Pith. sign in

REVIEW 5 major objections 7 minor 70 references

DrivingGaussian++: Towards Realistic Reconstruction and Editable Simulation for Surrounding Dynamic Driving Scenes

T0 review · 5 major / 7 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read DrivingGaussian++ claims a single composite 3D Gaussian representation can reconstruct dynamic surround-view driving scenes at state-of-the-art quality and edit them training-free in minutes.

desk verdict Useful editing extension of the authors' prior DrivingGaussian, but the central multi-view consistency claim rests on an unvalidated depth back-projection and a metric that never measures consistency. read the letter →

arxiv 2508.20965 v1 pith:VBBTE4S3 submitted 2025-08-28 cs.CV

Yajiao Xiong , Xiaoyu Zhou , Yongtao Wan , Deqing Sun , Ming-Hsuan Yang This is my paper
classification cs.CV
keywords 3DGaussianSplattingdynamicscenereconstructionautonomousdrivingsimulationsurround-viewsynthesisLiDARpriorcontrollableeditingweatherobjectmanipulation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

DrivingGaussian++ is trying to show that one 3D representation can do both jobs needed for autonomous-driving simulation: faithfully reconstruct a large-scale dynamic surround-view scene and let it be edited in many ways without retraining. The representation is Composite Gaussian Splatting, with incremental 3D Gaussians for the static background, a composite dynamic Gaussian graph for moving objects, and LiDAR-based initialization for geometric accuracy. On nuScenes the reconstructed model reports PSNR 28.74 / SSIM 0.865 versus EmerNeRF's 26.75 / 0.760, and editing a scene takes about 8 minutes compared with roughly 1 to 4.5 hours for the diffusion-based baselines, with higher CLIP-direction consistency. If true, this would make large-scale, diverse driving-scene simulation practical for training and testing perception systems.

What carries the argument

Composite Gaussian Splatting: the scene is split into Incremental Static 3D Gaussians, reconstructed bin-by-bin along the ego trajectory, and a Composite Dynamic Gaussian Graph, where each moving object stores its own Gaussians together with a transform, opacity, and temporal state. The LiDAR prior seeds the static Gaussians, and the global renderer uses the standard 3D Gaussian splatting rasterizer; editing operates directly on this explicit representation through inverse projection from rendered depth, so multi-view consistency is carried by geometry instead of per-frame 2D editing.

What would settle it

Render a known planar surface (e.g., the road) from the reconstructed Gaussian model and compare the rendered depth against the LiDAR sweep at the same timestamp; a systematic depth bias at object boundaries or in distant regions would predict cross-view misalignment of edited content. A more direct test: edit one object from a single viewpoint, then render the edited scene from the opposite surround camera and measure the position shift of the edited texture against a manually composited reference; a visible shift or ghosting would falsify the depth-based inverse projection assumption.

Watch

Extended reading notes

Core claim

The paper's central claim is that a composite explicit Gaussian scene, built by reconstructing the static background incrementally and each moving object as a node in a dynamic Gaussian graph, can be both state-of-the-art in reconstruction and a substrate for fast training-free multi-task editing. LiDAR points are used not merely as depth supervision but as the initialization of the Gaussians, which the authors argue recovers cleaner geometry and better multi-camera consistency. Editing is done at the level of the Gaussians themselves: textures are back-projected from 2D edits using rendered depth, weather is added as particle Gaussians with physical trajectories, and object insertion uses a

Load-bearing premise

The editing pipeline assumes the depth maps rendered by the reconstructed 3D Gaussian field are accurate enough to back-project edited 2D content into correctly placed 3D Gaussians; if rendered depth is biased at object boundaries or in distant regions, edited textures and snow cover will be geometrically misplaced across views.

Editorial extensions

If this is right

  • Reconstruction and editing are decoupled: any pre-reconstructed 3D Gaussian scene can be edited without retraining, bringing editing cost down to minutes.
  • The same framework handles texture modification, weather simulation, object removal and insertion, and dynamic trajectory prediction, replacing several task-specific editing pipelines.
  • Using LiDAR as Gaussian initialization rather than only as depth supervision improves geometry and multi-view consistency; the ablation and monocular KITTI-360 results indicate the method is not strictly dependent on LiDAR.
  • Edited scenes can produce hard cases such as foggy conditions, cracks, and added objects that degrade object detectors, making the pipeline useful for robustness testing of perception systems.
  • Higher CLIP-direction similarity than the compared baselines suggests edits align with text intent while remaining consistent across surrounding views.

Reading between the lines

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

  • An open test is whether the training-free editing generalizes when rendered depth is inaccurate: measuring back-projected edit alignment against LiDAR ground truth at object boundaries would directly probe the load-bearing depth assumption.
  • Because editing works on explicit Gaussians, the same inverse-projection recipe could transfer to other explicit scene representations and other domains, provided a reliable depth prior exists.
  • The LLM-based trajectory prediction is a black-box component; physical plausibility, collision avoidance, and interaction with existing traffic are not systematically evaluated, leaving a natural extension for dynamic insertion.
  • Editing consistency is currently quantified on a small set of scenes with CLIP-direction similarity; longer sequences and human perceptual studies would be a stronger test of temporal consistency.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 7 minor

Summary. This paper extends the CVPR 2024 DrivingGaussian framework to DrivingGaussian++. It reconstructs large-scale dynamic driving scenes using incremental static 3D Gaussians and a composite dynamic Gaussian graph, initialized with LiDAR points and supervised by TSSIM, robust, and LiDAR losses. On nuScenes and KITTI-360 it reports PSNR/SSIM/LPIPS improvements over EmerNeRF, 3DGS, SUDS, and other baselines. The new claimed contribution is a training-free multi-task editing layer: texture modification by 2D diffusion editing and depth back-projection, weather simulation via procedural particle Gaussians and snow coverage, and object insertion/deletion using a foreground asset bank; LLM (GPT-4o) predicts motion trajectories, and refinement uses inpainting and shadow synthesis. The paper reports large speedups and CLIP-direction scores for editing.

Significance. If the editing pipeline works as claimed, this would be a useful step toward practical dynamic-driving simulation, because it decouples reconstruction from editing and avoids per-task optimization. The reconstruction formulation has plausible components (LiDAR prior, incremental bins, dynamic Gaussian graph), and the reported reconstruction gains over EmerNeRF are substantial. Editing is demonstrated across several tasks with appealing qualitative examples and dramatic speedups. However, the paper's central new claim—consistent and realistic training-free editing—is not yet supported by the evaluation: there is no multi-view consistency metric, no depth-error validation of the back-projection step, no held-out frame evaluation, and the editing protocols are under-specified. Reconstruction numbers also need clarification as to whether they are training-frame or held-out results. With revisions adding targeted experiments, the contribution could become significant.

major comments (5)
  1. [Sec. 4.4.2 / Tables 2 and 4] The editing evaluation does not measure the property that is claimed. CLIP direction similarity compares text and image embeddings and cannot detect cross-view misplacement or geometric inconsistency of back-projected edits. For a 'consistent and realistic editing' claim, the paper needs a multi-view consistency metric (e.g., reprojection error of edited Gaussians, LPIPS/PSNR across overlapping edited cameras, or a user study) and per-task protocols (prompts, masks, number of scenes, seed). The single CLIP-dir number in Table 2 aggregates four tasks, making it impossible to interpret.
  2. [Sec. 3.3.1 / Eq. (14)] The depth-backprojection step is load-bearing for texture and snow editing. D_opt = Average_x(D_ori) assumes rendered 3DGS depth is unbiased and that horizontal flattening preserves edited-surface geometry. The paper itself notes depth discrepancies ('surface ... uneven and unrealistic') but does not validate Eq. (14). Please report depth error of D_ori and D_opt against LiDAR for edited regions, separated by surface type (road, vertical sign, vehicle), and show edited Gaussians rendered from novel viewpoints. Without this, the geometric-consistency claim for the novel editing layer is unsupported.
  3. [Sec. 4.2 / Table 1] The reconstruction evaluation does not state whether metrics are computed on held-out timestamps or views; 'report the average results of all camera frames in the scenes' suggests training frames were used. Since the abstract claims 'photorealistic surround-view synthesis,' held-out evaluation is needed. Also add variance over scene splits or seeds; margins over EmerNeRF are large but the editing comparison in Table 2 has no error bars.
  4. [Sec. 4.4.2 / Tables 2 and 4] The comparison protocol is under-specified. Prompts, masks, editing regions, scene identities, and LLM/diffusion settings are not given for any baseline or for Ours. Table 4(b) says 'LPIPS and FID' but reports only LPIPS. ClimateNeRF supports one task while Table 2 labels it as multi-task. A fair comparison needs per-task numbers and a clearly specified protocol for every method.
  5. [Sec. 3.3 / Eqs. (15), (19), (20)] The new editing pipeline has several free parameters and stochastic components that determine the results: particle range parameters zeta and epsilon, inpainting threshold d_thr, KNN pruning threshold, the GPT-4o prompt/temperature/seed, and the multi-camera fusion weight omega in Eq. (5). These are not specified, so the reported 8-minute editing time and qualitative outputs are not reproducible. Please report these settings or include a sensitivity analysis.
minor comments (7)
  1. [Throughout] Typos and wording issues: 'Cotrollable' in Sec. 2.2, 'matirx' after Eq. (5), 'denots' after Eq. (9), 'perforamnce' in Tables 1 and 3, 'T rajj' in Eqs. (16) and (20), and '1th' in Eq. (19).
  2. [Eq. (15)] The definition of s_{k,y} appears to be a typo: s_{k,y} = min(min(s_{k,x}, s_{k,x}), 0) + epsilon ignores s_{k,y} and likely should be min(s_{k,x}, s_{k,y}). Please correct.
  3. [Sec. 3.1.3 / Eq. (8)] The opacity adjustment formula uses cot(a_o) without defining the angle or deriving the expression. Please clarify the geometric meaning.
  4. [Table 4(b)] The text mentions 'LPIPS and FID' for object removal, but the table reports only LPIPS. Add FID or remove the claim.
  5. [Abstract] Two different project-page URLs appear in the abstract; please ensure they are consistent.
  6. [Sec. 4.2] The hardware description '8 RTX8000 with 384 GB memory' is ambiguous: is this 8 GPUs with 48 GB each, or a different configuration? Clarify for reproducibility.
  7. [Sec. 4.4.2] Define CLIP-direction similarity precisely: which CLIP model, which text prompts, and how direction is computed between source and edited images.

Circularity Check

1 steps flagged · score 2.0 of 10

Reconstruction is externally benchmarked and editing is training-free, but the CLIP-direction editing metric partially inherits the 2D diffusion editor's own prompt alignment, giving the headline editing comparison a mildly circular evaluative component.

  1. renaming known result [Section 3.3.1 (Texture Modification) and Section 4.4.2 (Quantitative Results, Tables 2 and 4)]
    "Next, we define the 2D mask of the target region and apply the diffusion model or image processing software to modify the image in a 2D space, producing the target image. Using the target image and the mask, we generate target Gaussians and assign appropriate attributes through inverse projection. ... We evaluate editing consistency using CLIP direction similarity metric for texture, weather editing and object insertion."

    For texture and snow-coverage editing, the target Gaussians are constructed by inverse-projecting a 2D diffusion-edited target image. Re-rendering the composite scene from the same viewpoint used to create that target image largely reproduces the 2D diffusion editor's output. The reported editing metric, CLIP-direction similarity, then measures the 2D editor's own text-to-image alignment rather than the success of the 3D back-projection. Since Tables 2 and 4 use CLIP-dir as the headline quantitative evidence and no explicit multi-view consistency metric is reported, part of the claimed 3D editing advantage reduces to the pre-aligned 2D editing input.

full rationale

The core reconstruction claim is not circular: Table 1 and Table 3 benchmark against external methods (EmerNeRF, Mip-NeRF360, 3DGS, SUDS, etc.) on nuScenes and KITTI-360, and the LiDAR and SfM ablations are genuine comparisons of initialization strategies. The editing pipeline is training-free and does not fit parameters to the edited images, so there is no fitted-input-called-prediction pattern in the strict sense. The paper's own admitted limitation in Section 3.3.1, that rendered 3DGS depth may not match the true object surface and Eq. 14 is an unvalidated horizontal-depth equalization, is a correctness and robustness concern rather than a circularity. No load-bearing self-citation chain or imported uniqueness theorem appears: references to prior work, including the CVPR 2024 DrivingGaussian precursor, are not used to forbid alternatives or supply the main derivation. The only mild circularity is evaluative: for texture and snow editing, the target Gaussians are produced by inverse-projection of a 2D diffusion-edited image, and the CLIP-direction metric on the source viewpoint substantially reflects the 2D editor's own prompt alignment. This weakens but does not eliminate the editing evidence, since other viewpoints and the 3D particle weather effects provide independent content. Overall score 2: one minor evaluative circularity, with the central reconstruction and most of the editing mechanism remaining externally grounded.

Assumptions & free parameters 5 free parameters · 5 assumptions · 2 invented entities

The central reconstruction claims rest on standard domain assumptions about sensor calibration, annotation quality, and the transferability of 3DGS hyperparameters. The novel editing claims rest additionally on the accuracy of rendered depth for inverse projection and on the validity of CLIP-direction similarity as an evaluation metric. The paper introduces no speculative physical entities; the weather particles and foreground bank are procedural constructs with limited external validation.

free parameters (5)
  • Weather particle design parameters zeta and epsilon (Eq. 15) = not reported
    zeta scales random snow Gaussian positions; epsilon perturbs color and scale (c_k=(1,1,1)+epsilon, s_k,y=min(min(s_k,x,s_k,x),0)+epsilon as written can be negative). Chosen by hand, no sensitivity analysis.
  • Inpainting distance threshold d_thr (Eq. 21) = not reported
    Determines which Gaussians around a removed object are marked for inpainting; the value is never stated.
  • Loss weights for L_TSSIM, L_Robust, L_LiDAR (Eqs. 11-13) = not reported
    The total loss is described as the sum of three terms but no weighting coefficients are given, so the relative influence of the LiDAR prior on geometry is unspecified.
  • LLM trajectory generation configuration (GPT-4o prompt, temperature, seed) = not reported
    Eq. 20 depends on GPT-4o output; the call is non-deterministic and the configuration is unspecified, so trajectory results are not reproducible.
  • Multi-camera fusion weight omega (Eq. 5) = not reported
    Weighted averaging across surrounding views is introduced but how omega is computed or set is not explained.
assumptions (5)
  • domain assumption Accurate extrinsic calibration between the six cameras and the LiDAR sweeps
    Section 3.1.1 projects LiDAR points through [R_i^t | T_i^t] (Eq. 1); any registration error propagates into Gaussian initialization and the L_LiDAR loss.
  • domain assumption Dataset-provided 3D bounding boxes correctly identify and localize all dynamic objects
    Section 3.1.3 builds the dynamic Gaussian graph using bounding boxes and object IDs from nuScenes; misdetections would mis-assign Gaussians to static or dynamic graphs.
  • domain assumption The 3DGS differentiable rasterizer, training schedule, and densification heuristics of [14] transfer to unbounded driving scenes
    Section 3.2 adopts the renderer from [14] and Section 4.2 tunes 3DGS hyperparameters; the method inherits 3DGS's assumptions about sorting and opacity-based compositing.
  • domain assumption CLIP direction similarity is a valid proxy for editing quality and is comparable across heterogeneous tasks
    Tables 2 and 4 use CLIP-dir as the headline metric without validation against human judgment; aggregation across tasks, prompts, and scenes is not specified.
  • domain assumption Depth rendered by the reconstruction is accurate enough for inverse projection of edited 2D content
    Section 3.3.1 acknowledges surface discrepancies between 3DGS and real objects and patches them with horizontal depth equalization (Eq. 14); the patch is ad hoc and unvalidated.
invented entities (2)
  • Procedural weather particle Gaussians (raindrops, snowflakes, fog)
    purpose: Simulate meteorological effects by inserting shape and color scripted Gaussians with hand-assigned trajectories (Sec 3.3.1, Eqs. 15-16).
    Particles are defined by hand-set distributions; there is no quantitative physical validation such as comparison to real weather imagery or optical measurements.
  • 3D Gaussian foreground asset bank
    purpose: Provides vehicles, pedestrians, and animals for object insertion, built from Blender models, nuScenes sparse reconstruction, and image-based 3D generation (Sec 4.4.3).
    The bank is a dataset rather than a physical postulate, but it is not released and has no independent falsifiable handle, so evidence outside the paper is unavailable.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DrivingGaussian++: Towards Realistic Reconstruction and Editable Simulation for Surrounding Dynamic Driving Scenes." pith.science (2026). https://pith.science/paper/VBBTE4S3

@misc{pith2026250820965,
  author       = {Pith},
  title        = {Pith review of: DrivingGaussian++: Towards Realistic Reconstruction and Editable Simulation for Surrounding Dynamic Driving Scenes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VBBTE4S3}},
  note         = {Machine review of arXiv:2508.20965}
}
read the original abstract

We present DrivingGaussian++, an efficient and effective framework for realistic reconstructing and controllable editing of surrounding dynamic autonomous driving scenes. DrivingGaussian++ models the static background using incremental 3D Gaussians and reconstructs moving objects with a composite dynamic Gaussian graph, ensuring accurate positions and occlusions. By integrating a LiDAR prior, it achieves detailed and consistent scene reconstruction, outperforming existing methods in dynamic scene reconstruction and photorealistic surround-view synthesis. DrivingGaussian++ supports training-free controllable editing for dynamic driving scenes, including texture modification, weather simulation, and object manipulation, leveraging multi-view images and depth priors. By integrating large language models (LLMs) and controllable editing, our method can automatically generate dynamic object motion trajectories and enhance their realism during the optimization process. DrivingGaussian++ demonstrates consistent and realistic editing results and generates dynamic multi-view driving scenarios, while significantly enhancing scene diversity. More results and code can be found at the project site: https://xiong-creator.github.io/DrivingGaussian_plus.github.io

Figures

Figures reproduced from arXiv: 2508.20965 by the authors.

Figure 1
Figure 1. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Comparison between the traditional and our editing strategy. Our method differs from previous work in three key aspects: 1) supporting diverse multi-task 3D simulation; 2) enabling training-free controllable editing; 3) demon￾strating strong performance in complex dynamic driving environments. structs the static background incrementally, while the latter models multiple dynamic objects with a Gaussian graph. (ii) We… view at source ↗
Figure 3
Figure 3. Overview Pipeline of our method. DrivingGaussian++ facilitates the reconstruction and controllable editing of surrounding dynamic scenes in autonomous driving by leveraging a compositional, controllable 3D representation with a unified global optimization strategy. In contrast, DrivingGaussian++ employs a training-free paradigm to effectively address the challenges of existing methods in dynamic driving scene editin… view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Pipeline of our reconstruction method. Left: DrivingGaussian++ takes sequential data from multi-sensor, including multi-camera images and LiDAR, as input. Middle: To represent large-scale dynamic driving scenes, we propose Composite Gaussian Splatting, which consists o…
Figure 5
Figure 5. Figure 5: Composite Gaussian Splatting with Incremental Static 3D Gaussians and Dynamic Gaussian Graph. We adopt Composite Gaussian Splatting to decompose the whole scene into static background and dynamic foreground objects, reconstructing each part separately and integrating t…
Figure 6
Figure 6. Figure 6: Pipeline of our editing framework. Left: We separately determine the target Gaussians for diverse tasks. For texture modification, first edit images and then conduct inverse projection with depth information. For weather simulation, design the attributes and distributi…
Figure 7
Figure 7. Figure 7: Failed Object Detection Cases Simulation. We use GroundingDINO [47] as the object detection model. It struggles to completely identify vehicles in foggy weather and fails to accurately recognize textures, such as cracks. appear uneven and unrealistic, thereby compromis…
Figure 9
Figure 9. Figure 9: Qualitative comparison with EmerNeRF [36] and 3DGS [14] on dynamic reconstruction for 4D driving scenes of nuScenes. DrivingGaussian++ enables the high-quality reconstruction of dynamic objects at high speed while maintaining temporal consistency. We also implement the…
Figure 10
Figure 10. Figure 10: Trajectory Prediction with LLM. With the sky direction, trajectory prompt, and initial position, we utilize LLM to predict the future trajectory of the added car. relative positions at time step t generated by LLM, dirsky is the sky direction and des denotes the descr…
Figure 12
Figure 12. Figure 12: Editing Results of nuScenes dataset. We demonstrate the results of DrivingGaussian++ across different tasks. DrivingGaussian++ enables realistic and consistent 3D editing of texture, weather, and objects in driving scenes. Urban-NeRF [53] uses depth cues from LiDAR in…
Figure 13
Figure 13. Figure 13: Dynamic Simulation of Snow. We first add dynamic particles at each time step. Secondly, we estimate the surface normal to obtain the particle deposition position and specify the particle motion trajectory. Our method can generate snowy scenes that follow physical prin…
Figure 14
Figure 14. Figure 14: Qualitative Comparison of Weather Simulation. We compare our method with IN2N, IGS2GS, and ClimateNeRF for weather simulation on 4D driving scenes from the nuScenes dataset. DrivingGaussian++ delivers realistic and coherent weather editing while maintaining high effic…
Figure 15
Figure 15. Figure 15: Dynamic Object Insertion and Scene Integration. We generate deformable foreground objects and seamlessly insert them into the scene. (a) A rigid car is added. (b) A generated 4D excavator is integrated. (c) Multiple objects are inserted with correct occlusion. The res…
Figure 16
Figure 16. Figure 16: Multiple Objects Insertion with Foreground Bank. We insert distinct objects from the foreground bank into each scene across various viewpoints. The results exhibit strong cross-view consistency in both geometry and appearance, highlighting the robustness of our method…
Figure 17
Figure 17. Figure 17: Rendering with or w/o the Incremental Static 3D Gaussians (IS3G) and Composite Dynamic Gaussian Graph (CDGG). IS3G ensures good geometry and topological integrity for static backgrounds in large-scale driving scenes. CDGG enables the reconstruction of dynamic objects …
Figure 18
Figure 18. Figure 18: Visualization comparison using different initial￾ization methods on KITTI-360. Compared to initialization with SfM points [14], using LiDAR prior allows Gaussians to restore more accurate geometric structures in the scene. ated by the pre-trained NeRF model, and LiDAR…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

70 extracted references · 54 canonical work pages

  1. [1]

    Instruct-nerf2nerf: Editing 3d scenes with instructions,

    A. Haque, M. Tancik, A. A. Efros, A. Holynski, and A. Kanazawa, “Instruct-nerf2nerf: Editing 3d scenes with instructions,” in ICCV, 2023, pp. 19 740–19 750. 1, 2, 3, 13

    work page 2023

  2. [2]

    Infusion: Inject and attention fusion for multi concept zero-shot text-based video editing,

    A. Khandelwal, “Infusion: Inject and attention fusion for multi concept zero-shot text-based video editing,” in ICCV, 2023, pp. 3017–3026. 1, 4

    work page 2023

  3. [3]

    Gaussctrl: multi-view consistent text-driven 3d gaus- sian splatting editing,

    J. Wu, J.-W. Bian, X. Li, G. Wang, I. Reid, P . Torr, and V . A. Prisacariu, “Gaussctrl: multi-view consistent text-driven 3d gaus- sian splatting editing,” arXiv preprint arXiv:2403.08733 , 2024. 1, 3

    arXiv 2024

  4. [4]

    Gaussianeditor: Swift and con- trollable 3d editing with gaussian splatting,

    Y. Chen, Z. Chen, C. Zhang, F. Wang, X. Yang, Y. Wang, Z. Cai, L. Yang, H. Liu, and G. Lin, “Gaussianeditor: Swift and con- trollable 3d editing with gaussian splatting,” in CVPR, 2024, pp. 21 476–21 485. 1, 3

    work page 2024

  5. [5]

    Vica-nerf: View-consistency-aware 3d editing of neural radiance fields,

    J. Dong and Y.-X. Wang, “Vica-nerf: View-consistency-aware 3d editing of neural radiance fields,” NIPS, vol. 36, 2024. 1, 3

    work page 2024

  6. [6]

    Climaten- erf: Extreme weather synthesis in neural radiance field,

    Y. Li, Z.-H. Lin, D. Forsyth, J.-B. Huang, and S. Wang, “Climaten- erf: Extreme weather synthesis in neural radiance field,” in ICCV, 2023, pp. 3227–3238. 1, 2, 4, 13

    work page 2023

  7. [7]

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

    B. Mildenhall, P . P . Srinivasan, M. Tancik, J. T. Barron, R. Ra- mamoorthi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,” Communications of the ACM , pp. 99–106, 2021. 2, 12

    work page 2021

  8. [8]

    Block-nerf: Scalable large scene neural view synthesis,

    M. Tancik, V . Casser, X. Yan, S. Pradhan, B. Mildenhall, P . P . Srinivasan, J. T. Barron, and H. Kretzschmar, “Block-nerf: Scalable large scene neural view synthesis,” in CVPR, 2022, pp. 8248–8258. 2

    work page 2022

Show all 70 references
  1. [9]

    Mega-nerf: Scal- able construction of large-scale nerfs for virtual fly-throughs,

    H. Turki, D. Ramanan, and M. Satyanarayanan, “Mega-nerf: Scal- able construction of large-scale nerfs for virtual fly-throughs,” in CVPR, 2022, pp. 12 922–12 931. 2

    2022

  2. [10]

    Neural fields meet explicit geometric representations for inverse rendering of urban scenes,

    Z. Wang, T. Shen, J. Gao, S. Huang, J. Munkberg, J. Hasselgren, Z. Gojcic, W. Chen, and S. Fidler, “Neural fields meet explicit geometric representations for inverse rendering of urban scenes,” in CVPR, 2023, pp. 8370–8380. 2

    2023

  3. [11]

    Switch-nerf: Learning scene decomposi- tion with mixture of experts for large-scale neural radiance fields,

    M. Zhenxing and D. Xu, “Switch-nerf: Learning scene decomposi- tion with mixture of experts for large-scale neural radiance fields,” in The Eleventh International Conference on Learning Representations ,

  4. [12]

    Neural scene graphs for dynamic scenes,

    J. Ost, F. Mannan, N. Thuerey, J. Knodt, and F. Heide, “Neural scene graphs for dynamic scenes,” in CVPR, 2021, pp. 2856–2865. 2, 3, 12

    2021

  5. [13]

    Towards efficient neural scene graphs by learning consistency fields,

    Y. Song, C. Kong, S. Lee, N. Kwak, and J. Lee, “Towards efficient neural scene graphs by learning consistency fields,” arXiv preprint arXiv:2210.04127, 2022. 2, 3

    2022 arXiv

  6. [14]

    3D Gaus- sian splatting for real-time radiance field rendering,

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, and G. Drettakis, “3D Gaus- sian splatting for real-time radiance field rendering,” TOG, vol. 42, no. 4, pp. 1–14, 2023. 2, 3, 6, 7, 9, 10, 11, 12, 14, 15

    2023

  7. [15]

    Instruct-gs2gs: Editing 3d gaussian splats with instructions (2024),

    C. Vachha and A. Haque, “Instruct-gs2gs: Editing 3d gaussian splats with instructions (2024),” URL https://instruct-gs2gs.github.io,

    2024

  8. [16]

    Instant neural graphics primitives with a multiresolution hash encoding,

    T. M ¨uller, A. Evans, C. Schied, and A. Keller, “Instant neural graphics primitives with a multiresolution hash encoding,” TOG, vol. 41, no. 4, pp. 1–15, 2022. 2, 10, 11

    2022

  9. [17]

    Plenoxels: Radiance fields without neural net- works,

    S. Fridovich-Keil, A. Yu, M. Tancik, Q. Chen, B. Recht, and A. Kanazawa, “Plenoxels: Radiance fields without neural net- works,” in CVPR, 2022, pp. 5501–5510. 2

    2022

  10. [18]

    Fastnerf: High-fidelity neural rendering at 200fps,

    S. J. Garbin, M. Kowalski, M. Johnson, J. Shotton, and J. Valentin, “Fastnerf: High-fidelity neural rendering at 200fps,” inICCV, 2021, pp. 14 346–14 355. 2

    2021

  11. [19]

    Barf: Bundle- adjusting neural radiance fields,

    C.-H. Lin, W.-C. Ma, A. Torralba, and S. Lucey, “Barf: Bundle- adjusting neural radiance fields,” in ICCV, 2021, pp. 5741–5751. 2

    2021

  12. [20]

    Nerf–: Neural radiance fields without known camera parameters,

    Z. Wang, S. Wu, W. Xie, M. Chen, and V . A. Prisacariu, “Nerf–: Neural radiance fields without known camera parameters,” arXiv preprint arXiv:2102.07064, 2021. 2

    2021 arXiv

  13. [21]

    Nope- nerf: Optimising neural radiance field with no pose prior,

    W. Bian, Z. Wang, K. Li, J.-W. Bian, and V . A. Prisacariu, “Nope- nerf: Optimising neural radiance field with no pose prior,” in CVPR, 2023, pp. 4160–4169. 2

    2023

  14. [22]

    Nerf for outdoor scene relighting,

    V . Rudnev, M. Elgharib, W. Smith, L. Liu, V . Golyanik, and C. Theobalt, “Nerf for outdoor scene relighting,” in ECCV. Springer, 2022, pp. 615–631. 2

    2022

  15. [23]

    D- nerf: Neural radiance fields for dynamic scenes,

    A. Pumarola, E. Corona, G. Pons-Moll, and F. Moreno-Noguer, “D- nerf: Neural radiance fields for dynamic scenes,” in CVPR, 2021, pp. 10 318–10 327. 2

    2021

  16. [24]

    Hdr- nerf: High dynamic range neural radiance fields,

    X. Huang, Q. Zhang, Y. Feng, H. Li, X. Wang, and Q. Wang, “Hdr- nerf: High dynamic range neural radiance fields,” in CVPR, 2022, pp. 18 398–18 408. 2

    2022

  17. [25]

    Nerf in the wild: Neural radiance fields for unconstrained photo collections,

    R. Martin-Brualla, N. Radwan, M. S. Sajjadi, J. T. Barron, A. Doso- vitskiy, and D. Duckworth, “Nerf in the wild: Neural radiance fields for unconstrained photo collections,” in CVPR, 2021, pp. 7210–7219. 2

    2021

  18. [26]

    Mip-nerf: A multiscale represen- tation for anti-aliasing neural radiance fields,

    J. T. Barron, B. Mildenhall, M. Tancik, P . Hedman, R. Martin- Brualla, and P . P . Srinivasan, “Mip-nerf: A multiscale represen- tation for anti-aliasing neural radiance fields,” in ICCV, 2021, pp. 5855–5864. 2, 10, 11

    2021

  19. [27]

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

    J. T. Barron, B. Mildenhall, D. Verbin, P . P . Srinivasan, and P . Hed- man, “Mip-nerf 360: Unbounded anti-aliased neural radiance fields,” in CVPR, 2022, pp. 5470–5479. 2, 10, 11, 12

    2022

  20. [28]

    Grid-guided neural radiance fields for large urban scenes,

    L. Xu, Y. Xiangli, S. Peng, X. Pan, N. Zhao, C. Theobalt, B. Dai, and D. Lin, “Grid-guided neural radiance fields for large urban scenes,” in CVPR, 2023, pp. 8296–8306. 2

    2023

  21. [29]

    Streetsurf: Extending multi-view implicit surface recon- struction to street views,

    J. Guo, N. Deng, X. Li, Y. Bai, B. Shi, C. Wang, C. Ding, D. Wang, and Y. Li, “Streetsurf: Extending multi-view implicit surface recon- struction to street views,” arXiv preprint arXiv:2306.04988 , 2023. 2

    2023 arXiv

  22. [30]

    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,” in CVPR, 2023, pp. 13–23. 3

    2023

  23. [31]

    Progressively optimized local radiance fields for robust view synthesis,

    A. Meuleman, Y.-L. Liu, C. Gao, J.-B. Huang, C. Kim, M. H. Kim, and J. Kopf, “Progressively optimized local radiance fields for robust view synthesis,” in CVPR, 2023, pp. 16 539–16 548. 3

    2023

  24. [32]

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

    Z. Wu, T. Liu, L. Luo, Z. Zhong, J. Chen, H. Xiao, C. Hou, H. Lou, Y. Chen, R. Yanget al., “Mars: An instance-aware, modular and realistic simulator for autonomous driving,” arXiv preprint arXiv:2307.15058, 2023. 3

    2023 arXiv

  25. [33]

    Unisim: A neural closed-loop sensor simulator,

    Z. Yang, Y. Chen, J. Wang, S. Manivasagam, W.-C. Ma, A. J. Yang, and R. Urtasun, “Unisim: A neural closed-loop sensor simulator,” in CVPR, 2023, pp. 1389–1399. 3

    2023

  26. [34]

    S-nerf: Neural radiance fields for street views,

    Z. Xie, J. Zhang, W. Li, F. Zhang, and L. Zhang, “S-nerf: Neural radiance fields for street views,” arXiv preprint arXiv:2303.00749 ,

    arXiv

  27. [35]

    Suds: Scalable urban dynamic scenes,

    H. Turki, J. Y. Zhang, F. Ferroni, and D. Ramanan, “Suds: Scalable urban dynamic scenes,” in CVPR, 2023, pp. 12 375–12 385. 3, 10, 11, 12

    2023

  28. [36]

    Emernerf: Emergent spatial- temporal scene decomposition via self-supervision,

    J. Yang, B. Ivanovic, O. Litany, X. Weng, S. W. Kim, B. Li, T. Che, D. Xu, S. Fidler, M. Pavone et al. , “Emernerf: Emergent spatial- temporal scene decomposition via self-supervision,” arXiv preprint arXiv:2311.02077, 2023. 3, 9, 10, 11, 12 17

    2023 arXiv

  29. [37]

    De- formable 3d gaussians for high-fidelity monocular dynamic scene reconstruction,

    Z. Yang, X. Gao, W. Zhou, S. Jiao, Y. Zhang, and X. Jin, “De- formable 3d gaussians for high-fidelity monocular dynamic scene reconstruction,” arXiv preprint arXiv:2309.13101 , 2023. 3

    2023 arXiv

  30. [38]

    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,” arXiv preprint arXiv:2310.08528 , 2023. 3

    2023 arXiv

  31. [39]

    Dynamic 3d gaussians: Tracking by persistent dynamic view synthesis,

    J. Luiten, G. Kopanas, B. Leibe, and D. Ramanan, “Dynamic 3d gaussians: Tracking by persistent dynamic view synthesis,” arXiv preprint arXiv:2308.09713, 2023. 3

    2023 arXiv

  32. [40]

    Dreameditor: Text- driven 3d scene editing with neural fields,

    J. Zhuang, C. Wang, L. Lin, L. Liu, and G. Li, “Dreameditor: Text- driven 3d scene editing with neural fields,” in SIGGRAPH Asia 2023 Conference Papers , 2023, pp. 1–10. 3

    2023

  33. [41]

    Hifa: High-fidelity text-to- 3d generation with advanced diffusion guidance,

    J. Zhu, P . Zhuang, and S. Koyejo, “Hifa: High-fidelity text-to- 3d generation with advanced diffusion guidance,” arXiv preprint arXiv:2305.18766, 2023. 3

    2023 arXiv

  34. [42]

    Structure-from-motion revis- ited,

    J. L. Schonberger and J.-M. Frahm, “Structure-from-motion revis- ited,” in CVPR, 2016, pp. 4104–4113. 4, 14

    2016

  35. [43]

    R3d3: Dense 3d reconstruction of dynamic scenes from multiple cameras,

    A. Schmied, T. Fischer, M. Danelljan, M. Pollefeys, and F. Yu, “R3d3: Dense 3d reconstruction of dynamic scenes from multiple cameras,” in ICCV, 2023, pp. 3216–3226. 5

    2023

  36. [44]

    Moisst: Multi-modal optimization of implicit scene for spatiotemporal calibration,

    Q. Herau, N. Piasco, M. Bennehar, L. Rold ˜ao, D. Tsishkou, C. Migniot, P . Vasseur, and C. Demonceaux, “Moisst: Multi-modal optimization of implicit scene for spatiotemporal calibration,” arXiv preprint arXiv:2303.03056 , 2023. 5

    2023 arXiv

  37. [45]

    Grounded sam: Assembling open-world models for diverse visual tasks,

    T. Ren, S. Liu, A. Zeng, J. Lin, K. Li, H. Cao, J. Chen, X. Huang, Y. Chen, F. Yan, Z. Zeng, H. Zhang, F. Li, J. Yang, H. Li, Q. Jiang, and L. Zhang, “Grounded sam: Assembling open-world models for diverse visual tasks,” arXiv preprint arXiv:2401.14159 , 2024. 6

    2024 arXiv

  38. [46]

    S3im: Stochastic structural similarity and its unreason- able effectiveness for neural fields,

    Z. Xie, X. Yang, Y. Yang, Q. Sun, Y. Jiang, H. Wang, Y. Cai, and M. Sun, “S3im: Stochastic structural similarity and its unreason- able effectiveness for neural fields,” in ICCV, 2023, pp. 18 024– 18 034. 7

    2023

  39. [47]

    Grounding dino: Marrying dino with grounded pre-training for open-set object detection,

    S. Liu, Z. Zeng, T. Ren, F. Li, H. Zhang, J. Yang, C. Li, J. Yang, H. Su, J. Zhu et al. , “Grounding dino: Marrying dino with grounded pre-training for open-set object detection,” arXiv preprint arXiv:2303.05499, 2023. 7, 8

    2023 arXiv

  40. [48]

    Depth anything: Unleashing the power of large-scale unlabeled data,

    L. Yang, B. Kang, Z. Huang, X. Xu, J. Feng, and H. Zhao, “Depth anything: Unleashing the power of large-scale unlabeled data,” in CVPR, 2024, pp. 10 371–10 381. 9

    2024

  41. [49]

    Design of an image edge detection filter using the sobel operator,

    N. Kanopoulos, N. Vasanthavada, and R. L. Baker, “Design of an image edge detection filter using the sobel operator,” IEEE Journal of solid-state circuits , vol. 23, no. 2, pp. 358–367, 1988. 9

    1988

  42. [50]

    Editable scene simulation for autonomous driving via collaborative llm-agents,

    Y. Wei, Z. Wang, Y. Lu, C. Xu, C. Liu, H. Zhao, S. Chen, and Y. Wang, “Editable scene simulation for autonomous driving via collaborative llm-agents,” in CVPR, 2024, pp. 15 077–15 087. 9, 14

    2024

  43. [51]

    Gpt-4 technical report,

    J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkat et al., “Gpt-4 technical report,” arXiv preprint arXiv:2303.08774 , 2023. 9, 10

    2023 arXiv

  44. [52]

    Arshad- owgan: Shadow generative adversarial network for augmented reality in single light scenes,

    D. Liu, C. Long, H. Zhang, H. Yu, X. Dong, and C. Xiao, “Arshad- owgan: Shadow generative adversarial network for augmented reality in single light scenes,” in CVPR, 2020, pp. 8139–8148. 10

    2020

  45. [53]

    Urban radiance fields,

    K. Rematas, A. Liu, P . P . Srinivasan, J. T. Barron, A. Tagliasacchi, T. Funkhouser, and V . Ferrari, “Urban radiance fields,” in CVPR, 2022, pp. 12 932–12 942. 10, 11

    2022

  46. [54]

    nuscenes: A multimodal dataset for autonomous driving,

    H. Caesar, V . Bankiti, A. H. Lang, S. Vora, V . E. Liong, Q. Xu, A. Krishnan, Y. Pan, G. Baldan, and O. Beijbom, “nuscenes: A multimodal dataset for autonomous driving,” in CVPR, 2020. 10

    2020

  47. [55]

    Kitti-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d,

    Y. Liao, J. Xie, and A. Geiger, “Kitti-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d,” IEEE T ransactions on Pattern Analysis and Machine Intelligence , vol. 45, no. 3, pp. 3292–3310, 2022. 10

    2022

  48. [56]

    Point-nerf: Point-based neural radiance fields,

    Q. Xu, Z. Xu, J. Philip, S. Bi, Z. Shu, K. Sunkavalli, and U. Neu- mann, “Point-nerf: Point-based neural radiance fields,” in CVPR, 2022, pp. 5438–5448. 12

    2022

  49. [57]

    Urban radiance field representation with deformable neural mesh primi- tives,

    F. Lu, Y. Xu, G. Chen, H. Li, K.-Y. Lin, and C. Jiang, “Urban radiance field representation with deformable neural mesh primi- tives,” in ICCV, 2023, pp. 465–476. 12

    2023

  50. [58]

    Dreamgaussian4d: Generative 4d gaussian splatting,

    J. Ren, L. Pan, J. Tang, C. Zhang, A. Cao, G. Zeng, and Z. Liu, “Dreamgaussian4d: Generative 4d gaussian splatting,” arXiv preprint arXiv:2312.17142 , 2023. 13, 14

    2023 arXiv

  51. [59]

    Paint by example: Exemplar-based image editing with diffusion models,

    B. Yang, S. Gu, B. Zhang, T. Zhang, X. Chen, X. Sun, D. Chen, and F. Wen, “Paint by example: Exemplar-based image editing with diffusion models,” in CVPR, 2023, pp. 18 381–18 391. 13, 14

    2023

  52. [60]

    Anydoor: Zero-shot object-level image customization,

    X. Chen, L. Huang, Y. Liu, Y. Shen, D. Zhao, and H. Zhao, “Anydoor: Zero-shot object-level image customization,” in CVPR, 2024, pp. 6593–6602. 13, 14

    2024

  53. [61]

    High-resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P . Esser, and B. Ommer, “High-resolution image synthesis with latent diffusion models,” in CVPR, 2022, pp. 10 684–10 695. 13, 14

    2022

  54. [62]

    Instructdiffusion: A generalist modeling interface for vision tasks,

    Z. Geng, B. Yang, T. Hang, C. Li, S. Gu, T. Zhang, J. Bao, Z. Zhang, H. Li, H. Hu et al. , “Instructdiffusion: A generalist modeling interface for vision tasks,” in CVPR, 2024, pp. 12 709–12 720. 13, 14

    2024

  55. [63]

    Instructpix2pix: Learning to follow image editing instructions,

    T. Brooks, A. Holynski, and A. A. Efros, “Instructpix2pix: Learning to follow image editing instructions,” in CVPR, 2023, pp. 18 392– 18 402. 13, 14

    2023

  56. [64]

    Towards under- standing cross and self-attention in stable diffusion for text-guided image editing,

    B. Liu, C. Wang, T. Cao, K. Jia, and J. Huang, “Towards under- standing cross and self-attention in stable diffusion for text-guided image editing,” in CVPR, 2024, pp. 7817–7826. 13, 14

    2024

  57. [65]

    Ultraedit: Instruction-based fine-grained image editing at scale,

    H. Zhao, X. Ma, L. Chen, S. Si, R. Wu, K. An, P . Yu, M. Zhang, Q. Li, and B. Chang, “Ultraedit: Instruction-based fine-grained image editing at scale,” arXiv preprint arXiv:2407.05282 , 2024. 14

    2024 arXiv

  58. [66]

    Resolution-robust large mask inpainting with fourier convolutions,

    R. Suvorov, E. Logacheva, A. Mashikhin, A. Remizova, A. Ashukha, A. Silvestrov, N. Kong, H. Goka, K. Park, and V . Lem- pitsky, “Resolution-robust large mask inpainting with fourier convolutions,” in Proceedings of the IEEE/CVF winter conference on applications of computer vis...

    2022

  59. [67]

    Spin-nerf: Multiview segmentation and perceptual inpainting with neural radiance fields,

    A. Mirzaei, T. Aumentado-Armstrong, K. G. Derpanis, J. Kelly, M. A. Brubaker, I. Gilitschenski, and A. Levinshtein, “Spin-nerf: Multiview segmentation and perceptual inpainting with neural radiance fields,” in CVPR, 2023, pp. 20 669–20 679. 14

    2023

  60. [68]

    Splatter image: Ultra-fast single-view 3d reconstruction,

    S. Szymanowicz, C. Rupprecht, and A. Vedaldi, “Splatter image: Ultra-fast single-view 3d reconstruction,” in CVPR, 2024, pp. 10 208–10 217. 14

    2024

  61. [69]

    Segment everything everywhere all at once,

    X. Zou, J. Yang, H. Zhang, F. Li, L. Li, J. Gao, and Y. J. Lee, “Segment everything everywhere all at once,” arXiv preprint arXiv:2304.06718, 2023. 14

    2023 arXiv

  62. [70]

    Dreamgaussian: Generative gaussian splatting for efficient 3d content creation,

    J. Tang, J. Ren, H. Zhou, Z. Liu, and G. Zeng, “Dreamgaussian: Generative gaussian splatting for efficient 3d content creation,” arXiv preprint arXiv:2309.16653 , 2023. 14

    2023 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.