang. Hdrgs: High dynamic range gaussian splatting. arXiv preprint arXiv:2408.06543, 2024. 2

    2024 arXiv

  43. [51]

    Mars: An instance-aware, modular and realistic simulator for autonomous driving

    Zirui Wu, Tianyu Liu, Liyi Luo, Zhide Zhong, Jianteng Chen, Hongmin Xiao, Chao Hou, Haozhe Lou, Yuantao Chen, Runyi Yang, et al. Mars: An instance-aware, modular and realistic simulator for autonomous driving. In CAAI International Conference on Artificial Intelligence , pages...

    2023

  44. [52]

    Multi-view image enhancement inconsistency decoupling guided 3d gaussian splatting

    Lu Xiao, Jiahao Wu, Zhanke Wang, Guanhua Wu, Runling Liu, Zhiyan Wang, and Ronggang Wang. Multi-view image enhancement inconsistency decoupling guided 3d gaussian splatting. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), ...

    2025

  45. [53]

    Instant gaussian stream: Fast and generalizable streaming of dynamic scene reconstruction via gaussian splatting

    Jinbo Yan, Rui Peng, Zhiyan Wang, Luyang Tang, Jiayu Yang, Jie Liang, Jiahao Wu, and Ronggang Wang. Instant gaussian stream: Fast and generalizable streaming of dynamic scene reconstruction via gaussian splatting. In Proceedings of the Computer Vision and Pattern Recognition C...

    2025

  46. [54]

    Nerf-ds: Neural radiance fields for dynamic specular objects

    Zhiwen Yan, Chen Li, and Gim Hee Lee. Nerf-ds: Neural radiance fields for dynamic specular objects. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8285–8295, 2023. 1, 3

    2023

  47. [55]

    Real-time photorealistic dynamic scene representation and rendering with 4d gaussian splatting

    Zeyu Yang, Hongye Yang, Zijie Pan, Xiatian Zhu, and Li Zhang. Real-time photorealistic dynamic scene representation and rendering with 4d gaussian splatting. arXiv preprint arXiv:2310.10642, 2023. 3, 6

    2023 arXiv

  48. [56]

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

    Ziyi Yang, Xinyu Gao, Wen Zhou, Shaohui Jiao, Yuqing Zhang, and Xiaogang Jin. Deformable 3d gaussians for high-fidelity monocular dynamic scene reconstruction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20331–20341, 2024. 3, 4

    2024

  49. [57]

    Viewcrafter: Taming video diffusion models for high-fidelity novel view synthesis

    Wangbo Yu, Jinbo Xing, Li Yuan, Wenbo Hu, Xiaoyu Li, Zhipeng Huang, Xiangjun Gao, Tien-Tsin Wong, Ying Shan, and Yonghong Tian. Viewcrafter: Taming video diffusion models for high-fidelity novel view synthesis. arXiv preprint arXiv:2409.02048, 2024. 2

    2024 arXiv

  50. [58]

    Gaussian opacity fields: Efficient and compact surface reconstruction in unbounded scenes

    Zehao Yu, Torsten Sattler, and Andreas Geiger. Gaussian opacity fields: Efficient and compact surface reconstruction in unbounded scenes. arXiv preprint arXiv:2404.10772, 2024. 2, 6

    2024 arXiv

  51. [59]

    Rade-gs: Rasterizing depth in gaussian splatting

    Baowen Zhang, Chuan Fang, Rakesh Shrestha, Yixun Liang, Xiaoxiao Long, and Ping Tan. Rade-gs: Rasterizing depth in gaussian splatting. arXiv preprint arXiv:2406.01467, 2024. 2

    2024 arXiv

  52. [60]

    Gaussianprediction: Dynamic 3d gaussian prediction for motion extrapolation and free view synthesis

    Boming Zhao, Yuan Li, Ziyu Sun, Lin Zeng, Yujun Shen, Rui Ma, Yinda Zhang, Hujun Bao, and Zhaopeng Cui. Gaussianprediction: Dynamic 3d gaussian prediction for motion extrapolation and free view synthesis. In ACM SIGGRAPH 2024 Conference Papers, pages 1–12, 2024. 3

    2024

  53. [61]

    Nice-slam: Neural implicit scalable encoding for slam

    Zihan Zhu, Songyou Peng, Viktor Larsson, Weiwei Xu, Hujun Bao, Zhaopeng Cui, Martin R Oswald, and Marc Pollefeys. Nice-slam: Neural implicit scalable encoding for slam. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12786–12796, 2022. 2

    2022

  54. [62]

    Fsgs: Real-time few-shot view synthesis using gaussian splatting

    Zehao Zhu, Zhiwen Fan, Yifan Jiang, and Zhangyang Wang. Fsgs: Real-time few-shot view synthesis using gaussian splatting. In European Conference on Computer Vision, pages 145–163. Springer, 2025. 2

    2025

  55. [63]

    High-quality video view interpolation using a layered representation

    C Lawrence Zitnick, Sing Bing Kang, Matthew Uyttendaele, Simon Winder, and Richard Szeliski. High-quality video view interpolation using a layered representation. ACM transactions on graphics (TOG), 23(3):600–608, 2004. 2

    2004

  56. [64]

    Ewa splatting

    Matthias Zwicker, Hanspeter Pfister, Jeroen Van Baar, and Markus Gross. Ewa splatting. IEEE Transactions on Visualization and Computer Graphics, 8(3):223–238, 2002. 3

    2002

Pith tools

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