Pith. sign in

REVIEW 3 major objections 5 minor 47 references

AniGS animates a static 3D scene reconstruction into a free-viewpoint video with natural ambient motion, while keeping rigid structures still.

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

2026-08-01 15:05 UTC pith:DNYSPBNX

load-bearing objection Genuine systems contribution for scene-level 3DGS animation, but the headline FVD metric is partly circular; the qualitative and multiview-consistency evidence carries the load. the 3 major comments →

arxiv 2607.18539 v1 pith:DNYSPBNX submitted 2026-07-20 cs.CV

AniGS: Bridging Rendering and Diffusion Prior for 3D Scene Animation

Yen-Chi Cheng , Chen Gao , Chuhan Chen , Tuotuo Li , Rajvi Shah , Ayush Saraf , Changil Kim , Liangyan Gui
show 3 more authors
Alexander Schwing Johannes Kopf Hung-Yu Tseng
This is my paper
classification cs.CV
keywords scene animation3D Gaussian splattingvideo diffusion modelambient dynamicsdeformation fieldnovel view synthesisdataset-model updatefree-viewpoint video
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 tries to establish that a static 3D Gaussian Splatting reconstruction of a large outdoor scene can be turned into a dynamic, free-viewpoint video with plausible ambient motion (vegetation swaying, leaves moving) without any ground-truth dynamic supervision. It does so by coupling the static representation with a time-conditioned deformation field and a pretrained video diffusion model that supplies motion cues, using an iterative dataset–model update loop to bootstrap temporally consistent multi-view supervision across the whole scene. The central claim is that this yields natural, scene-wide dynamics while rigid structures remain stable, supported by FVD scores, a multiview-consistency reconstruction check, and a user study on five real-world scenes.

Core claim

The central discovery is that scene-level animation of 3DGS reconstructions can be bootstrapped from a static canonical model plus a video diffusion prior, via an iterative process: render camera-fixed clips from the current model, refine them with a composed video-to-video diffusion step that freezes static regions, add these as training targets, then optimize a DCT-parameterized deformation field over the canonical Gaussians. The key insight is that by iterating dataset updates and model updates while expanding viewpoints, the method avoids the need for synchronized multi-camera dynamic capture, producing temporally coherent ambient motion across a large navigable scene.

What carries the argument

The time-conditioned deformation field: each Gaussian's position and rotation are offset by a discrete cosine transform (DCT) basis expansion of time, which is storage-efficient and smooth. The iterative dataset–model update loop with incremental viewpoint expansion and composed video-to-video refinement (ComposedV2V) is the mechanism that generates multi-view consistent pseudo-targets; ComposedV2V pastes the static region from the first frame into all frames to prevent drift in static areas.

Load-bearing premise

The load-bearing assumption is that FVD computed against videos generated by pretrained video diffusion models (especially LTX, which also generates the training targets) is a valid measure of motion realism; if that proxy does not reflect genuine scene dynamics, the central quantitative evidence loses its footing.

What would settle it

A concrete test: record or obtain real dynamic video of one of the same outdoor scenes (real wind-driven foliage) with a synchronized multi-camera rig, render AniGS's animation from matching viewpoints, and run a two-alternative forced choice comparing AniGS's motion against the real captured motion, or compute FVD against this independent reference. If AniGS is not preferred or its FVD is not lower than a static baseline, the claim of natural ambient dynamics would be falsified.

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

If this is right

  • If correct, reconstructed 3D environments from casual captures can be rendered as animated free-viewpoint videos, making VR/AR walkthroughs more immersive.
  • The method establishes a template for lifting video diffusion priors into 3D scene-level motion without per-scene dynamic capture, potentially extensible to other generative priors.
  • The iterative dataset–model update could serve as a general recipe for distilling 2D generative models into consistent 3D representations.
  • The composed refinement shows a way to constrain generative video models to move only desired regions, which may transfer to video editing tasks.
  • The multiview consistency of generated motion suggests the deformation field learns a true 3D motion, enabling novel-view synthesis of the animation.

Where Pith is reading between the lines

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

  • A testable extension: applying the same loop to indoor scenes with different animatable objects (the paper shows one such result) could verify whether the method's reliance on the diffusion model's prior limits it to vegetation-like oscillatory motion.
  • The FVD-LTX metric is partly self-referential because LTX is also the teacher; an independent evaluation against real captured dynamic footage or a blind comparison with a held-out scene would strengthen or weaken the claim.
  • The method may inherit biases from the video diffusion model (e.g., a tendency for certain motion patterns or a preference for particular lighting), and the iterative updates could amplify these biases; this is an open question not addressed in the paper.
  • The deformation field with DCT basis limits the motion to reversible, small deformations, so the claim naturally stops short of state changes; extending to non-reversible dynamics would require a different representation.

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

3 major / 5 minor

Summary. The paper proposes AniGS, a method for animating static 3D Gaussian Splatting reconstructions of large, cluttered outdoor scenes. The method combines a canonical 3DGS with a time-conditioned deformation field parameterized by DCT coefficients, and uses a pretrained flow-based video diffusion model (LTX) as a motion prior. Training is bootstrapped through an iterative dataset–model update: a dataset of camera-fixed rendered clips is progressively expanded and refined by a composed video-to-video (ComposedV2V) scheme that keeps static regions fixed, and the model is optimized against these pseudo-targets with a photometric loss plus canonical regularization and an SDS loss. Experiments on five collected scenes and DL3DV scenes compare against Gaussians2Life and PhysDreamer using FVD against diffusion-generated references, a multiview-consistency PSNR check, and a 2AFC user study. The authors claim natural ambient dynamics and high-quality novel-view videos at scene scale without ground-truth dynamic supervision.

Significance. If validated, AniGS addresses a real gap: adding plausible ambient motion to large, navigable 3D scene reconstructions without synchronized multi-camera dynamic capture. The proposed iterative viewpoint expansion and ComposedV2V static-preservation mechanism are sensible and potentially useful to the 4D-scene community. The multiview-consistency PSNR experiment (Table 3) is a good positive check, and the qualitative space-time slices illustrate the intended behavior. However, the quantitative evaluation currently relies on Fréchet Video Distance computed against reference videos generated by the same or similar video diffusion models. Since LTX is both the teacher that produces the pseudo-training targets (Sec. 3.4/3.5) and the generator of the reference distribution for FVD-LTX (Sec. 4.3), the primary numerical evidence is circular and does not establish that the synthesized motion is natural or realistic relative to real-world dynamics. This is a load-bearing weakness that requires a non-circular evaluation before the central claim can be accepted.

major comments (3)
  1. [Sec. 4.3; Sec. 3.4/3.6] FVD-LTX in Tables 1, 2, and 5 is a self-comparison. The pseudo-training targets are produced by LTX via ComposedV2V (Eqs. 9–12 and Sec. 3.6), and the reference distribution for FVD-LTX is also generated by LTX (Sec. 4.3). A low score therefore measures how well the model imitates its own teacher distribution, not how natural the motion is with respect to real scenes. Because the abstract's central claim of 'natural ambient dynamics' is supported primarily by these FVD numbers, an independent reference—for example real captured dynamic video, a non-teacher video model, or a metric grounded in human judgment—is needed. The ablations in Table 5 inherit this issue and cannot validate the design choices.
  2. [Sec. 4.3, Tables 1–2] FVD-Cosmos and FVD-DynamiCrafter are also computed against reference videos generated by pretrained video diffusion models. While these models are not the teacher for AniGS, they are still generative priors rather than ground-truth observations of the actual scenes. The paper acknowledges that ground-truth scene-level dynamic videos are unavailable (Sec. 4.3), but that does not justify treating diffusion-generated distributions as a gold standard for 'realistic motion'. The metric can reward matching a generic generator's motion prior while failing to capture the true dynamics of the scene. A concrete test would be to evaluate on a real dynamic outdoor capture (even with a single moving camera or limited multi-camera data) or to report human agreement against real footage.
  3. [Sec. 4.5, Table 4] The user study reports preference percentages (e.g., 91.7% in Fireplace) but does not report the number of participants, the number of comparison pairs, the variance across participants, or any statistical significance test. Without this information, the percentages cannot be distinguished from noise, especially when the numbers are close for some scenes (e.g., 74.5% in Trail). Please report sample sizes, confidence intervals, and a significance test (e.g., paired bootstrap or Wilcoxon).
minor comments (5)
  1. [Sec. 1] Typo: 'modeling the the synthesized motion' should read 'modeling the synthesized motion'.
  2. [Sec. 3.6] Hyperparameters such as DCT basis size K, loss weights λ_cano and λ_SDS, and the SDS clip-length parameter m are not specified. Please provide these values for reproducibility.
  3. [Sec. 4.1, Table 3] The 'Original capture' PSNR row includes DL3DV scenes (Gazebo, Courtyard) but the caption does not clarify whether these are from the same capture protocol as the collected scenes. Please specify the source and processing for each scene.
  4. [General] The baseline name is inconsistently spelled as 'Gaussians2Life' and 'Gaussians2life' in Tables 1 and 4 and in the text. Please unify.
  5. [References] Reference [38] contains an editorial note ('Use this entry for both ... bullets'), which should be removed.

Circularity Check

1 steps flagged

FVD-LTX is a self-comparison: the model is trained to match LTX-refined targets and then evaluated against LTX-generated references; the 'natural dynamics' claim is only partially independent via Cosmos/DynamiCrafter and a user study.

specific steps
  1. fitted input called prediction [Sec. 3.4 (Eqs. 8-12), Sec. 3.5 (Eqs. 13, 17), Sec. 3.6, Sec. 4.3]
    "we use LTX-Video [4] as the pretrained video diffusion model... training supervises rendered videos with LTX-refined targets via L_ani (Eq. 13) and LTX SDS (Eq. 17); Sec. 4.3: 'we measure FVD against a reference video distribution generated by pretrained video diffusion models... LTX [4], Cosmos [1], and DynamiCrafter [35]'."

    AniGS's pseudo-training targets A (Eqs. 9/12) are produced by LTX, and the deformation field is optimized to match them (Eq. 13) and to follow LTX's score (Eq. 17). FVD-LTX then compares rendered videos to a reference distribution also generated by LTX. A low FVD-LTX is therefore evidence that the renderer mimics its own teacher, not that the motion matches real scene dynamics; the metric is aligned with the training objective by construction. FVD-Cosmos and FVD-DynamiCrafter are less circular, but they are still generative references, not captured scene dynamics.

full rationale

The core derivation is not internally circular: AniGS explicitly bootstraps pseudo-supervision from a pretrained video diffusion prior, so the claim is essentially that the renderable representation can distill LTX's motion prior into a multi-view consistent 3D deformation field. The render-and-refine loop and the SDS loss are standard self-distillation mechanisms, and the method does not present its training targets as ground-truth measurements. The main circularity is in evaluation: the paper's headline FVD-LTX compares the model's output to a reference distribution generated by the same LTX model that produced the training pseudo-targets. Lower FVD-LTX is thus partly a self-consistency result, not independent evidence of natural dynamics. The paper mitigates this by also reporting FVD-Cosmos and FVD-DynamiCrafter, which are not the training teacher, and by including a user study; these provide some independent support, although all references are synthetic rather than real captured dynamics. The lack of participant counts and significance tests in the user study is a statistical reporting weakness, not a circularity per se. Overall, the central claim is partially supported by independent evaluation, so the circularity score is moderate rather than severe.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 0 invented entities

AniGS's contribution is a training/supervision recipe; the load-bearing assumptions are that a video generator's synthetic motion is a good teacher, that SAM2 masks localize it, that per-view refinements cohere into multi-view supervision, and that diffusion-generated reference videos can serve as ground truth for evaluation. No new physical entities are introduced.

free parameters (6)
  • DCT basis size K
    Controls the smoothness of the deformation field in Eq. 4; no value is reported, so the temporal behavior cannot be exactly reproduced.
  • Loss weights λ_cano and λ_SDS
    Balance the canonical reconstruction loss and the SDS loss in L = L_ani + λ_cano L_cano + λ_SDS L_SDS (Sec 3.5); values not given.
  • Number of camera locations N = 4
    N=4 camera locations used for the iterative dataset update (Sec 3.6); scene coverage depends on this hand-set choice.
  • SDS clip length parameter m
    Renders clips of 4m+1 frames for the SDS loss (Sec 3.5); m is never specified.
  • Training schedule = 70k iterations; update every 2500 steps
    The dataset update interval and total iterations are hand-set; no sensitivity analysis is provided.
  • Diffusion sampling settings = 8 steps, 121 frames, 1024×1408, FOV 95°
    Inference-time parameters for the diffusion prior (Sec 3.6); no ablation on these choices.
axioms (6)
  • domain assumption The pretrained LTX video diffusion model provides a reliable motion prior for scene-level animation.
    Sec 3.2/3.4 use LTX to generate pseudo-target videos; if its imagined motion is not plausible or multi-view consistent, the training signal is wrong.
  • domain assumption SAM2's first-frame segmentation correctly separates dynamic (foliage) from static regions.
    Eq. 8 composes static pixels using mask M0; a wrong mask either locks motion in dynamic areas or allows drift in static areas.
  • domain assumption ComposedV2V outputs from different viewpoints are temporally consistent enough to jointly supervise a single deformation field.
    Sec 3.4 builds the training dataset from per-view diffusion refinements; the paper itself notes naive per-view clips are inconsistent, so the composition is assumed to fix this.
  • ad hoc to paper FVD against diffusion-generated reference videos is a valid proxy for motion realism.
    Sec 4.3 introduces this metric because ground-truth dynamic videos are unavailable; the proxy is not validated against real captures.
  • domain assumption The iterative dataset–model update converges to full scene coverage without drift.
    Sec 3.4–3.5 alternates dataset expansion and model fitting; no convergence guarantee or failure analysis is given beyond ablations.
  • domain assumption DCT-basis deformation can represent the target ambient motion.
    Eq. 4 models scalar deformation as a DCT series; this borrows AmbientGS [25] and presumes the motion is smooth and low-frequency.

pith-pipeline@v1.3.0-alltime-deepseek · 14927 in / 12316 out tokens · 121533 ms · 2026-08-01T15:05:29.545732+00:00 · methodology

0 comments
read the original abstract

Novel view rendering of large and complex reconstructed scenes is becoming increasingly photorealistic. However, most reconstructions remain static and lack the ambient motion that makes environments immersive. We present AniGS, a method for scene-level animation of 3D Gaussian Splatting (3DGS) reconstructions that adds subtle, distributed dynamics, e.g., vegetation motion, while preserving rigid structures. Unlike existing 3D animation techniques which are limited to object-centric subjects or small regions, AniGS is designed for large, cluttered, navigable scenes. AniGS represents the scene with a canonical 3DGS and models motion using a time-conditioned deformation field. To animate the entire scene, we leverage a pretrained video diffusion model and introduce an iterative dataset--model update strategy that progressively expands viewpoint coverage and repeatedly updates camera-fixed training videos using a render-and-refine scheme. To prevent artifacts from unintended motion in static areas, we further introduce a composed video-to-video refinement scheme that restricts motion to desired regions. Experiments on five real-world, large-scale outdoor scenes demonstrate that AniGS produces natural ambient dynamics and high-quality novel view videos, enabling more immersive viewing experiences of reconstructed environments.

Figures

Figures reproduced from arXiv: 2607.18539 by Alexander Schwing, Ayush Saraf, Changil Kim, Chen Gao, Chuhan Chen, Hung-Yu Tseng, Johannes Kopf, Liangyan Gui, Rajvi Shah, Tuotuo Li, Yen-Chi Cheng.

Figure 1
Figure 1. Figure 1: 3D Scene Animation. Given a static 3D scene (top-left) reconstructed with 3D Gaussian Splatting (3DGS), our AniGS method synthesizes an animated 3D scene consisting of ambient motion that supports novel view video rendering. For each example rendered view, we highlight regions of interest and visualize the induced dynamics using space–time slices (𝑥/𝑦–𝑡), showing natural motion in foliage. ∗Word was done w… view at source ↗
Figure 2
Figure 2. Figure 2: Method overview. (left) AniGS uses a canonical 3DGS to represent scene geometry and appearance, and a time-conditioned deformation field to model the synthesized motion. The deformation field predicts discrete cosine transform (DCT) coefficients to offset and rotate each splat in the canonical 3DGS. (right) To animate the entire 3D scene, we design an iterative dataset–model update scheme with incremental … view at source ↗
Figure 3
Figure 3. Figure 3: Composed Video-to-Video (ComposedV2V). Slight motion intro￾duced by the video diffusion model in stationary regions can be amplified over the iterative dataset–model update process. We therefore propose ComposedV2V to re-stabilize the static region in each dataset update. Given a rendered (camera-fixed) video, we estimate the dynamic mask from the first frame, and create the composed video by copying the s… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results. We present the novel view rendering results on various scenes, and highlight the synthesized dynamics using space-time slices. t y x t x t x t Gazebo Park Courtyard t y t y [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative results on DL3DV. We present the novel view rendering results on scenes from DL3DV and highlight the animation. , Vol. 1, No. 1, Article . Publication date: July 2026 [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparisons. We present the novel view rendering results produced by different approaches: ours, Gaussians2life [30] and Phys￾Dreamer [39]. The synthesized dynamics is shown using space-time slices. x t [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Limitation example. AniGS cannot synthesize large, non-reversible dynamics. We show an example of the strong time-lapse effects of lighting changes. AniGS failed to model the global changes produced by the video diffusion model in this case and produce noticeable artifacts. x t t y x t [PITH_FULL_IMAGE:figures/full_fig_p011_7.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

47 extracted references · 14 linked inside Pith

  1. [1]

    Arslan Ali, Junjie Bai, Maciej Bala, Yogesh Balaji, Aaron Blakeman, Tiffany Cai, Jiaxin Cao, Tianshi Cao, Elizabeth Cha, Yu-Wei Chao, et al. 2025. World simulation with video foundation models for physical ai.arXiv preprint arXiv:2511.00062 (2025)

    Pith/arXiv arXiv 2025

  2. [2]

    Sherwin Bahmani, Ivan Skorokhodov, Victor Rong, Gordon Wetzstein, Leonidas Guibas, Peter Wonka, Sergey Tulyakov, Jeong Joon Park, Andrea Tagliasacchi, and David B. Lindell. 2024. 4D-fy: Text-to-4D Generation Using Hybrid Score Distillation Sampling. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2024

  3. [3]

    Joao Carreira and Andrew Zisserman. 2017. Quo vadis, action recognition? a new model and the kinetics dataset. Inproceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 6299–6308

    2017

  4. [4]

    Yoav HaCohen, Nisan Chiprut, Benny Brazowski, Daniel Shalem, Dudu Moshe, Eitan Richardson, Eran Levin, Guy Shiran, Nir Zabari, Ori Gordon, et al . 2024. Ltx-video: Realtime video latent diffusion.arXiv preprint arXiv:2501.00103(2024)

    Pith/arXiv arXiv 2024

  5. [5]

    Ayaan Haque, Matthew Tancik, Alexei A Efros, Aleksander Holynski, and Angjoo Kanazawa. 2023. Instruct-nerf2nerf: Editing 3d scenes with instructions. InPro- ceedings of the IEEE/CVF international conference on computer vision. 19740–19750

    2023

  6. [6]

    Jan Held, Renaud Vandeghen, Adrien Deliege, Abdullah Hamdi, Silvio Giancola, Anthony Cioppa, Andrea Vedaldi, Bernard Ghanem, Andrea Tagliasacchi, and Marc Van Droogenbroeck. 2025. Triangle Splatting for Real-Time Radiance Field Rendering.arXiv preprint arXiv:2505.19175(2025)

    Pith/arXiv arXiv 2025

  7. [7]

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. 2017. Gans trained by a two time-scale update rule converge to a local nash equilibrium.Advances in neural information processing systems30 (2017)

    2017

  8. [8]

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2024. 2d gaussian splatting for geometrically accurate radiance fields. InACM SIG- GRAPH 2024 conference papers. 1–11

    2024

  9. [9]

    Hanzhuo Huang, Yuan Liu, Ge Zheng, Jiepeng Wang, Zhiyang Dou, and Sibei Yang. 2025. MVTokenFlow: High-quality 4D Content Generation using Multiview Token Flow. InInternational Conference on Learning Representations (ICLR)

    2025

  10. [10]

    Yifan Jiang et al. 2024. Animate3D: Animating Any 3D Model with Multi-view Video Diffusion. InAdvances in Neural Information Processing Systems (NeurIPS)

    2024

  11. [11]

    Zeren Jiang, Chuanxia Zheng, Iro Laina, Diane Larlus, and Andrea Vedaldi. 2025. Geo4D: Leveraging Video Generators for Geometric 4D Scene Reconstruction. arXiv preprint arXiv:2504.07961(2025)

    Pith/arXiv arXiv 2025

  12. [12]

    In-Hwan Jin, Haesoo Choo, Seong-Hun Jeong, Heemoon Park, Junghwan Kim, Oh-joon Kwon, and Kyeongbo Kong. 2025. Optimizing 4D Gaussians for Dynamic Scene Video from Single Landscape Images.arXiv preprint arXiv:2504.05458 (2025)

    Pith/arXiv arXiv 2025

  13. [13]

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis

  14. [14]

    Mert Kiray, Paul Uhlenbruck, and Benjamin Busam. 2025. PromptVFX: Text-Driven Fields for Open-World 3D Gaussian Animation.arXiv preprint arXiv:2506.01091(2025)

    arXiv 2025

  15. [15]

    Feng, and Jia-Bin Huang

    Yao-Chih Lee, Yi-Ting Chen, Andrew Wang, Ting-Hsuan Liao, Brandon Y. Feng, and Jia-Bin Huang. 2024. VividDream: Generating 3D Scene with Ambient Dy- namics.arXiv preprint arXiv:2405.20334(2024)

    Pith/arXiv arXiv 2024

  16. [16]

    Xiang Li, Xinyu Ma, et al. 2025. Articulated Kinematics Distillation from Video Diffusion Models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2025

  17. [17]

    Lu Ling, Yichen Sheng, Zhi Tu, Wentian Zhao, Cheng Xin, Kun Wan, Lantao Yu, Qianyu Guo, Zixun Yu, Yawen Lu, et al . 2024. Dl3dv-10k: A large-scale scene dataset for deep learning-based 3d vision. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 22160–22169

    2024

  18. [18]

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

    2024

  19. [19]

    Linzhan Mou, Jiahui Lei, Chen Wang, Lingjie Liu, and Kostas Daniilidis. 2025. DIMO: Diverse 3D Motion Generation for Arbitrary Objects. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    2025

  20. [20]

    Oded Rahamim et al. 2024. Bringing Objects to Life: Training-Free 4D Generation from 3D Objects through View-Consistent Noise.arXiv preprint arXiv:2412.20422 (2024)

    Pith/arXiv arXiv 2024

  21. [21]

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman Rädle, Chloe Rolland, Laura Gustafson, et al. 2024. Sam 2: Segment anything in images and videos.arXiv preprint arXiv:2408.00714(2024)

    Pith/arXiv arXiv 2024

  22. [22]

    Jiawei Ren, Liang Pan, Jiaxiang Tang, Chi Zhang, Ang Cao, Gang Zeng, and Ziwei Liu. 2023. DreamGaussian4D: Generative 4D Gaussian Splatting.arXiv preprint arXiv:2312.17142(2023)

    Pith/arXiv arXiv 2023

  23. [23]

    Johannes L Schonberger and Jan-Michael Frahm. 2016. Structure-from-motion revisited. InProceedings of the IEEE conference on computer vision and pattern recognition. 4104–4113

    2016

  24. [24]

    Kaifeng Sheng, Zheng Zhou, Yingliang Peng, and Qianwei Wang. 2025. 2D Triangle Splatting for Direct Differentiable Mesh Training.arXiv preprint arXiv:2506.18575(2025)

    Pith/arXiv arXiv 2025

  25. [25]

    Meng-Li Shih et al . 2024. Modeling Ambient Scene Dynamics for Free-view Synthesis.arXiv preprint arXiv:2406.09395(2024)

    Pith/arXiv arXiv 2024

  26. [26]

    Qi Sun, Zhiyang Guo, Ziyu Wan, Jing Nathan Yan, Shengming Yin, Wengang Zhou, Jing Liao, and Houqiang Li. 2025. EG4D: Explicit Generation of 4D Object without Score Distillation. InInternational Conference on Learning Representations (ICLR)

    2025

  27. [27]

    Wenqiang Sun, Shuo Chen, Fangfu Liu, Zilong Chen, Yueqi Duan, Jun Zhang, and Yikai Wang. 2025. DimensionX: Create Any 3D and 4D Scenes from a Single Image with Controllable Video Diffusion. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    2025

  28. [28]

    Thomas Unterthiner, Sjoerd Van Steenkiste, Karol Kurach, Raphael Marinier, Marcin Michalski, and Sylvain Gelly. 2018. Towards accurate generative models of video: A new metric & challenges.arXiv preprint arXiv:1812.01717(2018)

    Pith/arXiv arXiv 2018

  29. [29]

    Qianqian Wang, Vickie Ye, Hang Gao, Weijia Zeng, and Angjoo Kanazawa

  30. [30]

    Thomas Wimmer, Michael Oechsle, Michael Niemeyer, and Federico Tombari

  31. [31]

    Guanjun Wu et al. 2024. 4D Gaussian Splatting for Real-Time Dynamic Scene Rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2024

  32. [32]

    Zizheng Wu et al . 2025. AnimateAnyMesh: A Feed-Forward 4D Foundation Model for Text-Driven Universal Mesh Animation. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    2025

  33. [33]

    Tianyi Xie, Zeshun Zong, Yuxing Qiu, Xuan Li, Yutao Feng, Yin Yang, and Chen- fanfu Jiang. 2024. Physgaussian: Physics-integrated 3d gaussians for generative dynamics. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 4389–4398

    2024

  34. [34]

    Yiming Xie, Chun-Han Yao, Vikram Voleti, Huaizu Jiang, and Varun Jampani

  35. [35]

    Jinbo Xing, Menghan Xia, Yong Zhang, Haoxin Chen, Wangbo Yu, Hanyuan Liu, Gongye Liu, Xintao Wang, Ying Shan, and Tien-Tsin Wong. 2024. Dynami- crafter: Animating open-domain images with video diffusion priors. InEuropean Conference on Computer Vision. Springer, 399–417

    2024

  36. [36]

    Chun-Han Yao, Yiming Xie, Vikram Voleti, Huaizu Jiang, and Varun Jampani. 2025. SV4D 2.0: Enhancing Spatio-Temporal Consistency in Multi-View Video Diffusion for High-Quality 4D Generation. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    2025

  37. [37]

    Zehao Yu, Anpei Chen, Binbin Huang, Torsten Sattler, and Andreas Geiger. 2024. Mip-splatting: Alias-free 3d gaussian splatting. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 19447–19456

    2024

  38. [38]

    InInternational Conference on Learning Representations (ICLR)

    SV4D: Dynamic 3D Content Generation with Multi-Frame and Multi-View Consistency. InInternational Conference on Learning Representations (ICLR)

  39. [39]

    Tao Zhang et al. 2024. PhysDreamer: Physics-Based Interaction with 3D Objects via Video Generation. InEuropean Conference on Computer Vision (ECCV)

    2024

  40. [40]

    Yuan Zhao, Shih-Yang Lin, Xinhang Li, Siyu Huang, Anpei Chen, et al . 2025. GenXD: Generating Any 3D and 4D Scenes. InInternational Conference on Learning Representations (ICLR). Introduces the CamVid-30K dataset

    2025

  41. [41]

    Yuchen Zhi et al. 2025. MV-Performer: Taming Video Diffusion Model for Faithful and Synchronized Multi-view Performer Synthesis.arXiv preprint arXiv:2510.07190 (2025)

    arXiv 2025

  42. [42]

    video-to-4D feedforward

    Bowen Zhang, Sicheng Xu, Chuxin Wang, Jiaolong Yang, Feng Zhao, Dong Chen, and Baining Guo. 2025. Gaussian Variation Field Diffusion for High-fidelity Video- to-4D Synthesis.arXiv preprint arXiv:2507.23785(2025). Use this entry for both “video-to-4D feedforward” and “single video to 4D” bullets

    Pith/arXiv arXiv 2025

  43. [46]

    Hanxin Zhu, Tianyu He, Xiqian Yu, Junliang Guo, Zhibo Chen, and Jiang Bian

  44. [47]

    AR4D: Autoregressive 4D Generation from Monocular Videos.arXiv preprint arXiv:2501.01722(2025). , Vol. 1, No. 1, Article . Publication date: July 2026. 10•Cheng et al. FireplaceFlowersTrail Garden t y x t x t x t x t t y x t x t t y x t t y t y Fig. 4.Qualitative results. We present the novel view rendering results on various scenes, and highlight the syn...

    Pith/arXiv arXiv 2025

  45. [2023]

    Graph.42, 4 (2023), 139–1

    3D Gaussian splatting for real-time radiance field rendering.ACM Trans. Graph.42, 4 (2023), 139–1

    2023

  46. [2024]

    Shape of Motion: 4D Reconstruction from a Single Video.arXiv preprint arXiv:2407.13764(2024)

    arXiv 2024

  47. [2025]

    Gaussians-to-Life: Text-Driven Animation of 3D Gaussian Splatting Scenes. , Vol. 1, No. 1, Article . Publication date: July 2026. AniGS: Bridging Rendering and Diffusion Prior for 3D Scene Animation•9 InInternational Conference on 3D Vision (3DV). 958–968

    2026