arxiv: 2604.04331 · v1 · submitted 2026-04-06 · 💻 cs.CV · cs.AI

Recognition: 2 theorem links

· Lean Theorem

GA-GS: Generation-Assisted Gaussian Splatting for Static Scene Reconstruction

Yedong Shen , Shiqi Zhang , Sha Zhang , Yifan Duan , Xinran Zhang , Wenhao Yu , Lu Zhang , Jiajun Deng

show 1 more author

Yanyong Zhang

Authors on Pith no claims yet

Pith reviewed 2026-05-10 20:00 UTC · model grok-4.3

classification 💻 cs.CV cs.AI

keywords static scene reconstructionGaussian splattingdiffusion modelocclusion handlingmonocular videodynamic objects3D reconstructioninpainting

0 comments

The pith

Generation-assisted Gaussian splatting recovers static scenes hidden by dynamic objects using diffusion inpainting

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

The paper establishes that static 3D scene reconstruction from monocular videos can be improved by assisting Gaussian Splatting with generated content for occluded areas. It segments dynamic objects, inpaints the background with a diffusion model to create supervision signals, and uses a learnable scalar to weight the authenticity of each Gaussian during optimization and rendering. This approach matters for applications like virtual reality and autonomous driving where complete static models are needed even when the video contains moving foreground elements. Experiments on the DAVIS dataset and a new Trajectory-Match dataset demonstrate better performance in handling large-scale persistent occlusions compared to previous methods.

Core claim

GA-GS reconstructs static scenes by removing dynamic objects via a motion-aware module, inpainting occluded regions with a diffusion model to provide pseudo-ground-truth, and modulating each Gaussian primitive's opacity with a learnable authenticity scalar to balance real and generated data in rendering and supervision.

What carries the argument

Learnable authenticity scalar for each Gaussian primitive that dynamically modulates opacity to enable authenticity-aware rendering and supervision.

Load-bearing premise

The regions inpainted by the diffusion model provide accurate enough pseudo-ground-truth to guide the reconstruction without introducing errors in the static scene model.

What would settle it

Quantitative metrics or visual artifacts in the reconstructed occluded regions on the Trajectory-Match dataset that exceed those of baseline methods without generation assistance would falsify the effectiveness of the pseudo-ground-truth supervision.

Figures

Figures reproduced from arXiv: 2604.04331 by Jiajun Deng, Lu Zhang, Sha Zhang, Shiqi Zhang, Wenhao Yu, Xinran Zhang, Yanyong Zhang, Yedong Shen, Yifan Duan.

**Figure 1.** Figure 1: Comparison between the previous pipeline (a) and our proposed GA-GS (b). Previous methods supervise 3D Gaussian primitives solely based on background regions after dynamic object removal. In contrast, our GAGS leverages a diffusion model to generate occluded content for auxiliary supervision and introduces authenticity-driven rendering to balance real and generated information. is critical for downstream … view at source ↗

**Figure 2.** Figure 2: An overview of our GA-GS pipeline. We use VGGT [13] to obtain accurate camera poses and per-pixel 3D positions, Then we employ a motion-aware SAM-based module to segment moving regions, and use a diffusion model to inpaint occlusions, providing pseudo-ground-truth supervision. In the opacity blending stage, the parameter θ is used to control the opacity of each Gaussian primitive, and the image space mask … view at source ↗

**Figure 3.** Figure 3: Visualization on the DAVIS dataset. Since ground truth for the static background is unavailable, the first row only shows the input containing dynamic objects. Compared to the baselines, our method achieves better visual results in both background reconstruction and occlusion region recovery. rendering, while those from inpainted areas serve as auxiliary support. Specifically, for a given Gaussian primitiv… view at source ↗

**Figure 4.** Figure 4: Data acquisition process of the Trajectory-Match dataset. (a) We employ a robot-mounted camera platform to ensure precise and repeatable camera trajectories. (b) For each scene, we first capture a dynamic sequence containing moving objects such as pedestrians or vehicles. (c) We then record a corresponding static sequence along the same trajectory after removing dynamic elements, which serves as ground-tru… view at source ↗

**Figure 5.** Figure 5: Visualizations on Trajectory-Match Dataset. The second row presents the ground truth of recorded static scene, serving as a reference for comparing the reconstructions of GA-GS and the baseline in the third and fourth rows. level accuracy, SSIM (Structural Similarity Index) for structural consistency, and LPIPS (Learned Perceptual Image Patch Similarity) for perceptual similarity based on deep features. I… view at source ↗

read the original abstract

Reconstructing static 3D scene from monocular video with dynamic objects is important for numerous applications such as virtual reality and autonomous driving. Current approaches typically rely on background for static scene reconstruction, limiting the ability to recover regions occluded by dynamic objects. In this paper, we propose GA-GS, a Generation-Assisted Gaussian Splatting method for Static Scene Reconstruction. The key innovation of our work lies in leveraging generation to assist in reconstructing occluded regions. We employ a motion-aware module to segment and remove dynamic regions, and thenuse a diffusion model to inpaint the occluded areas, providing pseudo-ground-truth supervision. To balance contributions from real background and generated region, we introduce a learnable authenticity scalar for each Gaussian primitive, which dynamically modulates opacity during splatting for authenticity-aware rendering and supervision. Since no existing dataset provides ground-truth static scene of video with dynamic objects, we construct a dataset named Trajectory-Match, using a fixed-path robot to record each scene with/without dynamic objects, enabling quantitative evaluation in reconstruction of occluded regions. Extensive experiments on both the DAVIS and our dataset show that GA-GS achieves state-of-the-art performance in static scene reconstruction, especially in challenging scenarios with large-scale, persistent occlusions.

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.

Desk Editor's Note private letter to a colleague

GA-GS adds diffusion inpainting and per-Gaussian authenticity scalars to handle persistent occlusions in static reconstruction, but the abstract gives no numbers or checks to support the SOTA claim.

read the letter

The main thing to know is that GA-GS removes dynamic objects with a motion-aware step, uses a diffusion model to fill in the occluded background as pseudo-ground-truth, and adds a learnable authenticity scalar per Gaussian primitive to modulate how much the generated content contributes during rendering and loss. They also built the Trajectory-Match dataset with paired robot trajectories to allow quantitative checks on the hidden regions. This setup directly targets the gap where standard Gaussian splatting or NeRF methods either leave holes or assume clean background footage. The combination of inpainting for supervision and the scalar for balancing real versus generated data is not a trivial add-on, and the new dataset lets them measure something previous work could not. The approach is practical for VR and driving scenes where occlusions are common and long-lasting. The soft spot is the complete absence of any metrics, ablations, or validation that the inpainted regions actually match the true static geometry. Diffusion models frequently produce plausible but geometrically wrong content in large occluded areas, and a jointly learned scalar could easily over-trust bad inpaintings or under-use real observations. Without those checks the central claim stays unverified. This paper is for people working on monocular video to 3D reconstruction who need to deal with moving foregrounds. It deserves a serious referee because the pipeline and dataset are fresh enough to warrant review, even if the current version needs substantial added evidence on inpainting accuracy and quantitative results. I would send it to peer review with a request for those details.

Referee Report

2 major / 2 minor

Summary. The paper introduces GA-GS, a Generation-Assisted Gaussian Splatting method for reconstructing static 3D scenes from monocular videos containing dynamic objects. It employs a motion-aware module to segment and remove dynamic regions, uses a diffusion model to inpaint occluded areas as pseudo-ground-truth supervision, and introduces a learnable authenticity scalar per Gaussian primitive to dynamically modulate opacity and balance real versus generated content during rendering and loss computation. The authors construct the Trajectory-Match dataset (fixed-path robot recordings with/without dynamic objects) to enable quantitative evaluation of occluded-region reconstruction and report state-of-the-art performance on DAVIS and their dataset, especially under large-scale persistent occlusions.

Significance. If the inpainted pseudo-ground-truth proves geometrically and semantically accurate and the authenticity scalar reliably suppresses artifacts, the approach could meaningfully advance static scene reconstruction in occluded settings relevant to VR and autonomous driving. The Trajectory-Match dataset is a clear strength, providing the first verifiable quantitative benchmark for occluded-region fidelity; this addresses a long-standing evaluation gap. The generation-assisted supervision idea is a promising direction worth further exploration.

major comments (2)

[Method (inpainting and pseudo-ground-truth supervision) and Experiments] The central claim that diffusion-based inpainting supplies reliable pseudo-ground-truth for large occluded regions is load-bearing yet unsupported by direct validation. The Trajectory-Match dataset records both occluded and clean static views, but no quantitative metrics (e.g., PSNR, depth error, or geometric consistency) are reported comparing the inpainted content against the true background geometry in those regions. Without this check, it remains unclear whether the supervision signal is accurate enough to support the SOTA claim in persistent-occlusion scenarios.
[Method (authenticity scalar) and Experiments] No ablation or analysis is provided for the learnable authenticity scalar. The paper states that the scalar modulates opacity for authenticity-aware rendering and supervision, yet there are no results showing its learned values, its effect on artifact suppression, or performance drop when it is removed or fixed. This is critical because the scalar is the only mechanism claimed to prevent erroneous generated content from corrupting the Gaussian optimization.

minor comments (2)

[Abstract] Abstract contains a typo: 'thenuse' should read 'then use'.
[Method] The paper would benefit from explicit equations or pseudocode for the authenticity scalar's integration into the splatting and loss terms to clarify its exact functional form.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and positive remarks on the Trajectory-Match dataset and the overall direction of the work. We address each major comment below and will revise the manuscript to incorporate the suggested analyses.

read point-by-point responses

Referee: The central claim that diffusion-based inpainting supplies reliable pseudo-ground-truth for large occluded regions is load-bearing yet unsupported by direct validation. The Trajectory-Match dataset records both occluded and clean static views, but no quantitative metrics (e.g., PSNR, depth error, or geometric consistency) are reported comparing the inpainted content against the true background geometry in those regions. Without this check, it remains unclear whether the supervision signal is accurate enough to support the SOTA claim in persistent-occlusion scenarios.

Authors: We agree that direct quantitative validation of the inpainted regions against the clean ground-truth views available in Trajectory-Match would strengthen the paper. While the end-to-end improvements in occluded-region reconstruction metrics on this dataset provide indirect support for the quality of the pseudo-ground-truth, we did not report explicit comparisons (e.g., PSNR, SSIM, or depth consistency) between the diffusion-inpainted content and the true static background. In the revised manuscript we will add these metrics, computed on the held-out clean views, together with qualitative comparisons, to directly assess inpainting fidelity. revision: yes
Referee: No ablation or analysis is provided for the learnable authenticity scalar. The paper states that the scalar modulates opacity for authenticity-aware rendering and supervision, yet there are no results showing its learned values, its effect on artifact suppression, or performance drop when it is removed or fixed. This is critical because the scalar is the only mechanism claimed to prevent erroneous generated content from corrupting the Gaussian optimization.

Authors: We acknowledge the absence of dedicated analysis for the authenticity scalar. In the revision we will add: (i) visualizations of the learned scalar values across Gaussians in occluded versus visible regions, (ii) quantitative ablations showing performance when the scalar is removed or fixed to constant values, and (iii) qualitative results illustrating its role in suppressing artifacts from generated content. These experiments will demonstrate the scalar's contribution to balancing real and pseudo-ground-truth supervision during optimization. revision: yes

Circularity Check

0 steps flagged

No circularity; method integrates external diffusion inpainting and new dataset without self-referential reductions

full rationale

The paper describes a pipeline that segments dynamic objects via a motion-aware module, uses an off-the-shelf diffusion model to generate inpainted pseudo-ground-truth for occluded regions, and introduces a learnable per-Gaussian authenticity scalar to modulate rendering and loss. No equations, derivations, or fitted parameters are presented as 'predictions' that reduce to the inputs by construction. The Trajectory-Match dataset is newly collected for quantitative evaluation rather than being used to fit the core claims. No load-bearing self-citations or uniqueness theorems from prior author work are invoked to justify the architecture. The central claims rest on empirical performance against external benchmarks and the assumption that the diffusion model provides usable supervision, which is an external dependency rather than a circular reduction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on the assumption that diffusion inpainting yields usable pseudo-ground-truth and that a per-Gaussian scalar can reliably separate real from generated content; no free parameters are explicitly fitted in the abstract description beyond the learnable scalar itself.

free parameters (1)

authenticity scalar
Learnable per-Gaussian value introduced to modulate opacity and balance real versus inpainted regions during rendering and supervision.

axioms (1)

domain assumption Diffusion models can generate inpainted background regions sufficiently accurate to serve as pseudo-ground-truth for occluded areas.
Invoked when the method uses the diffusion output to supervise reconstruction of regions hidden by dynamic objects.

invented entities (1)

authenticity scalar no independent evidence
purpose: Dynamically adjusts each Gaussian primitive's contribution to distinguish real background from generated inpainted content.
New per-primitive parameter not present in standard Gaussian splatting formulations.

pith-pipeline@v0.9.0 · 5541 in / 1451 out tokens · 64653 ms · 2026-05-10T20:00:05.855384+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

we introduce a learnable authenticity scalar for each Gaussian primitive, which dynamically modulates opacity during splatting for authenticity-aware rendering and supervision
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

GA-GS achieves state-of-the-art performance in static scene reconstruction, especially in challenging scenarios with large-scale, persistent occlusions

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

33 extracted references · 7 canonical work pages

[1]

Nerf: Representing scenes as neural radiance fields for view synthesis,

B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,”Communications of the ACM, vol. 65, no. 1, pp. 99–106, 2021

2021
[2]

3d gaussian splatting for real-time radiance field rendering

B. Kerbl, G. Kopanas, T. Leimk ¨uhler, and G. Drettakis, “3d gaussian splatting for real-time radiance field rendering.”ACM Trans. Graph., vol. 42, no. 4, pp. 139–1, 2023

2023
[3]

Spatialsplat: Efficient Se- mantic 3d from Sparse Unposed Images.arXiv preprint arXiv:2505.23044, 2025

Y . Sheng, J. Deng, X. Zhang, Y . Zhang, B. Hua, Y . Zhang, and J. Ji, “Spatialsplat: Efficient semantic 3d from sparse unposed images,”arXiv preprint arXiv:2505.23044, 2025

work page arXiv 2025
[4]

Vr-splatting: Foveated radiance field rendering via 3d gaussian splatting and neural points,

L. Franke, L. Fink, and M. Stamminger, “Vr-splatting: Foveated radiance field rendering via 3d gaussian splatting and neural points,”Proceedings of the ACM on Computer Graphics and Interactive Techniques, vol. 8, no. 1, pp. 1–21, 2025

2025
[5]

Taming 3dgs: High-quality radiance fields with limited resources,

S. S. Mallick, R. Goel, B. Kerbl, M. Steinberger, F. V . Carrasco, and F. De La Torre, “Taming 3dgs: High-quality radiance fields with limited resources,” inSIGGRAPH Asia 2024 Conference Papers, 2024, pp. 1– 11

2024
[6]

Hybrid motion representation learning for prediction from raw sensor data,

D. Meng, C. Yu, J. Deng, D. Qian, H. Li, and D. Ren, “Hybrid motion representation learning for prediction from raw sensor data,” IEEE Transactions on Multimedia, vol. 25, pp. 8868–8879, 2023

2023
[7]

Gaussnav: Gaussian splatting for visual navigation,

X. Lei, M. Wang, W. Zhou, and H. Li, “Gaussnav: Gaussian splatting for visual navigation,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

2025
[8]

Flashgs: Efficient 3d gaussian splatting for large-scale and high-resolution rendering,

G. Feng, S. Chen, R. Fu, Z. Liao, Y . Wang, T. Liu, B. Hu, L. Xu, Z. Pei, H. Liet al., “Flashgs: Efficient 3d gaussian splatting for large-scale and high-resolution rendering,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 26 652–26 662

2025
[9]

Gaussianeditor: Editing 3d gaussians delicately with text instructions,

J. Wang, J. Fang, X. Zhang, L. Xie, and Q. Tian, “Gaussianeditor: Editing 3d gaussians delicately with text instructions,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 20 902–20 911

2024
[10]

Infusion: Inpainting 3d gaussians via learning depth completion from diffusion prior.arXiv preprint arXiv:2404.11613, 2024

Z. Liu, H. Ouyang, Q. Wang, K. L. Cheng, J. Xiao, K. Zhu, N. Xue, Y . Liu, Y . Shen, and Y . Cao, “Infusion: Inpainting 3d gaussians via learning depth completion from diffusion prior,”arXiv preprint arXiv:2404.11613, 2024

work page arXiv 2024
[11]

Wildgaussians: 3d gaussian splatting in the wild.arXiv preprint arXiv:2407.08447, 2024

J. Kulhanek, S. Peng, Z. Kukelova, M. Pollefeys, and T. Sattler, “Wildgaussians: 3d gaussian splatting in the wild,”arXiv preprint arXiv:2407.08447, 2024

work page arXiv 2024
[12]

Robust 3d gaussian splatting for novel view synthesis in presence of distractors,

P. Ungermann, A. Ettenhofer, M. Nießner, and B. Roessle, “Robust 3d gaussian splatting for novel view synthesis in presence of distractors,” inDAGM German Conference on Pattern Recognition. Springer, 2024, pp. 153–167

2024
[13]

Vggt: Visual geometry grounded transformer,

J. Wang, M. Chen, N. Karaev, A. Vedaldi, C. Rupprecht, and D. Novotny, “Vggt: Visual geometry grounded transformer,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 5294– 5306

2025
[14]

Segment anything,

A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Loet al., “Segment anything,” inProceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 4015–4026

2023
[15]

Gs-sfs: Joint gaussian splatting and shape-from-silhouette for multiple human reconstruction in large-scale sports scenes,

Y . Jiang, J. Li, H. Qin, Y . Dai, J. Liu, G. Zhang, C. Zhang, and T. Yang, “Gs-sfs: Joint gaussian splatting and shape-from-silhouette for multiple human reconstruction in large-scale sports scenes,”IEEE Transactions on Multimedia, vol. 26, pp. 11 095–11 110, 2024

2024
[16]

Bugs: Universal 3d gaussian splatting with a bi-directional gaussian growing mechanism,

F. Duan, Y . Zhang, X. Li, X. Tan, J. Wang, and L. Chen, “Bugs: Universal 3d gaussian splatting with a bi-directional gaussian growing mechanism,”IEEE Transactions on Multimedia, 2026

2026
[17]

4d gaussian splatting for real-time dynamic scene rendering,

G. Wu, T. Yi, J. Fang, L. Xie, X. Zhang, W. Wei, W. Liu, Q. Tian, and X. Wang, “4d gaussian splatting for real-time dynamic scene rendering,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 20 310–20 320

2024
[18]

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

Y . Duan, F. Wei, Q. Dai, Y . He, W. Chen, and B. Chen, “4d-rotor gaussian splatting: towards efficient novel view synthesis for dynamic scenes,” inACM SIGGRAPH 2024 Conference Papers, 2024, pp. 1–11

2024
[19]

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

Y . Lin, Z. Dai, S. Zhu, and Y . Yao, “Gaussian-flow: 4d reconstruction with dynamic 3d gaussian particle,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2024, pp. 21 136–21 145

2024
[20]

Street gaussians: Modeling dynamic urban scenes with gaussian splatting,

Y . Yan, H. Lin, C. Zhou, W. Wang, H. Sun, K. Zhan, X. Lang, X. Zhou, and S. Peng, “Street gaussians: Modeling dynamic urban scenes with gaussian splatting,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 156–173

2024
[21]

Og-gaussian: Occupancy based street gaussians for autonomous driv- ing,

Y . Shen, X. Zhang, Y . Duan, S. Zhang, H. Li, Y . Wu, J. Ji, and Y . Zhang, “Og-gaussian: Occupancy based street gaussians for autonomous driv- ing,”arXiv preprint arXiv:2502.14235, 2025

work page arXiv 2025
[22]

S3r-gs: Streamlining the pipeline for large-scale street scene reconstruction,

G. Zheng, J. Deng, X. Chu, Y . Yuan, H. Li, and Y . Zhang, “S3r-gs: Streamlining the pipeline for large-scale street scene reconstruction,” arXiv preprint arXiv:2503.08217, 2025

work page arXiv 2025
[23]

Robust dynamic radiance fields,

Y .-L. Liu, C. Gao, A. Meuleman, H.-Y . Tseng, A. Saraf, C. Kim, Y .-Y . Chuang, J. Kopf, and J.-B. Huang, “Robust dynamic radiance fields,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 13–23

2023
[24]

Nerf on-the-go: Exploiting uncertainty for distractor-free nerfs in the wild,

W. Ren, Z. Zhu, B. Sun, J. Chen, M. Pollefeys, and S. Peng, “Nerf on-the-go: Exploiting uncertainty for distractor-free nerfs in the wild,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 8931–8940

2024
[25]

Hybridgs: Decoupling transients and statics with 2d and 3d gaussian splatting,

J. Lin, J. Gu, L. Fan, B. Wu, Y . Lou, R. Chen, L. Liu, and J. Ye, “Hybridgs: Decoupling transients and statics with 2d and 3d gaussian splatting,” inProceedings of the Computer Vision and Pattern Recogni- tion Conference, 2025, pp. 788–797

2025
[26]

Emerging properties in self-supervised vision transformers,

M. Caron, H. Touvron, I. Misra, H. J ´egou, J. Mairal, P. Bojanowski, and A. Joulin, “Emerging properties in self-supervised vision transformers,” inProceedings of the International Conference on Computer Vision (ICCV), 2021

2021
[27]

Spotlesssplats: Ignoring distractors in 3d gaussian splatting,

S. Sabour, L. Goli, G. Kopanas, M. Matthews, D. Lagun, L. Guibas, A. Jacobson, D. Fleet, and A. Tagliasacchi, “Spotlesssplats: Ignoring distractors in 3d gaussian splatting,”ACM Transactions on Graphics, vol. 44, no. 2, pp. 1–11, 2025

2025
[28]

Das3r: Dynamics-aware gaussian splatting for static scene reconstruction

K. Xu, T. H. E. Tse, J. Peng, and A. Yao, “Das3r: Dynamics- aware gaussian splatting for static scene reconstruction,”arXiv preprint arXiv:2412.19584, 2024

work page arXiv 2024
[29]

Structure-from-motion revisited,

J. L. Schonberger and J.-M. Frahm, “Structure-from-motion revisited,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 4104–4113

2016
[30]

Moving object segmen- tation: All you need is sam (and flow),

J. Xie, C. Yang, W. Xie, and A. Zisserman, “Moving object segmen- tation: All you need is sam (and flow),” inProceedings of the Asian conference on computer vision, 2024, pp. 162–178

2024
[31]

Raft: Recurrent all-pairs field transforms for optical flow,

Z. Teed and J. Deng, “Raft: Recurrent all-pairs field transforms for optical flow,” inEuropean conference on computer vision. Springer, 2020, pp. 402–419

2020
[32]

Diffueraser: A diffusion model for video inpainting,

X. Li, H. Xue, P. Ren, and L. Bo, “Diffueraser: A diffusion model for video inpainting,”arXiv preprint arXiv:2501.10018, 2025

work page arXiv 2025
[33]

A benchmark dataset and evaluation methodol- ogy for video object segmentation,

F. Perazzi, J. Pont-Tuset, B. McWilliams, L. Van Gool, M. Gross, and A. Sorkine-Hornung, “A benchmark dataset and evaluation methodol- ogy for video object segmentation,” inComputer Vision and Pattern Recognition, 2016

2016