Pith. sign in

REVIEW 4 major objections 5 minor 72 references

FillGS argues that spatiotemporal observation gaps in 4D Gaussian splatting should be located by a motion-aware deficiency score and filled with diffusion-refined views, improving rendering fidelity under sparse view coverage.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 09:58 UTC pith:SPQCMD6J

load-bearing objection A solid engineering pipeline for active sparse-view 4DGS refinement; the central deficiency score is plausible but under-validated and the selection ablation is confounded with the weighting scheme. the 4 major comments →

arxiv 2607.29284 v1 pith:SPQCMD6J submitted 2026-07-31 cs.CV

FillGS: Filling Observation Gaps in 4D Gaussian Splatting via Viewpoint-Time Selection and Generative Refinement

Takashi Otonari , Toshihiko Yamasaki This is my paper
classification cs.CV
keywords 4D Gaussian Splattingsparse-view dynamic reconstructionvirtual viewpoint selectiongenerative refinementmotion-aware observation densityvideo diffusion modelnovel view synthesisobservation gap filling
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper claims that under sparse camera coverage, dynamic scenes rendered with 4D Gaussian Splatting suffer artifacts because some spatiotemporal regions are barely observed, and existing generative-refinement methods pick virtual viewpoints by heuristics like interpolation or spatial coverage that ignore where the gaps actually are. FillGS instead actively selects virtual viewpoints by scoring candidates according to a motion-aware deficiency signal per Gaussian, then refines the rendered views with a video diffusion model and fine-tunes the 4D model using only pixels that are consistent with real observations. The central claim is that this active spatiotemporal selection and reliability-weighted fine-tuning consistently improve rendering quality over prior viewpoint-selection and fine-tuning strategies. The authors introduce new sparse train/test splits with only two training cameras and report gains across PSNR, SSIM, LPIPS, FID, and DINOv2 similarity in both interpolation and extrapolation settings.

Core claim

The paper's central claim is that the effectiveness of generative refinement for 4D Gaussian Splatting depends on which virtual viewpoints are chosen, and that the right criterion is spatiotemporal observation sparsity rather than spatial coverage or interpolation. It defines a per-Gaussian, per-time deficiency score D = speed / (1 + observation count), which marks fast-moving and rarely seen Gaussians as under-constrained. Candidate virtual viewpoints are scored by rendering this deficiency field and averaging it over the view, and the highest-scoring candidates are refined with a video diffusion model. During fine-tuning, the same deficiency score down-weights already well-observed regions

What carries the argument

The motion-aware observation deficiency score D_gi(t) = s_gi(t) / (1 + C_gi(t)), where s_gi(t) is the speed of Gaussian g_i at time t and C_gi(t) is its observation count. This single scalar plays two roles: it selects virtual viewpoints by ranking candidates that observe many high-deficiency Gaussians, and it weights the fine-tuning loss so that sparsely observed, fast-moving regions receive stronger supervision from generated images while already-constrained regions are left mostly unchanged. A sensitivity-based starting-point selection (a finite-difference approximation to local rendering instability) seeds the exploration, and a consistency mask from feature matching filters unreliable g

Load-bearing premise

The load-bearing premise is that a Gaussian's deficiency, computed as speed divided by (1 + observation count), faithfully measures how much a new virtual observation of that Gaussian would improve the reconstruction; if this score misranks viewpoints, both the selection of where to generate and the weighting of what to fine-tune are pointed at the wrong places.

What would settle it

Train a 4DGS model on a sparse two-camera split, compute D for each Gaussian, and render several candidate virtual viewpoints with known deficiency scores. For each candidate, generate a diffusion-refined image, fine-tune the model using only that image (with the proposed masks), and measure the actual PSNR or LPIPS improvement on held-out views. If the rank correlation between the deficiency score and the achieved improvement is not positive, the selection signal is not doing the claimed work.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Under two-camera training splits, the full pipeline outperforms interpolation-, Fisher-information-, and coverage-based selection on PSNR, SSIM, LPIPS, FID, and DINOv2 similarity in both interpolation and extrapolation test settings.
  • Ablations show that each component—starting-point selection, candidate scoring, the consistency mask, and the co-visibility mask—contributes to the final gain; removing any one degrades results.
  • The method improves fine-tuning over prior weighting schemes when the 4DGS backbone and the generative model are held fixed, indicating that the gains come from selection and weighting rather than from a stronger generator.
  • The approach transfers to a different 4DGS backbone and a different diffusion generator, and also improves monocular capture settings, suggesting the mechanism is not tied to one architecture.
  • The method is sample-efficient: even a single virtual trajectory yields competitive results, and ten trajectories outperform twenty-trajectory baselines at lower generation cost.

Where Pith is reading between the lines

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

  • If the deficiency score is a valid proxy for information gain, the same selection criterion could be reused for physical next-best-view planning in dynamic capture, since it targets under-constrained Gaussians rather than spatial coverage and could guide where to point real cameras.
  • The consistency-masked fine-tuning suggests a general recipe for any generative-depth or diffusion-based supervision: only back-propagate pixels that agree with real observations, treating generated content as a constraint on under-observed regions rather than as ground truth.
  • A direct test of the score's validity would be to compare D_gi(t) against the actual reduction in reconstruction error obtained by adding a generated view; if the correlation is weak, the selection stage may be succeeding for other reasons, such as the global candidate sampling or the fine-tuning masks.
  • The method's dependence on the diffusion model's generation quality is acknowledged in the paper; as generators improve, the same selection-and-filtering pipeline could be expected to yield larger gains without architectural changes.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes FillGS, a pipeline for 4D Gaussian Splatting (4DGS) under sparse multi-view dynamic captures. It actively selects spatiotemporal virtual viewpoints using a motion-aware observation-deficiency score D_gi(t) = s_gi(t)/(1 + C_gi(t)), where s is Gaussian speed and C is the observation count (Eq. 4). Rendered images from these viewpoints are refined by a video diffusion model, and the 4DGS model is fine-tuned using consistency masks and co-visibility masks that down-weight unreliable or already well-constrained regions. The paper introduces new sparse train/test splits on Neural 3D Video, Technicolor, and Nvidia datasets and reports consistent improvements over E-D3DGS, interpolation-based selection, FisherRF-based selection, and coverage-based selection, with ablations on the viewpoint-selection components and fine-tuning strategy.

Significance. If validated, FillGS addresses a real gap: prior diffusion-based 4DGS refinement methods use heuristic viewpoint selection (random, interpolation, or spatial coverage) and do not account for the spatiotemporal nature of observation sparsity. The paper has several strengths: the experimental protocol controls the backbone and generative model across baselines; the ablations examine both viewpoint selection and fine-tuning; and the supplementary provides experiments with a different backbone (Ex4DGS), a different diffusion model (Difix3D+), a monocular setting, computational cost, and failure cases. The main weakness is that the load-bearing deficiency score in Eq. (4) is never validated against actual reconstruction gain, and because the same score is used in the fine-tuning weights, the selection ablations are confounded. The paper is promising but the central active-selection claim is not yet fully supported.

major comments (4)
  1. [Sec. 3.3 and Sec. 3.5, Eq. (4)] D_gi(t)=s_gi(t)/(1+C_gi(t)) is presented as the central novelty, but the paper never validates that this score is a proxy for information gain or reconstruction improvement. The only quantitative support is Table 3, where removing speed or count, or inverting the score, degrades results. However, the same D is used as the co-visibility weight in fine-tuning (Sec. 3.5), so the selection ablations are confounded: a change in selection also changes the loss weighting. To support the active-selection claim, the authors should (i) fix the fine-tuning weighting while varying only the selection score, and (ii) provide a direct validation of D against actual per-pixel reconstruction error or information gain, e.g., by comparing candidate viewpoints with high vs. low D after adding a generated view.
  2. [Tables 1-4 and supplementary tables] All results are reported as single runs with no error bars, confidence intervals, or significance tests. In sparse-view settings, FID is especially noisy when computed over a small number of test views; the reported margins (e.g., Table 1 interpolation PSNR 21.73 vs. 21.15 for interpolation-based selection, SSIM 0.792 vs. 0.776) may be within run-to-run variation. Please report at least three seeds with mean and standard deviation, or paired significance tests.
  3. [Sec. 4.1, Fig. 3] The evaluation uses new train/test splits that are not released. Without the exact camera/time indices or code to generate them, other researchers cannot reproduce or compare against the numbers. Releasing the splits (and the evaluation protocol) is necessary for the empirical claim to be verifiable.
  4. [Abstract and Sec. 1] The claim that FillGS improves rendering 'without relying on hand-crafted heuristics' is overstated. Eq. (4) is itself a hand-specified heuristic, and the pipeline has several manually chosen constants: the normalization constant L-bar in Eq. (2), perturbation weights w_delta and perturbation magnitudes in Eq. (3) and the supplementary, and candidate-generation hyperparameters (e.g., 12 local candidates, 10-degree rotations). Please either soften the claim or provide evidence that the results are robust to reasonable variation in these choices.
minor comments (5)
  1. [Sec. 3.2, Eqs. (2)-(3)] The normalization constants are described verbally ('dataset-level mean is 1', 'averages to 1'); please give explicit formulas or a clear pointer to where they are defined.
  2. [Table 3] The row 'Negative selection' should state explicitly whether it inverts the score only for viewpoint selection or also for the fine-tuning co-visibility weights, since both use D.
  3. [Fig. 2] The labels 'GS0.1 m/s GS1 m/s' near the density color bar are confusing; rephrase to clarify what the color scale represents.
  4. [Sec. 3.4] The patch construction (3 frames x 4 views) and conditioning are only sketched. Please provide precise implementation details in the main text or a more explicit pointer to the supplementary, as this is essential for reproducing the diffusion model.
  5. [Sec. 3.2 and Supp. B.3] The text says the sensitivity score is 'analyzed on synthetic data in the supplementary,' but Supp. B.3 only visualizes the score over space and time; it does not quantitatively correlate the score with reconstruction error. The wording should be adjusted to match what is actually shown.

Circularity Check

0 steps flagged

No significant circularity: the empirical pipeline does not reduce its outputs to its inputs.

full rationale

The paper's derivation chain is empirical rather than definitional. The motion-aware observation-deficiency score D_gi(t)=s_gi(t)/(1+C_gi(t)) (Eq. 4) is explicitly introduced as a heuristic: 'a Gaussian with few observations and high speed receives a high deficiency score, indicating that it is insufficiently constrained by the captured views.' The same score is used both for virtual-viewpoint selection (Sec. 3.3) and for normalizing fine-tuning weights (Sec. 3.5), which is a potential experimental confound for attributing gains specifically to viewpoint selection, but it is not a circular derivation: the reported PSNR/SSIM/FID values are measured on held-out test views after optimization and are not algebraic consequences of D. The 'Relation to information gain' paragraph is explicitly interpretive ('can be interpreted as a proxy'), not a proof that D equals information gain. The method is evaluated against external baselines (interpolation-based selection, FisherRF, ExploreGS coverage) and component ablations, and the diffusion model is initialized from DL3DV and fine-tuned on other dynamic datasets rather than the test benchmarks. The authors do not cite their own prior work, invoke no uniqueness theorem, and import no load-bearing result from self-citations. No equation or predicted quantity reduces to its own input by construction, so there is no significant circularity.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

The pipeline is built on several hand-specified proxies and domain assumptions. The most important are the observation-deficiency score, the sensitivity-to-uncertainty link, and the trustworthiness of the diffusion model's outputs after masking. These are not derived from first principles and are not independently validated, so they carry much of the explanatory burden of the paper.

free parameters (5)
  • L-bar normalization constant = dataset-level mean of L set to 1
    Eq. 2: ad hoc normalization so the error term and sensitivity term are comparable; affects starting-viewpoint ranking.
  • perturbation weights w_delta = normalized so each delta contributes 1/|D| on average
    Eq. 3: hand-set to balance translation, rotation, and time sensitivity channels.
  • sensitivity perturbation magnitudes = 1/5 voxel, 0.2 deg, 1 frame interval
    Supp. A: chosen by hand; V(v) depends on them, and no grid search is reported.
  • candidate-generation hyperparameters = 12 local + 12 global candidates, 64 grid divisions, step = 2x body diagonal
    Sec. 3.3 and Supp. A: discretization choices that define which virtual viewpoints are reachable.
  • fine-tuning schedule = 30k iterations, 20 virtual trajectories every 6k iterations
    Sec. 4.1: chosen without a reported ablation on these values; Table 5 explores trajectory count only.
axioms (4)
  • ad hoc to paper Rendering sensitivity under small pose/time perturbations is a proxy for reconstruction uncertainty and error-prone neighborhoods
    Sec. 3.2: asserted as a connection to epistemic uncertainty; only synthetic supporting evidence is given.
  • ad hoc to paper D_gi(t)=s_gi(t)/(1+C_gi(t)) measures how much a new observation improves conditioning
    Sec. 3.3: the motion-aware observation-deficiency score is defined, not derived; the 'relation to information gain' is interpretive.
  • domain assumption Video diffusion model refined images, after consistency and co-visibility masking, can be used as reliable pseudo-observations
    Sec. 3.4–3.5: correctness depends on the generator being good enough, acknowledged as a limitation.
  • domain assumption The new 2-camera train/test splits from Neural 3D Video and Technicolor are a valid proxy for real sparse-view settings
    Sec. 4.1: authors create these splits; no external validation that they capture real capture conditions.

pith-pipeline@v1.3.0-daily-deepseek · 18139 in / 12681 out tokens · 136125 ms · 2026-08-03T09:58:51.972542+00:00 · methodology

0 comments
read the original abstract

4D Gaussian Splatting (4DGS) can render dynamic scenes photorealistically. However, with limited viewpoint coverage, some spatiotemporal regions remain sparsely observed, leading to artifacts, particularly in scenes with large motion. Existing approaches leveraging generative models rely on heuristic virtual-viewpoint selection before refining rendered views. As a result, they cannot actively explore such sparsely observed regions. To address this issue, we propose a pipeline that actively selects spatiotemporal virtual viewpoints to improve 4DGS reconstruction. Our method selects virtual viewpoints for generative enhancement based on the rendering sensitivity and motion-aware observation density of 4D Gaussians, prioritizing views that alleviate observation sparsity. In the refined images, we filter out regions that conflict with captured observations or are likely to contain generative artifacts and then fine-tune 4DGS using only the reliable regions. We evaluate our method on multi-view video benchmarks using new train/test splits designed to induce observation gaps. Results show consistent improvements over prior viewpoint selection strategies and fine-tuning methods in both qualitative and quantitative evaluations, while reducing artifacts.

Figures

Figures reproduced from arXiv: 2607.29284 by Takashi Otonari, Toshihiko Yamasaki.

Figure 1
Figure 1. Figure 1: 4D Gaussian Splatting (4DGS) suffers from limited spatiotemporal observa￾tions. In this paper, we address the challenge of insufficient spatiotemporal observations by actively selecting viewpoints to identify spatiotemporal gaps and then refining the rendered images with a generative model. Our method enables high-fidelity rendering from novel spatiotemporal viewpoints. However, 4DGS is highly sensitive to… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our FillGS pipeline. (i) We first train a 4DGS model. (ii) We then actively select virtual viewpoints. To do so, we first identify training views that are locally unstable due to sparse observations (Sec. 3.2), and then generate candidate virtual viewpoints using both local and global strategies. Each candidate is scored using a motion-aware observation-density metric, and we prioritize candida… view at source ↗
Figure 3
Figure 3. Figure 3: Our sparse-view camera split. We train on two cameras and evaluate on spatially separated views under both interpolation and extrapolation splits. Red cam￾eras indicate training cameras, and blue cameras indicate testing cameras. 4 Experiments 4.1 Experimental Settings Implementation details. We adopt E-D3DGS [2] as our baseline, initializing it according to the original training configuration. The model i… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison on Neural 3D Video Dataset [23]. [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison on Technicolor Dataset [45]. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of virtual trajectory sampling. [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Evaluation of a starting-viewpoint selection strategy using a syn [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative results of the ablation study on our viewpoint selection. [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative results of the ablation study on our fine-tuning methods. [PITH_FULL_IMAGE:figures/full_fig_p023_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative comparison on the Nvidia Video Dataset [70]. [PITH_FULL_IMAGE:figures/full_fig_p024_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Failure cases. Our pipeline has difficulty reconstructing complex motions. B.7 Failure Cases Our pipeline may struggle to accurately reconstruct scenes due to the limited representational capacity of the underlying 4DGS representation and the insuf￾ficient quality of the generated results. As shown in [PITH_FULL_IMAGE:figures/full_fig_p025_11.png] view at source ↗

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

72 extracted references · 8 linked inside Pith

  1. [1]

    In: 3DV (2026)

    Azzarelli, A., Anantrasirichai, N., Bull, D.R.: Splatography: Sparse multi-view dy- namic gaussian splatting for film-making challenges. In: 3DV (2026)

    2026

  2. [2]

    In: ECCV (2024)

    Bae, J., Kim, S., Yun, Y., Lee, H., Bang, G., Uh, Y.: Per-gaussian embedding-based deformation for deformable 3d gaussian splatting. In: ECCV (2024)

    2024

  3. [3]

    In: ICCV (2025)

    Bose, S., Dutta, A., Nag, S., Zhang, J., Li, J., Karydis, K., Chowdhury, A.K.R.: Uncertainty-aware diffusion-guided refinement of 3d scenes. In: ICCV (2025)

    2025

  4. [4]

    Cao,A.,Johnson,J.:Hexplane:Afastrepresentationfordynamicscenes.In:CVPR (2023)

    2023

  5. [5]

    In: SIGGRAPH (2024)

    Duan, Y., Wei, F., Dai, Q., He, Y., Chen, W., Chen, B.: 4d-rotor gaussian splatting: Towards efficient novel-view synthesis for dynamic scenes. In: SIGGRAPH (2024)

    2024

  6. [6]

    arXiv preprint arXiv:2511.15706 (2025)

    Edstedt, J., Nordström, D., Zhang, Y., Bökman, G., Astermark, J., Larsson, V., Heyden, A., Kahl, F., Wadenbäck, M., Felsberg, M.: Roma v2: Harder better faster denser feature matching. arXiv preprint arXiv:2511.15706 (2025)

    Pith/arXiv arXiv 2025

  7. [7]

    arXiv preprint arXiv:2503.14665 (2025)

    Ewen, P., Chen, H., Isaacson, S., Wilson, J., Skinner, K.A., Vasudevan, R.: These magic moments: Differentiable uncertainty quantification of radiance field models. arXiv preprint arXiv:2503.14665 (2025)

    Pith/arXiv arXiv 2025

  8. [8]

    In: NeurIPS (2022)

    Gao, H., Li, R., Tulsiani, S., Russell, B., Kanazawa, A.: Monocular dynamic view synthesis: A reality check. In: NeurIPS (2022)

    2022

  9. [9]

    In: ICLR (2026)

    Guo, F., Hsu, C.C., Ding, S., Zhang, C.: Uncertainty matters in dynamic gaussian splatting for monocular 4d reconstruction. In: ICLR (2026)

    2026

  10. [10]

    In: NeurIPS (2017)

    Heusel,M.,Ramsauer,H.,Unterthiner,T.,Nessler,B.,Hochreiter,S.:Ganstrained by a two time-scale update rule converge to a local nash equilibrium. In: NeurIPS (2017)

    2017

  11. [11]

    arXiv preprint arXiv:2512.25073 (2025)

    Huang, Y.C., Chien, H.J., Lin, C.Y., Chen, Y.H., Liu, Y.L.: Gamo: Geometry- aware multi-view diffusion outpainting for sparse-view 3d reconstruction. arXiv preprint arXiv:2512.25073 (2025)

    Pith/arXiv arXiv 2025

  12. [12]

    In: ECCV (2024)

    Jiang, W., Lei, B., Daniilidis, K.: Fisherrf: Active view selection and mapping with radiance fields using fisher information. In: ECCV (2024)

    2024

  13. [13]

    In: IROS (2023) 16 T

    Jin, L., Chen, X., Rückin, J., Popović, M.: Neu-nbv: Next best view planning using uncertainty estimation in image-based neural rendering. In: IROS (2023) 16 T. Otonari and T. Yamasaki

    2023

  14. [14]

    In: ECCV (2024)

    Katsumata, K., Vo, D.M., Nakayama, H.: A compact dynamic 3d gaussian repre- sentation for real-time dynamic view synthesis. In: ECCV (2024)

    2024

  15. [15]

    ACM TOG42(4) (2023)

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM TOG42(4) (2023)

    2023

  16. [16]

    In: NeurIPS (2024)

    Kim, M., Lim, J., Han, B.: 4d gaussian splatting in the wild with uncertainty-aware regularization. In: NeurIPS (2024)

    2024

  17. [17]

    In: ICCV (2025)

    Kim, M., Jeon, S., Cho, I., Yoo, M., Kim, S.J.: Exploregs: Explorable 3d scene reconstruction with virtual camera samplings and diffusion priors. In: ICCV (2025)

    2025

  18. [18]

    In: CVPR (2025)

    Kong, H., Yang, X., Wang, X.: Generative sparse-view gaussian splatting. In: CVPR (2025)

    2025

  19. [19]

    In: VMV (2023)

    Kopanas, G., Drettakis, G.: Improving nerf quality by progressive camera place- ment for free-viewpoint navigation. In: VMV (2023)

    2023

  20. [20]

    In: ECCV (2024)

    Kratimenos, A., Lei, J., Daniilidis, K.: Dynmf: Neural motion factorization for real-time dynamic view synthesis with 3d gaussian splatting. In: ECCV (2024)

    2024

  21. [21]

    In: NeurIPS (2024)

    Lee, J., Won, C., Jung, H., Bae, I., Jeon, H.G.: Fully explicit dynamic gaussian splatting. In: NeurIPS (2024)

    2024

  22. [22]

    In: CVPR (2025)

    Lei, J., Weng, Y., Harley, A., Guibas, L., Daniilidis, K.: Mosca: Dynamic gaussian fusion from casual videos via 4d motion scaffolds. In: CVPR (2025)

    2025

  23. [23]

    In: CVPR (2022)

    Li, T., Slavcheva, M., Zollhoefer, 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: CVPR (2022)

    2022

  24. [24]

    arXiv preprint arXiv:2511.23044 (2025)

    Li, Y., Cao, J., Ruan, P., Saxena, D., Zhu, S., Cao, Y.: Geometry-consistent 4d gaussian splatting for sparse-input dynamic view synthesis. arXiv preprint arXiv:2511.23044 (2025)

    arXiv 2025

  25. [25]

    IEEE RA-L10(8), 8099–8106 (2025)

    Li, Y., Kuang, Z., Li, T., Hao, Q., Yan, Z., Zhou, G., Zhang, S.: Activesplat: High- fidelity scene reconstruction through active gaussian splatting. IEEE RA-L10(8), 8099–8106 (2025)

    2025

  26. [26]

    In: CVPR (2024)

    Li, Z., Chen, Z., Li, Z., Xu, Y.: Spacetime gaussian feature splatting for real-time dynamic view synthesis. In: CVPR (2024)

    2024

  27. [27]

    In: CVPR (2021)

    Li, Z., Niklaus, S., Snavely, N., Wang, O.: Neural scene flow fields for space-time view synthesis of dynamic scenes. In: CVPR (2021)

    2021

  28. [28]

    In: CVPR (2023)

    Li, Z., Wang, Q., Cole, F., Tucker, R., Snavely, N.: Dynibar: Neural dynamic image- based rendering. In: CVPR (2023)

    2023

  29. [29]

    In: CVPR (2025)

    Liang, Y., Xu, T., Kikuchi, Y.: Himor: Monocular deformable gaussian reconstruc- tion with hierarchical motion representation. In: CVPR (2025)

    2025

  30. [30]

    In: CVPR (2024)

    Lin, Y., Dai, Z., Zhu, S., Yao, Y.: Gaussian-flow: 4d reconstruction with dynamic 3d gaussian particle. In: CVPR (2024)

    2024

  31. [31]

    In: CVPR (2024)

    Ling, L., Sheng, Y., Tu, Z., Zhao, W., Xin, C., Wan, K., Yu, L., Guo, Q., Yu, Z., Lu, Y., Li, X., Sun, X., Ashok, R., Mukherjee, A., Kang, H., Kong, X., Hua, G., Zhang, T., Benes, B., Bera, A.: Dl3dv-10k: A large-scale scene dataset for deep learning-based 3d vision. In: CVPR (2024)

    2024

  32. [32]

    arXiv preprint arXiv:2403.11396 (2024)

    Liu, G., Jiang, W., Lei, B., Pandey, V., Daniilidis, K., Motee, N.: Beyond uncer- tainty: Risk-aware active view acquisition for safe robot navigation and 3d scene understanding with fisherrf. arXiv preprint arXiv:2403.11396 (2024)

    Pith/arXiv arXiv 2024

  33. [33]

    In: ICLR (2025)

    Liu, Q., Liu, Y., Wang, J., Lyv, X., Wang, P., Wang, W., Hou, J.: Modgs: Dynamic gaussian splatting from casually-captured monocular videos with depth priors. In: ICLR (2025)

    2025

  34. [34]

    In: NeurIPS (2024) FillGS 17

    Liu, X., Zhou, C., Huang, S.: 3dgs-enhancer: Enhancing unbounded 3d gaussian splatting with view-consistent 2d diffusion priors. In: NeurIPS (2024) FillGS 17

    2024

  35. [35]

    In: CVPR (2024)

    Lu, Z., Guo, X., Hui, L., Chen, T., Yang, M., Tang, X., Zhu, F., Dai, Y.: 3d geometry-aware deformable gaussian splatting for dynamic view synthesis. In: CVPR (2024)

    2024

  36. [36]

    In: IROS (2024)

    Marza, P., Matignon, L., Simonin, O., Batra, D., Wolf, C., Chaplot, D.S.: Autonerf: Training implicit scene representations with autonomous agents. In: IROS (2024)

    2024

  37. [37]

    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

  38. [38]

    In: NeurIPS (2025)

    Nazarczuk, M., Catley-Chandar, S., Tanay, T., Zhang, Z., Slabaugh, G., Pérez- Pellitero, E.: Vidar: Video diffusion-aware 4d reconstruction from monocular in- puts. In: NeurIPS (2025)

    2025

  39. [39]

    TMLR (2024)

    Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., Assran, M., Ballas, N., Galuba, W.,Howes,R.,Huang,P.Y.,Li,S.W.,Misra,I.,Rabbat,M.,Sharma,V.,Synnaeve, G., Xu, H., Jegou, H., Mairal, J., Labatut, P., Joulin, A., Bojanowski, P.: Dinov2: Learning robust visual features without su...

    2024

  40. [40]

    In: ECCV (2022)

    Pan, X., Lai, Z., Song, S., Huang, G.: Activenerf: Learning where to see with uncertainty estimation. In: ECCV (2022)

    2022

  41. [41]

    In: ICCV (2021)

    Park, K., Sinha, U., Barron, J.T., Bouaziz, S., Goldman, D.B., Seitz, S.M., Martin- Brualla, R.: Nerfies: Deformable neural radiance fields. In: ICCV (2021)

    2021

  42. [42]

    ACM TOG40(6) (2021)

    Park,K.,Sinha,U.,Hedman,P.,Barron,J.T.,Bouaziz,S.,Goldman,D.B.,Martin- Brualla, R., Seitz, S.M.: Hypernerf: A higher-dimensional representation for topo- logically varying neural radiance fields. ACM TOG40(6) (2021)

    2021

  43. [43]

    In: CVPR (2016)

    Perazzi, F., Pont-Tuset, J., McWilliams, B., Gool, L.V., Gross, M., Sorkine- Hornung, A.: A benchmark dataset and evaluation methodology for video object segmentation. In: CVPR (2016)

    2016

  44. [44]

    In: CVPR (2021)

    Pumarola, A., Corona, E., Pons-Moll, G., Moreno-Noguer, F.: D-nerf: Neural ra- diance fields for dynamic scenes. In: CVPR (2021)

    2021

  45. [45]

    In: CVPRW (2017)

    Sabater, N., Boisson, G., Vandame, B., Kerbiriou, P., Babon, F., Hog, M., Gendrot, R., Langlois, T., Bureller, O., Schubert, A., Allié, V.: Dataset and pipeline for multi-view light-field video. In: CVPRW (2017)

    2017

  46. [46]

    In: WACV (2026)

    Safadoust, S., Tosi, F., Güney, F., Poggi, M.: Warprf: Multi-view consistency for training-free uncertainty quantification and applications in radiance fields. In: WACV (2026)

    2026

  47. [47]

    In: 3DV (2021)

    Shen,J.,Ruiz,A.,Agudo,A.,Moreno-Noguer,F.:Stochasticneuralradiancefields: Quantifying uncertainty in implicit 3d representations. In: 3DV (2021)

    2021

  48. [48]

    In: AAAI (2026)

    Shi, C., Yang, C., Hu, X., Chen, M., Pan, W., Yang, Y., Ding, J., Yu, Z., Yu, J.: Sparse4dgs: 4d gaussian splatting for sparse-frame dynamic scene reconstruction. In: AAAI (2026)

    2026

  49. [49]

    In: SIGGRAPH Asia (2024)

    Stearns, C., Harley, A., 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)

    2024

  50. [50]

    In: ICRA (2023)

    Sünderhauf, N., Abou-Chakra, J., Miller, D.: Density-aware nerf ensembles: Quan- tifying predictive uncertainty in neural radiance fields. In: ICRA (2023)

    2023

  51. [51]

    arXiv preprint arXiv:2105.05994 (2021)

    Wang, C., Eckart, B., Lucey, S., Gallo, O.: Neural trajectory fields for dynamic novel view synthesis. arXiv preprint arXiv:2105.05994 (2021)

    Pith/arXiv arXiv 2021

  52. [52]

    In: ICCV (2025)

    Wang, Q., Ye, V., Gao, H., Zeng, W., Austin, J., Li, Z., Kanazawa, A.: Shape of motion: 4d reconstruction from a single video. In: ICCV (2025)

    2025

  53. [53]

    In: AAAI (2025) 18 T

    Wang, S., Yang, X., Shen, Q., Jiang, Z., Wang, X.: Gflow: Recovering 4d world from monocular video. In: AAAI (2025) 18 T. Otonari and T. Yamasaki

    2025

  54. [54]

    TMLR (2025)

    Wang, Z., Zhou, H., Blaschko, M.B., Tuytelaars, T., Wu, M.: Diversity-driven view subset selection for indoor novel view synthesis. TMLR (2025)

    2025

  55. [55]

    IEEE TIP13(4), 600–612 (2004)

    Wang, Z., Bovik, A.C., Sheikh, H.R., Simoncelli, E.P.: Image quality assessment: from error visibility to structural similarity. IEEE TIP13(4), 600–612 (2004)

    2004

  56. [56]

    arXiv preprint arXiv:2506.19844 (2025)

    Wang, Z., Bhalgat, Y., Li, R., Prisacariu, V.A.: Active view selector: Fast and accurate active view selection with cross reference image quality assessment. arXiv preprint arXiv:2506.19844 (2025)

    Pith/arXiv arXiv 2025

  57. [57]

    In: 3DV (2026)

    Wei, J., Leutenegger, S., Schaefer, S.: Gsfix3d: Diffusion-guided repair of novel views in gaussian splatting. In: 3DV (2026)

    2026

  58. [58]

    In: CVPR (2025)

    Wilson, J., Almeida, M., Mahajan, S., Labrie, M., Ghaffari, M., Ghasemalizadeh, O., Sun, M., Kuo, C.H., Sen, A.: Pop-gs: Next best view in 3d-gaussian splatting with p-optimality. In: CVPR (2025)

    2025

  59. [59]

    In: CVPR (2024)

    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: CVPR (2024)

    2024

  60. [60]

    In: CVPR (2025)

    Wu, J.Z., Zhang, Y., Turki, H., Ren, X., Gao, J., Shou, M.Z., Fidler, S., Gojcic, Z., Ling, H.: Difix3d+: Improving 3d reconstructions with single-step diffusion models. In: CVPR (2025)

    2025

  61. [61]

    In: CVPR (2025)

    Wu, S., Xu, C., Huang, B., Geiger, A., Chen, A.: Genfusion: Closing the loop between reconstruction and generation via videos. In: CVPR (2025)

    2025

  62. [62]

    In: CVPR (2021)

    Xian, W., Huang, J.B., Kopf, J., Kim, C.: Space-time neural irradiance fields for free-viewpoint video. In: CVPR (2021)

    2021

  63. [63]

    In: CVPR (2024)

    Xiao, W., Cruz, R.S., Ahmedt-Aristizabal, D., Salvado, O., Fookes, C., Lebrat, L.: Nerf director: Revisiting view selection in neural volume rendering. In: CVPR (2024)

    2024

  64. [64]

    arXiv preprint arXiv:2508.02129 (2025)

    Xiao, Y., Lin, Z., Lu, C., Zhai, D., Jiang, K., Zhao, W., Zhang, W., Jiang, J., Wang, H., Liu, X.: Vdegaussian: Video diffusion enhanced 4d gaussian splatting for dynamic urban scenes modeling. arXiv preprint arXiv:2508.02129 (2025)

    Pith/arXiv arXiv 2025

  65. [65]

    In: CVPR (2024)

    Xu, Z., Peng, S., Lin, H., He, G., Sun, J., Shen, Y., Bao, H., Zhou, X.: 4k4d: Real-time 4d view synthesis at 4k resolution. In: CVPR (2024)

    2024

  66. [66]

    ACM TOG43(6) (2024)

    Xu, Z., Xu, Y., Yu, Z., Peng, S., Sun, J., Bao, H., Zhou, X.: Representing long volumetric video with temporal gaussian hierarchy. ACM TOG43(6) (2024)

    2024

  67. [67]

    In: ICLR (2024)

    Yang, Z., Yang, H., Pan, Z., Zhang, L.: Real-time photorealistic dynamic scene representation and rendering with 4d gaussian splatting. In: ICLR (2024)

    2024

  68. [68]

    In: CVPR (2024)

    Yang, Z., Gao, X., Zhou, W., Jiao, S., Zhang, Y., Jin, X.: Deformable 3d gaussians for high-fidelity monocular dynamic scene reconstruction. In: CVPR (2024)

    2024

  69. [69]

    In: ICML (2026)

    Yin, X., Zhang, Q., Chang, J., Feng, Y., Fan, Q., Yang, X., Pun, C.M., Zhang, H., Cun, X.: Gsfixer: Improving 3d gaussian splatting with reference-guided video diffusion priors. In: ICML (2026)

    2026

  70. [70]

    In: CVPR (2020)

    Yoon, J.S., Kim, K., Gallo, O., Park, H.S., Kautz, J.: Novel view synthesis of dynamic scenes with globally coherent depths from a monocular camera. In: CVPR (2020)

    2020

  71. [71]

    arXiv preprint arXiv:2211.12656 (2022)

    Zhan, H., Zheng, J., Xu, Y., Reid, I., Rezatofighi, H.: Activermap: Radiance field for active mapping and planning. arXiv preprint arXiv:2211.12656 (2022)

    Pith/arXiv arXiv 2022

  72. [72]

    Zhang, R., Isola, P., Efros, A.A., Shechtman, E., Wang, O.: The unreasonable effectiveness of deep features as a perceptual metric. In: CVPR (2018) FillGS 19 FillGS: Filling Observation Gaps in 4D Gaussian Splatting via Viewpoint-Time Selection and Generative Refinement Supplementary Material A Additional Implementation Details Video diffusion model.For t...

    2018