pith. sign in

arxiv: 2606.09187 · v1 · pith:DAMYDZXOnew · submitted 2026-06-08 · 💻 cs.CV

CP4D: Compositional Physics-aware 4D Scene Generation

Hanxin Zhu , Cong Wang , Tianyu He , Long Chen , Xin Jin , Chen Gao , Zhibo Chen This is my paper

Pith reviewed 2026-06-27 16:50 UTC · model grok-4.3

classification 💻 cs.CV
keywords 4D scene generationphysics-aware synthesiscompositional modelingvideo diffusion models3D scene compositiondynamic object trajectoriesphysical simulation
0
0 comments X

The pith

CP4D generates 4D scenes by composing static 3D environments with physically grounded dynamic objects.

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

Existing approaches to 4D generation often ignore physical laws and produce inconsistent or implausible motion. CP4D instead splits the task into a fixed 3D background generated separately from moving foreground objects whose paths are created by blending simulator rules with video-model intuition. A final automated step merges the parts into one scene. If the method works as described, users gain scenes they can navigate and manipulate while the objects follow believable dynamics and retain visual detail.

Core claim

The paper claims that reformulating 4D generation as the integration of a static 3D environment with physically grounded dynamic objects, produced via pre-trained expert models for each part, a hybrid motion synthesis step that merges simulator priors with video diffusion common sense, and an automated composition mechanism, yields photorealistic 4D scenes that respect complex physical dynamics.

What carries the argument

The hybrid motion synthesis strategy, which combines priors from physical simulators with the common sense embedded in video diffusion models to create object trajectories and interactions.

If this is right

  • Generated scenes support exploration and interaction while maintaining visual fidelity.
  • Dynamic objects exhibit strong physical plausibility in their motion and contacts.
  • Users obtain fine-grained control over the placement and behavior of objects in the final 4D output.
  • The resulting scenes outperform prior methods on standard visual and physical quality metrics.

Where Pith is reading between the lines

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

  • The separation of static and dynamic elements could allow reuse of existing 3D asset libraries without retraining an entire 4D model from scratch.
  • The same pipeline might support insertion of real captured objects into simulated environments for mixed-reality testing.
  • Scaling the hybrid synthesis to many interacting objects would test whether the current merging step remains stable.

Load-bearing premise

The hybrid motion synthesis strategy that integrates priors from physical simulators with the common sense embedded in video diffusion models will reliably produce trajectories and interactions that are both physically accurate and visually coherent without additional constraints or post-processing.

What would settle it

Generate a scene containing a dropped rigid object on a flat surface and measure whether the resulting trajectory shows penetration, floating, or other violations that the method claims to avoid.

Figures

Figures reproduced from arXiv: 2606.09187 by Chen Gao, Cong Wang, Hanxin Zhu, Long Chen, Tianyu He, Xin Jin, Zhibo Chen.

Figure 1
Figure 1. Figure 1: Pipeline of CP4D. Given a textual prompt, CP4D constructs a physically faithful 4D scene via a three-stage pipeline: 1) synthesizing 3D representations for both foreground objects and background environments (Sec. 4.1), 2) simulating foreground motions with physical grounding to ensure realistic dynamics (Sec. 4.2), and 3) automatically composing the foregrounds and background into a coherent and visually … view at source ↗
Figure 2
Figure 2. Figure 2: (a) Limited numerical precision in the physics solver leads to erroneous estimation of foreground geometry. (b) As a result, the solver reports collisions that are not visually manifested, producing spurious intersections. (c) Our method eliminates these inconsistencies, enabling interactions that are both visually coher￾ent and physically faithful. formally expressed as follows: Gb = F b 3d (Ib), Gf = F f… view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the depth-aware heuristic for initializing the scale S. Left: the foreground Gf is independently gener￾ated and may exceed the camera frustum at depth P z . Middle: to ensure full visibility, its projected extent in the x–y plane is constrained by the frustum bounds (B xmin, Bx max, By min, By max), yielding the maximum feasible scale S. Right: applying this ini￾tialized S guarantees that G… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparisons. The top row shows the given text prompt and the corresponding generated image. Our method generates temporally consistent and physically plausible videos, outperforming baseline approaches in both visual fidelity and physical realism. coordinate system of the background with the camera co￾ordinate system, the z-coordinate of P (i.e., P z ) is directly equal to the corresponding dep… view at source ↗
Figure 5
Figure 5. Figure 5: Ablation study. Results of ablation on optimizing VLM￾estimated physical parameters and foreground object positions. Original Sene Background Editing Object Editing [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Editing results. Examples of background environment and foreground object editing in generated 4D scenes. 5.2. Comparisons with State-Of-The-Art Methods As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Fixed structural tokens appear in black, background descriptors in red, and motion-related descriptors of foreground objects in blue. I would like you to evaluate the quality of a generated videos based on the following criteria: physical realism, photorealism, and semantic consistency. The evaluation will be based on 10 evenly sampled frames from each video. Given the original image and the following inst… view at source ↗
Figure 8
Figure 8. Figure 8: Prompt used for GPT-4o evaluation. B. More experimental details B.1. Dataset curation To construct a dataset capable of comprehensively evaluating adherence to physical laws, we adapt textual prompts from VideoPhy (Bansal et al., 2024) and design a set of 34 representative prompts. These prompts ensure coverage of at least two distinct examples for each physical category, including rigid-body, elastic, def… view at source ↗
Figure 9
Figure 9. Figure 9: Prompt template for automated material estimation. via the forward Euler scheme, is expressed as: mi (v n+1 i − v n i ) ∆t = − X p V 0 p ∂Ψ ∂F (FpE,n)F T pE,n∇w n ip + f ext i , (10) here, i and p denote the fields on the Eulerian grid and the Lagrangian particles, respectively; w n ip is the B-spline kernel defined on the i-th grid evaluated at x n p ; V 0 p represents the initial volume, and ∆t is the ti… view at source ↗
Figure 10
Figure 10. Figure 10: Prompt template for automated external forces estimation. D.2. Rigid collision simulation Our proposed method integrates traditional rigid body dynamics with 3D Gaussian particle rendering, establishing a unified framework for physics simulation and visual rendering. The core innovation lies in representing complex 3D objects as collections of Gaussian particles while maintaining rigid body constraints fo… view at source ↗
Figure 11
Figure 11. Figure 11: Physical simulation results of elastic bodies under different density and Young’s modulus [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Multi-puck rigid-body collision cases in shuffleboard simulation. where hground is the ground height and ϵ is a small tolerance for numerical stability. Impulse-Based Collision Response. The collision response employs an impulse-based approach: Linear Impulse: J = −(1 + e) v · n 1/m1 + 1/m2 , (16) Angular Impulse: τ = r × J, (17) where r is the vector from the center of mass to the collision point, and e … view at source ↗
Figure 13
Figure 13. Figure 13: An example of raindrop falling simulated using the PBD solver. Raindrops are visually scaled up here to facilitate observation of fluid motion dynamics. D.3. Position-Based Dynamics solver for fluid objects The PBD framework operates on the fundamental principle of constraint satisfaction rather than force integration. For fluid simulation, the primary constraint is density preservation: Ci(x1, x2, ..., x… view at source ↗
Figure 14
Figure 14. Figure 14: Qualitative comparisons of ablation studies on the position initialization and scale initialization illustrated in Sec. 4.3. Position Correction: ∆xi = λi∇Ci . (27) Iterative Refinement: x k+1 i = x k i + ∆xi , (28) where k denotes the iteration number, and ϵ is a small regularization term. Boundary Constraint Handling. Boundary constraints are enforced through position clamping and velocity reflection: P… view at source ↗
Figure 15
Figure 15. Figure 15: Visual comparison with the baseline that separately generates the foreground and background from text prompts. Independent synthesis often causes inconsistencies in style, texture, and geometry, leading to unrealistic compositions. Our method produces coherent and visually consistent foreground–background results. In these exploratory experiments, we leverage video generative models together with an SDS-b… view at source ↗
Figure 16
Figure 16. Figure 16: Qualitative visualization results of our method and the baselines. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Qualitative visualization results of our method and the baselines. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Qualitative visualization results of our method and the baselines. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Qualitative visualization results of our method and the baselines. 24 [PITH_FULL_IMAGE:figures/full_fig_p024_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Visualization results of the motion process observed from different viewpoints View 1 View 2 View 1 View 2 [PITH_FULL_IMAGE:figures/full_fig_p025_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Visualization results of the motion process observed from different viewpoints 25 [PITH_FULL_IMAGE:figures/full_fig_p025_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Visualization results of the motion process observed from different viewpoints View 1 View 2 View 1 View 2 [PITH_FULL_IMAGE:figures/full_fig_p026_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Visualization results of the motion process observed from different viewpoints 26 [PITH_FULL_IMAGE:figures/full_fig_p026_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: Visualization results from more diverse camera viewpoints [PITH_FULL_IMAGE:figures/full_fig_p027_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Visualization results of preliminary explorations towards achieving more complex foreground-background interactions. 27 [PITH_FULL_IMAGE:figures/full_fig_p027_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: Supplemented visualization results of different foreground dynamics. 28 [PITH_FULL_IMAGE:figures/full_fig_p028_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: Supplemented visualization results of various scenarios. 29 [PITH_FULL_IMAGE:figures/full_fig_p029_27.png] view at source ↗
read the original abstract

4D generation (\textit{i.e.}, dynamic 3D generation) has recently emerged as a rapidly growing research frontier due to its powerful spatiotemporal modeling capabilities. However, despite notable advances, existing approaches typically fail to capture the underlying physical principles, producing results that are both physically inconsistent and visually implausible. To overcome this limitation, we present CP4D, a novel paradigm for photorealistic 4D scene synthesis with faithful adherence to complex physical dynamics. Drawing inspiration from the compositional nature of real-world scenes, where immutable static backgrounds coexist with dynamic, physically plausible foregrounds, CP4D reformulates 4D generation as the integration of a static 3D environment with physically grounded dynamic objects. On this basis, our framework follows a three-stage pipeline: \textbf{1)} Firstly, we leverage pre-trained expert models to generate high-fidelity 3D representations of the environment and foreground objects respectively. \textbf{2)} Subsequently, to produce physically plausible trajectories and realistic interactions for these objects, we propose a hybrid motion synthesis strategy that integrates priors from physical simulators with the common sense embedded in video diffusion models. \textbf{3)} Finally, we develop an automated composition mechanism that seamlessly fuses the static environment and dynamic objects into coherent, physically consistent 4D scenes. Extensive experiments demonstrate that CP4D can generate explorable and interactive 4D scenes with high visual fidelity, strong physical plausibility, and fine-grained controllability, significantly outperforming existing methods. The project page: https://anonymous.4open.science/w/CP4D/.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces CP4D, a three-stage compositional framework for 4D scene generation: (1) separate generation of high-fidelity static 3D environments and dynamic foreground objects via pre-trained expert models, (2) hybrid motion synthesis that integrates priors from physical simulators with common sense from video diffusion models to produce trajectories and interactions, and (3) an automated composition mechanism to fuse them into coherent scenes. It claims this yields explorable, interactive 4D scenes with high visual fidelity, strong physical plausibility, fine-grained controllability, and significant outperformance over existing methods.

Significance. If the physical-plausibility and outperformance claims are substantiated with quantitative evidence, the work would represent a meaningful advance in 4D generation by explicitly incorporating physical dynamics rather than relying solely on learned priors, with potential impact on simulation, robotics, and interactive media.

major comments (2)
  1. [Method (stage 2)] The hybrid motion synthesis strategy (stage 2) is presented only at a high level as 'integrating priors from physical simulators with the common sense embedded in video diffusion models' together with an 'automated composition mechanism,' without any equations, algorithm, conflict-resolution procedure, or constraints. This mechanism is load-bearing for the central claim of 'strong physical plausibility' and 'without additional constraints or post-processing.'
  2. [Abstract / Experiments] The abstract states that 'extensive experiments demonstrate' outperformance and physical plausibility, yet supplies no quantitative metrics, baselines, error analysis, or evaluation protocols. This absence directly undermines assessment of the soundness of the physical-consistency and superiority claims.
minor comments (1)
  1. [Abstract] The project page is listed as anonymous; while acceptable for review, the absence of any concrete numerical results or protocol details in the abstract reduces immediate clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments correctly identify areas where additional technical detail and quantitative evidence are needed to support the central claims. We address each point below and will incorporate revisions accordingly.

read point-by-point responses

  1. Referee: [Method (stage 2)] The hybrid motion synthesis strategy (stage 2) is presented only at a high level as 'integrating priors from physical simulators with the common sense embedded in video diffusion models' together with an 'automated composition mechanism,' without any equations, algorithm, conflict-resolution procedure, or constraints. This mechanism is load-bearing for the central claim of 'strong physical plausibility' and 'without additional constraints or post-processing.'

    Authors: We agree that the high-level description in the current manuscript is insufficient to fully substantiate the physical-plausibility claims or enable reproduction. In the revised manuscript we will expand Section 3.2 to include the explicit equations for combining simulator priors with diffusion outputs, the full algorithm with pseudocode, the conflict-resolution procedure between physics and learned common sense, and the constraints (if any) applied during composition. revision: yes

  2. Referee: [Abstract / Experiments] The abstract states that 'extensive experiments demonstrate' outperformance and physical plausibility, yet supplies no quantitative metrics, baselines, error analysis, or evaluation protocols. This absence directly undermines assessment of the soundness of the physical-consistency and superiority claims.

    Authors: The observation is accurate: the abstract currently asserts results without accompanying numbers or protocols. We will revise the abstract to reference the specific quantitative metrics, baselines, and evaluation protocols used, and we will expand the experiments section to report all numerical results, error analyses, and comparison details that support the outperformance and physical-plausibility statements. revision: yes

Circularity Check

0 steps flagged

No circularity: engineering pipeline with no derivation chain

full rationale

The paper describes CP4D as a three-stage engineering pipeline (pre-trained 3D generation, hybrid motion synthesis integrating simulators and video diffusion, automated composition) without any equations, fitted parameters, predictions, or first-principles derivations. No self-definitional steps, fitted inputs renamed as predictions, or load-bearing self-citations appear in the provided text. The central claims rest on the integration of external pre-trained models and an unspecified hybrid strategy rather than any result that reduces to its own inputs by construction, making the method self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no equations, parameters, or explicit assumptions; ledger entries cannot be populated.

pith-pipeline@v0.9.1-grok · 5829 in / 1079 out tokens · 12539 ms · 2026-06-27T16:50:36.495614+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

67 extracted references · 12 linked inside Pith

  1. [1]

    arXiv preprint arXiv:2406.03520 , year=

    Videophy: Evaluating physical commonsense for video generation , author=. arXiv preprint arXiv:2406.03520 , year=

    Pith/arXiv arXiv

  2. [2]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Physgaussian: Physics-integrated 3d gaussians for generative dynamics , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  3. [3]

    arXiv preprint arXiv:2411.12789 , year=

    Efficient Physics Simulation for 3D Scenes via MLLM-Guided Gaussian Splatting , author=. arXiv preprint arXiv:2411.12789 , year=

    arXiv

  4. [4]

    2025 IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW) , pages=

    LIVE-GS: LLM Powers Interactive VR by Enhancing Gaussian Splatting , author=. 2025 IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW) , pages=. 2025 , organization=

    2025

  5. [5]

    Proceedings of the AAAI Conference on Artificial Intelligence , pages=

    DreamPhysics: Learning Physics-Based 3D Dynamics with Video Diffusion Priors , author=. Proceedings of the AAAI Conference on Artificial Intelligence , pages=

  6. [6]

    arXiv preprint arXiv:2406.04338 , year=

    Physics3d: Learning physical properties of 3d gaussians via video diffusion , author=. arXiv preprint arXiv:2406.04338 , year=

    arXiv

  7. [7]

    arXiv preprint arXiv:2501.18982 , year=

    OmniphysGS: 3d constitutive gaussians for general physics-based dynamics generation , author=. arXiv preprint arXiv:2501.18982 , year=

    arXiv

  8. [8]

    arXiv preprint arXiv:2409.07179 , year=

    Phy124: Fast physics-driven 4d content generation from a single image , author=. arXiv preprint arXiv:2409.07179 , year=

    arXiv

  9. [9]

    arXiv preprint arXiv:2411.16800 , year=

    Phys4DGen: Physics-Compliant 4D Generation with Multi-Material Composition Perception , author=. arXiv preprint arXiv:2411.16800 , year=

    arXiv

  10. [10]

    arXiv preprint arXiv:2411.17189 , year=

    Physmotion: Physics-grounded dynamics from a single image , author=. arXiv preprint arXiv:2411.17189 , year=

    arXiv

  11. [11]

    ACM Transactions on Graphics (TOG) , volume=

    A moving least squares material point method with displacement discontinuity and two-way rigid body coupling , author=. ACM Transactions on Graphics (TOG) , volume=. 2018 , publisher=

    2018

  12. [12]

    ACM Transactions on Graphics (TOG) , pages=

    Anisotropic elastoplasticity for cloth, knit and hair frictional contact , author=. ACM Transactions on Graphics (TOG) , pages=. 2017 , publisher=

    2017

  13. [13]

    ACM Transactions on Graphics (TOG) , pages=

    Taichi: a language for high-performance computation on spatially sparse data structures , author=. ACM Transactions on Graphics (TOG) , pages=. 2019 , publisher=

    2019

  14. [14]

    , author=

    3D Gaussian splatting for real-time radiance field rendering. , author=. ACM Trans. Graph. , pages=

  15. [15]

    arXiv preprint arXiv:2209.14988 , year=

    Dreamfusion: Text-to-3d using 2d diffusion , author=. arXiv preprint arXiv:2209.14988 , year=

    Pith/arXiv arXiv

  16. [16]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Uniscene: Unified occupancy-centric driving scene generation , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  17. [17]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Drivedreamer4d: World models are effective data machines for 4d driving scene representation , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  18. [18]

    arXiv preprint arXiv:2506.04225 , year=

    Voyager: Long-Range and World-Consistent Video Diffusion for Explorable 3D Scene Generation , author=. arXiv preprint arXiv:2506.04225 , year=

    arXiv

  19. [19]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Wonderworld: Interactive 3d scene generation from a single image , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  20. [20]

    arXiv preprint arXiv:2409.02048 , year=

    Viewcrafter: Taming video diffusion models for high-fidelity novel view synthesis , author=. arXiv preprint arXiv:2409.02048 , year=

    Pith/arXiv arXiv

  21. [21]

    ACM Transactions on Graphics (TOG) , volume=

    A material point method for snow simulation , author=. ACM Transactions on Graphics (TOG) , volume=. 2013 , publisher=

    2013

  22. [22]

    Acm siggraph 2016 courses , pages=

    The material point method for simulating continuum materials , author=. Acm siggraph 2016 courses , pages=

    2016

  23. [23]

    2006 , publisher=

    Computational Inelasticity , author=. 2006 , publisher=

    2006

  24. [24]

    Advances in neural information processing systems , volume=

    Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distillation , author=. Advances in neural information processing systems , volume=

  25. [25]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Physgen3d: Crafting a miniature interactive world from a single image , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  26. [26]

    European Conference on Computer Vision , pages=

    Physgen: Rigid-body physics-grounded image-to-video generation , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  27. [27]

    2025 , booktitle=

    CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer , author=. 2025 , booktitle=

    2025

  28. [28]

    arXiv preprint arXiv:2503.20314 , year=

    Wan: Open and advanced large-scale video generative models , author=. arXiv preprint arXiv:2503.20314 , year=

    Pith/arXiv arXiv

  29. [29]

    arXiv preprint arXiv:2312.17142 , year=

    Dreamgaussian4d: Generative 4d gaussian splatting , author=. arXiv preprint arXiv:2312.17142 , year=

    arXiv

  30. [30]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Vbench: Comprehensive benchmark suite for video generative models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  31. [31]

    arXiv preprint arXiv:2504.00983 , year=

    Worldscore: A unified evaluation benchmark for world generation , author=. arXiv preprint arXiv:2504.00983 , year=

    arXiv

  32. [32]

    Advances in neural information processing systems , volume=

    Denoising diffusion probabilistic models , author=. Advances in neural information processing systems , volume=

  33. [33]

    arXiv preprint arXiv:2010.02502 , year=

    Denoising diffusion implicit models , author=. arXiv preprint arXiv:2010.02502 , year=

    Pith/arXiv arXiv 2010

  34. [34]

    arXiv preprint arXiv:2407.17470 , year=

    Sv4d: Dynamic 3d content generation with multi-frame and multi-view consistency , author=. arXiv preprint arXiv:2407.17470 , year=

    arXiv

  35. [35]

    arXiv preprint arXiv:2311.02848 , year=

    Consistent4d: Consistent 360 \ deg \ dynamic object generation from monocular video , author=. arXiv preprint arXiv:2311.02848 , year=

    arXiv

  36. [36]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    4d-fy: Text-to-4d generation using hybrid score distillation sampling , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  37. [37]

    arXiv preprint arXiv:2406.13527 , year=

    4k4dgen: Panoramic 4d generation at 4k resolution , author=. arXiv preprint arXiv:2406.13527 , year=

    arXiv

  38. [38]

    arXiv preprint arXiv:2507.01099 , year=

    Geometry-aware 4D Video Generation for Robot Manipulation , author=. arXiv preprint arXiv:2507.01099 , year=

    Pith/arXiv arXiv

  39. [39]

    arXiv preprint arXiv:2506.01103 , year=

    DeepVerse: 4D Autoregressive Video Generation as a World Model , author=. arXiv preprint arXiv:2506.01103 , year=

    arXiv

  40. [40]

    arXiv preprint arXiv:2503.05638 , year=

    Trajectorycrafter: Redirecting camera trajectory for monocular videos via diffusion models , author=. arXiv preprint arXiv:2503.05638 , year=

    arXiv

  41. [41]

    arXiv preprint arXiv:2506.04590 , year=

    Follow-Your-Creation: Empowering 4D Creation through Video Inpainting , author=. arXiv preprint arXiv:2506.04590 , year=

    arXiv

  42. [42]

    European Conference on Computer Vision , pages=

    Tc4d: Trajectory-conditioned text-to-4d generation , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  43. [43]

    European Conference on Computer Vision , pages=

    Stag4d: Spatial-temporal anchored generative 4d gaussians , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  44. [44]

    Advances in Neural Information Processing Systems , volume=

    L4gm: Large 4d gaussian reconstruction model , author=. Advances in Neural Information Processing Systems , volume=

  45. [45]

    arXiv preprint arXiv:2405.16645 , year=

    Diffusion4d: Fast spatial-temporal consistent 4d generation via video diffusion models , author=. arXiv preprint arXiv:2405.16645 , year=

    arXiv

  46. [46]

    arXiv preprint arXiv:2503.11647 , year=

    Recammaster: Camera-controlled generative rendering from a single video , author=. arXiv preprint arXiv:2503.11647 , year=

    arXiv

  47. [47]

    arXiv preprint arXiv:2403.16993 , year=

    Comp4d: Llm-guided compositional 4d scene generation , author=. arXiv preprint arXiv:2403.16993 , year=

    arXiv

  48. [48]

    Advances in neural information processing systems , volume=

    Compositional 3d-aware video generation with llm director , author=. Advances in neural information processing systems , volume=

  49. [49]

    arXiv preprint arXiv:2501.01722 , year=

    Ar4d: Autoregressive 4d generation from monocular videos , author=. arXiv preprint arXiv:2501.01722 , year=

    arXiv

  50. [50]

    Advances in Neural Information Processing Systems , volume=

    Dreammesh4d: Video-to-4d generation with sparse-controlled gaussian-mesh hybrid representation , author=. Advances in Neural Information Processing Systems , volume=

  51. [51]

    arXiv preprint arXiv:2403.12365 , year=

    Gaussianflow: Splatting gaussian dynamics for 4d content creation , author=. arXiv preprint arXiv:2403.12365 , year=

    arXiv

  52. [52]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Nerfies: Deformable neural radiance fields , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  53. [53]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Objaverse: A universe of annotated 3d objects , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  54. [54]

    arXiv preprint arXiv:2407.02371 , year=

    Openvid-1m: A large-scale high-quality dataset for text-to-video generation , author=. arXiv preprint arXiv:2407.02371 , year=

    Pith/arXiv arXiv

  55. [55]

    arXiv preprint arXiv:2404.02101 , year=

    Cameractrl: Enabling camera control for text-to-video generation , author=. arXiv preprint arXiv:2404.02101 , year=

    Pith/arXiv arXiv

  56. [56]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    4d gaussian splatting for real-time dynamic scene rendering , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  57. [57]

    arXiv preprint arXiv:2303.08774 , year=

    Gpt-4 technical report , author=. arXiv preprint arXiv:2303.08774 , year=

    Pith/arXiv arXiv

  58. [58]

    arXiv preprint arXiv:2508.02324 , year=

    Qwen-image technical report , author=. arXiv preprint arXiv:2508.02324 , year=

    Pith/arXiv arXiv

  59. [59]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Segment anything , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  60. [60]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Depth anything: Unleashing the power of large-scale unlabeled data , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  61. [61]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Structured 3d latents for scalable and versatile 3d generation , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  62. [62]

    arXiv preprint arXiv:1412.6980 , year=

    Adam: A method for stochastic optimization , author=. arXiv preprint arXiv:1412.6980 , year=

    Pith/arXiv arXiv

  63. [63]

    arXiv preprint arXiv:2308.06571 , year=

    Modelscope text-to-video technical report , author=. arXiv preprint arXiv:2308.06571 , year=

    Pith/arXiv arXiv

  64. [64]

    arXiv preprint arXiv:2508.15376 , year=

    DriveSplat: Decoupled Driving Scene Reconstruction with Geometry-enhanced Partitioned Neural Gaussians , author=. arXiv preprint arXiv:2508.15376 , year=

    arXiv

  65. [65]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    FluidNexus: 3D fluid reconstruction and prediction from a single video , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  66. [66]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Unleashing the potential of multi-modal foundation models and video diffusion for 4d dynamic physical scene simulation , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  67. [67]

    arXiv preprint arXiv:2411.04989 , year=

    Sg-i2v: Self-guided trajectory control in image-to-video generation , author=. arXiv preprint arXiv:2411.04989 , year=

    arXiv