Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

GS-LiDAR: Generating Realistic LiDAR Point Clouds with Panoramic Gaussian Splatting

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Gaussian splatting renders realistic LiDAR point clouds from novel views faster and more accurately than NeRF-based simulators.

desk verdict A genuinely new and well-engineered Gaussian-splatting approach to LiDAR novel view synthesis, with a clean ray-splat intersection and real speedups, but the 'significantly surpasses' claim rests on a thin evaluation that likely uses a single scene per dataset with no variance. read the letter →

arxiv 2501.13971 v2 pith:CA4EXISH submitted 2025-01-22 cs.CV cs.GReess.IV

Junzhe Jiang , Chun Gu , Yurui Chen , Li Zhang This is my paper · ORCID
classification cs.CVcs.GReess.IV
keywords LiDARnovelviewsynthesisGaussiansplattingautonomousdrivingsimulationpanoramicrenderingpointcloudgenerationperiodicvibrationray-splatintersectionintensityandray-dropmodeling
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

GS-LiDAR aims to replace slow NeRF-based LiDAR simulators with a Gaussian-splatting representation that renders full 360-degree LiDAR views directly. The paper claims that 2D Gaussian primitives with a periodic vibration model reconstruct both static surfaces and moving vehicles and pedestrians, and that an explicit ray-splat intersection makes panoramic depth rendering geometrically accurate. On the KITTI-360 and nuScenes driving benchmarks, it reports lower point-cloud, depth, and intensity errors than the leading NeRF-based method LiDAR4D while training 1.67 times faster and rendering 31 times faster. If this holds, simulation for autonomous driving becomes fast enough for large-scale closed-loop training without hand-built virtual worlds.

What carries the argument

The load-bearing object is a 2D Gaussian primitive: a flat elliptic disk in space with position, orientation, scale, opacity, plus spherical-harmonic coefficients for intensity and ray-drop. Time enters through a periodic vibration model that shifts the disk's center along a learned direction with a sine of period $l$ and decays its opacity with a Gaussian over time, so each primitive can represent a slice of a moving surface. The rendering pass is panoramic Gaussian splatting: each LiDAR ray is written as the intersection of two orthogonal planes, and the ray-splat intersection $(u,v)$ is obtained by solving two linear equations, yielding an exact depth with no Jacobian approximation. Depth, intensity, and ray-drop maps are alpha-blended in tile-sorted order, supervised with both mean and median depth, and the ray-drop map is refined by a U-Net. This combination is what the paper claims carries the accuracy and speed gains.

What would settle it

Train GS-LiDAR on a driving sequence containing a hard brake, a lane change, or a pedestrian reversing, using the paper's 51-frame protocol, then measure Chamfer distance and depth RMSE on held-out timestamps; if errors climb sharply relative to the reported KITTI-360 and nuScenes numbers while static-scene errors stay low, the periodic-vibration relay is the failing component. A cheaper check is to grid-search the cycle-length hyperparameter $l$ on one KITTI-360 dynamic sequence: if results swing widely with $l$, the method depends on a tuned prior the paper does not disclose.

Watch

Extended reading notes

Core claim

The central discovery claimed is that LiDAR novel view synthesis does not need neural radiance fields. GS-LiDAR represents a driving scene as a set of 2D Gaussian primitives, each carrying spherical-harmonic coefficients for intensity and ray-drop probability, and renders them onto a panoramic range map by explicitly intersecting every LiDAR ray with each Gaussian disk. Because the intersection is computed by solving two plane equations rather than by approximating a Jacobian, the rendered depth is view-consistent and geometrically accurate. A periodic vibration model moves each Gaussian's center and opacity over time, so static and dynamic objects share one representation. Supervised by ground-truth depth, intensity, and ray-drop maps, with a U-Net refining ray-drop, the method reports depth RMSE reductions of about 10.7 to 13.1 percent over LiDAR4D together with large speedups. The claim is that this makes explicit Gaussian splatting, rather than NeRF, the practical basis for LiDAR simulation.

Load-bearing premise

The paper's dynamic-scene claim rests on assuming every moving object can be represented as Gaussians that vibrate back and forth with one shared, fixed cycle length, with complicated motions stitched from chains of these vibrations; the paper does not report that cycle length or test how well the stitching handles braking, lane changes, or reversing.

Editorial extensions

If this is right

  • LiDAR novel-view synthesis becomes fast enough for practical use: about 11 frames per second on a single A6000 GPU, versus 0.3 fps for LiDAR4D.
  • Autonomous-driving perception can be trained on simulated point clouds that include intensity and realistic dropout, not just geometry, since both are rendered explicitly.
  • Dynamic scenes can be modeled without separate object detection or tracking, because vibrating Gaussian primitives handle moving and static elements uniformly.
  • The same panoramic ray-splat rendering should extend to any spinning LiDAR configuration, given the beam count and vertical field of view, as the Waymo experiments suggest.
  • Because the pipeline is differentiable, the representation can be optimized not only to match recorded LiDAR but also end-to-end for downstream driving tasks.

Reading between the lines

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

  • The periodic vibration prior is the fragile part: in scenes with braking, lane changes, or reversing pedestrians, a single scene-level cycle length is likely a poor fit, and the method would need to lean on chains of Gaussians. A natural test is to replace the sine motion with per-Gaussian polynomial trajectories or to learn a cycle length per primitive and see whether the reported gains persist.
  • The explicit ray-splat intersection solves for the hit point of a beam against a disk in closed form, so the same renderer should transfer to other beam-based sensors such as radar or sonar whenever their beam geometry can be expressed as two plane equations; the paper does not explore this.
  • A joint Gaussian representation for cameras and LiDAR is an open next step: since 3DGS already models RGB appearance and GS-LiDAR models depth, intensity, and dropout, a shared primitive set could produce synchronized image and point-cloud novel views in real time, which this paper does not address.
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

4 major / 5 minor

Summary. The paper proposes GS-LiDAR, a Gaussian-splatting framework for LiDAR novel view synthesis. The scene is represented by 2D Gaussian primitives with periodic vibration attributes for dynamics, and a panoramic rendering procedure with explicit ray-splat intersection produces depth, intensity, and ray-drop maps. These maps are supervised with ground-truth LiDAR range and intensity data, with additional losses for depth distortion, normal consistency, and chamfer distance, plus a U-Net refinement step for ray-drop. Experiments on KITTI-360, nuScenes, and Waymo report lower chamfer distance and depth RMSE than LiDAR4D and large training/rendering speedups. The paper concludes that GS-LiDAR significantly surpasses previous NeRF-based approaches in both computational speed and simulation quality on KITTI-360 and nuScenes.

Significance. If the empirical claims are robust, GS-LiDAR is a meaningful contribution: it replaces costly NeRF ray marching with explicit Gaussian splatting, introduces a clean ray-splat intersection that avoids the Jacobian approximation of 3DGS for panoramic LiDAR projection, and reaches 11 fps rendering. The intensity and ray-drop SH attributes, together with U-Net refinement, target realistic LiDAR-specific effects, and the code is promised publicly. The core derivation in Section 3 is coherent and the ablations show that each component contributes. However, the current evidence for dataset-level superiority is limited by an apparently single-scene evaluation without variance, at least one apparent copy-paste error in the appendix, and an unreported key hyperparameter. These issues are fixable, but they currently prevent the paper's strongest claims from being fully supported.

major comments (4)
  1. [§4.1, Tables 1–3, §5] The experimental setup states that 51 consecutive frames are selected as a single scene with 4 held-out samples, but the paper never states how many scenes are used for KITTI-360 or nuScenes. The table captions ("Static Scene Sequence" for Table 1, and similar singular framing) suggest one sequence per dataset. No error bars, multiple seeds, or per-scene breakdowns are reported. The conclusion's claim of significant superiority "on the KITTI-360 and nuScenes datasets" is therefore an extrapolation from a small number of streets. Please either evaluate additional scenes and report variance, or explicitly limit the claims to the evaluated sequences.
  2. [Appendix A.2, Table 5] The Waymo results in Table 5 are numerically identical to the GS-LiDAR row in Table 3 for nuScenes (CD 0.2382, F-score 0.9055, RMSE 5.8925, and all depth/intensity metrics match exactly). This appears to be a copy-paste error. Because the Waymo experiments are cited as evidence of generalization and scalability, the correct numbers must be provided; as written, this supporting evidence is invalid.
  3. [§3.2, Eq. (3), §4.1] The periodic vibration model uses a scene-level cycle length l, described as a hyperparameter that serves as a prior for the scene, but its value is never reported in the implementation details and no sensitivity analysis is given. The claim that multiple Gaussians can relay to represent arbitrary non-periodic motion (braking, lane changes, pedestrians reversing) is not analyzed or measured. Since dynamic-scene quality is a central contribution and the motion model is the main assumption that differs from static reconstruction, the paper should report the value of l, test sensitivity to it, and discuss the failure modes of the sinusoidal model for non-periodic motion.
  4. [Tables 1–3, §5] The stated superiority in "simulation quality" is not uniform across the reported metrics. On KITTI-360, LiDAR4D has a higher point-cloud F-score in both static (0.9264 vs 0.9236) and dynamic (0.9272 vs 0.9231) settings. On nuScenes, GS-LiDAR has worse intensity LPIPS, SSIM, and PSNR than LiDAR4D (0.0627/0.7291/27.7420 vs 0.0459/0.7498/27.7977). The conclusion needs to either explain these trade-offs or qualify the superiority claim to the specific metrics where the gains hold, rather than making an unqualified claim about overall simulation quality.
minor comments (5)
  1. [Figure 1, Table 1] Figure 1's caption reports CD 0.0786 for GS-LiDAR, which does not match the static-scene CD 0.0847 in Table 1; please clarify which scene or condition Figure 1 reports.
  2. [§1, §4.3] The Introduction reports an 11.5% RMSE reduction on dynamic KITTI-360, while §4.3 reports 11.4%; Table 2 implies 11.47%, so the two numbers should be reconciled.
  3. [§1, Figure 1] The text states a 31x rendering speedup over LiDAR4D, but the numbers in Figure 1 (11 fps vs 0.3 fps) imply a 36.7x speedup; please clarify how the 31x figure is computed.
  4. [§1, §2] The claim that NeRF and its variants are "designed for symmetrical scenes" is unclear; presumably "static" or "bounded" scenes is intended, and the wording should be corrected.
  5. [§3.4] The U-Net used for ray-drop refinement is not described in terms of architecture, input normalization, or training schedule after Gaussian optimization, which makes the ray-drop component difficult to reproduce.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: held-out supervised evaluation with only a minor same-group component citation for periodic vibration.

full rationale

The derivation chain is self-contained with respect to the headline claims. GS-LiDAR optimizes Gaussian attributes, SH intensity/ray-drop coefficients, and a U-Net against ground-truth range/intensity/ray-drop maps from training frames (Eqs. 15-17, 19-21), then evaluates on 4 held-out frames per 51-frame scene (Sec. 4.1). No evaluation metric is a training target by construction: chamfer distance, F-score, RMSE, MedAE, LPIPS, SSIM, and PSNR are computed on held-out rendered maps against their GT counterparts. The periodic-vibration model (Eq. 3) is adopted from same-group prior work (Chen et al. 2023a) rather than derived, and the paper includes an ablation (Table 4, 'w/o periodic vibration') showing its contribution; this is a component choice backed by internal evidence and external comparisons, not a uniqueness theorem or a renamed input. The ray-drop U-Net refinement follows NeRF-LiDAR (same-group citation) but is trained with the same GT BCE loss on training frames, again a post-processing fit rather than a prediction of a fitted value. No equation in the paper equals its own input or reduces to a fitted parameter relabeled as a prediction. Score 2 reflects only the presence of same-group component citations (PVG, NeRF-LiDAR) that are not load-bearing in a circular sense.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central geometry claim rests on standard Gaussian-splatting assumptions and a domain-specific sensor model. The main free choices are the cycle length l for periodic vibration (unreported), the hand-set loss weights, and the initial number of Gaussians. No new physical entities are introduced; the periodic vibration is a mathematical device. The most fragile assumption is the sinusoidal motion model for dynamic scenes.

free parameters (3)
  • cycle length l for periodic vibration = not reported
    Hyperparameter controlling the period of sinusoidal motion in Eq. 3. The paper does not state its value or how it is chosen, yet it directly controls dynamic modeling. It is set by hand as a scene prior.
  • loss weights (lambda_d, lambda_int, lambda_drop, lambda_dist, lambda_n, lambda_ch) = 10, 0.05, 0.05, 0.1, 0.1, 0.1
    Chosen by hand; no sensitivity analysis is provided. The reported quantitative results depend on these settings.
  • initial number of Gaussian primitives = 1,000,000
    Randomly sampled from LiDAR points for initialization; no justification or ablation is given for this quantity.
assumptions (4)
  • domain assumption LiDAR rays emanate from a single origin with fixed angular resolution; the sensor model in Eq. 8 is exact.
    Used throughout Section 3.3 for panoramic rendering. Real LiDAR has beam divergence, multiple returns, and intensity-dependent range noise, which are ignored.
  • domain assumption Surfaces in driving scenes can be represented as locally flat 2D Gaussian disks whose tangent planes approximate the scene geometry.
    Inherited from 2DGS and used in Section 3.2. This fails for thin structures (leaves, poles) and sharp edges not aligned with the splats.
  • domain assumption Object motion over the training window is approximately sinusoidal with a single period l for all Gaussians.
    Equation 3 in Section 3.2. Complex motion must be assembled from multiple vibrating Gaussians, and no proof or analysis is given for the sufficiency of this model.
  • domain assumption The per-scene U-Net for ray-drop refinement transfers from training viewpoints to novel viewpoints.
    Section 3.4. The U-Net is optimized on training views only, so its generalization to held-out views is an unverified assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GS-LiDAR: Generating Realistic LiDAR Point Clouds with Panoramic Gaussian Splatting." pith.science (2026). https://pith.science/paper/CA4EXISH

@misc{pith2026250113971,
  author       = {Pith},
  title        = {Pith review of: GS-LiDAR: Generating Realistic LiDAR Point Clouds with Panoramic Gaussian Splatting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CA4EXISH}},
  note         = {Machine review of arXiv:2501.13971}
}
read the original abstract

LiDAR novel view synthesis (NVS) has emerged as a novel task within LiDAR simulation, offering valuable simulated point cloud data from novel viewpoints to aid in autonomous driving systems. However, existing LiDAR NVS methods typically rely on neural radiance fields (NeRF) as their 3D representation, which incurs significant computational costs in both training and rendering. Moreover, NeRF and its variants are designed for symmetrical scenes, making them ill-suited for driving scenarios. To address these challenges, we propose GS-LiDAR, a novel framework for generating realistic LiDAR point clouds with panoramic Gaussian splatting. Our approach employs 2D Gaussian primitives with periodic vibration properties, allowing for precise geometric reconstruction of both static and dynamic elements in driving scenarios. We further introduce a novel panoramic rendering technique with explicit ray-splat intersection, guided by panoramic LiDAR supervision. By incorporating intensity and ray-drop spherical harmonic (SH) coefficients into the Gaussian primitives, we enhance the realism of the rendered point clouds. Extensive experiments on KITTI-360 and nuScenes demonstrate the superiority of our method in terms of quantitative metrics, visual quality, as well as training and rendering efficiency.

Figures

Figures reproduced from arXiv: 2501.13971 by the authors.

Figure 1
Figure 1. GS-LiDAR achieves superior LiDAR simulation quality for novel view synthesis while maintaining fast training and rendering speed. ABSTRACT LiDAR novel view synthesis (NVS) has emerged as a novel task within LiDAR simulation, offering valuable simulated point cloud data from novel viewpoints to aid in autonomous driving systems. However, existing LiDAR NVS methods typically rely on neural radiance fields (NeRF) as th… view at source ↗
Figure 2
Figure 2. Overview of the GS-LiDAR framework: GS-LiDAR is based on 2D Gaussian primitives with periodic vibration properties, allowing for dynamic modeling of position and opacity along with accurate geometry. At a given timestamp, Gaussians query their states and utilize the proposed panoramic Gaussian splatting technique to render panoramic maps of depth, ray-drop, and intensity. For each ray and Gaussian primitive, we calc… view at source ↗
Figure 3
Figure 3. Our LiDAR coordinate system and two ways of depth rendering. The mean depth refers [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Panoramic Gaussian rasterization details. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Comparison of 3D LiDAR point cloud. GS-LiDAR produces a more cohesive LiDAR point cloud compared to LiDAR-NeRF (Tao et al., 2023) and LiDAR4D (Zheng et al., 2024) [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Comparison of the rendered depth map with competitors. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Comparison of the rendered intensity map with competitors. [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Comparison of w/ and w/o ray-splat intersection. A more detailed discussion is provided [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Ray-splat intersection v.s. 3D Gaussian splatting. [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]
Figure 10
Figure 10. Figure 10: Comparison of the rendered depth and intensity map with ground truth. The first two [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]
Figure 11
Figure 11. Figure 11: The visualization of LiDAR point clouds in three-dimensional space. [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Neural LiDAR Bundle Adjustment

    cs.RO 2026-07 conditional novelty 6.0 of 10

    Tailored volume-sampling density and a LiDAR-specific loss enable neural bundle adjustment that jointly optimizes LiDAR poses and maps better than prior BA and neural mapping baselines.

Reference graph

Works this paper leans on

54 extracted references · 39 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Barron, Ben Mildenhall, Matthew Tancik, Peter Hedman, Ricardo Martin-Brualla, and Pratul P

    Jonathan T. Barron, Ben Mildenhall, Matthew Tancik, Peter Hedman, Ricardo Martin-Brualla, and Pratul P. Srinivasan. Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields. In ICCV, 2021

    work page 2021

  3. [3]

    Barron, Ben Mildenhall, Dor Verbin, Pratul P

    Jonathan T. Barron, Ben Mildenhall, Dor Verbin, Pratul P. Srinivasan, and Peter Hedman. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. In CVPR, 2022

    work page 2022

  4. [4]

    Zip-nerf: Anti-aliased grid-based neural radiance fields

    Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P Srinivasan, and Peter Hedman. Zip-nerf: Anti-aliased grid-based neural radiance fields. In ICCV, 2023

    work page 2023

  5. [5]

    nuscenes: A multimodal dataset for autonomous driving

    Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh Vora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, and Oscar Beijbom. nuscenes: A multimodal dataset for autonomous driving. In CVPR, 2020

    2020

  6. [6]

    Tensorf: Tensorial radiance fields

    Anpei Chen, Zexiang Xu, Andreas Geiger, Jingyi Yu, and Hao Su. Tensorf: Tensorial radiance fields. In ECCV, 2022

    work page 2022

  7. [7]

    Periodic vibration gaussian: Dynamic urban scene reconstruction and real-time rendering

    Yurui Chen, Chun Gu, Junzhe Jiang, Xiatian Zhu, and Li Zhang. Periodic vibration gaussian: Dynamic urban scene reconstruction and real-time rendering. arXiv preprint, 2023 a

    work page 2023

  8. [8]

    Neurbf: A neural fields representation with adaptive radial basis functions

    Zhang Chen, Zhong Li, Liangchen Song, Lele Chen, Jingyi Yu, Junsong Yuan, and Yi Xu. Neurbf: A neural fields representation with adaptive radial basis functions. In ICCV, 2023 b

    work page 2023

Show all 54 references
  1. [9]

    Mobilenerf: Exploiting the polygon rasterization pipeline for efficient neural field rendering on mobile architectures

    Zhiqin Chen, Thomas Funkhouser, Peter Hedman, and Andrea Tagliasacchi. Mobilenerf: Exploiting the polygon rasterization pipeline for efficient neural field rendering on mobile architectures. In CVPR, 2023 c

    2023

  2. [10]

    Carla: An open urban driving simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. Carla: An open urban driving simulator. In CoRL, 2017

    2017

  3. [11]

    A point set generation network for 3d object reconstruction from a single image

    Haoqiang Fan, Hao Su, and Leonidas J Guibas. A point set generation network for 3d object reconstruction from a single image. In CVPR, 2017

    2017

  4. [12]

    Fast dynamic radiance fields with time-aware neural voxels

    Jiemin Fang, Taoran Yi, Xinggang Wang, Lingxi Xie, Xiaopeng Zhang, Wenyu Liu, Matthias Nie ner, and Qi Tian. Fast dynamic radiance fields with time-aware neural voxels. In SIGGRAPH Asia, 2022

    2022

  5. [13]

    Plenoxels: Radiance fields without neural networks

    Sara Fridovich-Keil, Alex Yu, Matthew Tancik, Qinhong Chen, Benjamin Recht, and Angjoo Kanazawa. Plenoxels: Radiance fields without neural networks. In CVPR, 2022

    2022

  6. [14]

    K-planes: Explicit radiance fields in space, time, and appearance

    Sara Fridovich-Keil, Giacomo Meanti, Frederik Rahb k Warburg, Benjamin Recht, and Angjoo Kanazawa. K-planes: Explicit radiance fields in space, time, and appearance. In CVPR, 2023

    2023

  7. [15]

    Relightable 3d gaussian: Real-time point cloud relighting with brdf decomposition and ray tracing

    Jian Gao, Chun Gu, Youtian Lin, Hao Zhu, Xun Cao, Li Zhang, and Yao Yao. Relightable 3d gaussian: Real-time point cloud relighting with brdf decomposition and ray tracing. In ECCV, 2024

    2024

  8. [16]

    Learning to simulate realistic lidars

    Beno \^ t Guillard, Sai Vemprala, Jayesh K Gupta, Ondrej Miksik, Vibhav Vineet, Pascal Fua, and Ashish Kapoor. Learning to simulate realistic lidars. In IROS, 2022

    2022

  9. [17]

    Baking neural radiance fields for real-time view synthesis

    Peter Hedman, Pratul P Srinivasan, Ben Mildenhall, Jonathan T Barron, and Paul Debevec. Baking neural radiance fields for real-time view synthesis. In ICCV, 2021

    2021

  10. [18]

    Tri-miprf: Tri-mip representation for efficient anti-aliasing neural radiance fields

    Wenbo Hu, Yuling Wang, Lin Ma, Bangbang Yang, Lin Gao, Xiao Liu, and Yuewen Ma. Tri-miprf: Tri-mip representation for efficient anti-aliasing neural radiance fields. In ICCV, 2023

    2023

  11. [19]

    2d gaussian splatting for geometrically accurate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically accurate radiance fields. In ACM SIGGRAPH, 2024

    2024

  12. [20]

    Neural kernel surface reconstruction

    Jiahui Huang, Zan Gojcic, Matan Atzmon, Or Litany, Sanja Fidler, and Francis Williams. Neural kernel surface reconstruction. In CVPR, 2023

    2023

  13. [21]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk \"u hler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM TOG, 2023

    2023

  14. [22]

    Design and use paradigms for gazebo, an open-source multi-robot simulator

    Nathan Koenig and Andrew Howard. Design and use paradigms for gazebo, an open-source multi-robot simulator. In IROS, 2004

    2004

  15. [23]

    Pcgen: Point cloud generator for lidar simulation

    Chenqi Li, Yuan Ren, and Bingbing Liu. Pcgen: Point cloud generator for lidar simulation. In ICRA, 2023

    2023

  16. [24]

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

    Yiyi Liao, Jun Xie, and Andreas Geiger. Kitti-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d. IEEE TPAMI, 2022

    2022

  17. [25]

    Neural sparse voxel fields

    Lingjie Liu, Jiatao Gu, Kyaw Zaw Lin, Tat-Seng Chua, and Christian Theobalt. Neural sparse voxel fields. NeurIPS, 2020

    2020

  18. [26]

    Lidarsim: Realistic lidar simulation by leveraging the real world

    Sivabalan Manivasagam, Shenlong Wang, Kelvin Wong, Wenyuan Zeng, Mikita Sazanovich, Shuhan Tan, Bin Yang, Wei-Chiu Ma, and Raquel Urtasun. Lidarsim: Realistic lidar simulation by leveraging the real world. In CVPR, 2020

    2020

  19. [27]

    Nerf: Representing scenes as neural radiance fields for view synthesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. In ECCV, 2020

    2020

  20. [28]

    Instant neural graphics primitives with a multiresolution hash encoding

    Thomas M \"u ller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding. ACM TOG, 2022

    2022

  21. [29]

    D-nerf: Neural radiance fields for dynamic scenes

    Albert Pumarola, Enric Corona, Gerard Pons-Moll, and Francesc Moreno-Noguer. D-nerf: Neural radiance fields for dynamic scenes. In CVPR, 2021

    2021

  22. [30]

    Kilonerf: Speeding up neural radiance fields with thousands of tiny mlps

    Christian Reiser, Songyou Peng, Yiyi Liao, and Andreas Geiger. Kilonerf: Speeding up neural radiance fields with thousands of tiny mlps. In ICCV, 2021

    2021

  23. [31]

    Merf: Memory-efficient radiance fields for real-time view synthesis in unbounded scenes

    Christian Reiser, Rick Szeliski, Dor Verbin, Pratul Srinivasan, Ben Mildenhall, Andreas Geiger, Jon Barron, and Peter Hedman. Merf: Memory-efficient radiance fields for real-time view synthesis in unbounded scenes. ACM TOG, 2023

    2023

  24. [32]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In MICCAI, 2015

    2015

  25. [33]

    Airsim: High-fidelity visual and physical simulation for autonomous vehicles

    Shital Shah, Debadeepta Dey, Chris Lovett, and Ashish Kapoor. Airsim: High-fidelity visual and physical simulation for autonomous vehicles. In Field and Service Robotics: Results of the 11th International Conference, 2018

    2018

  26. [34]

    Direct voxel grid optimization: Super-fast convergence for radiance fields reconstruction

    Cheng Sun, Min Sun, and Hwann - Tzong Chen. Direct voxel grid optimization: Super-fast convergence for radiance fields reconstruction. In CVPR, 2022

    2022

  27. [35]

    Scalability in perception for autonomous driving: Waymo open dataset

    Pei Sun, Henrik Kretzschmar, Xerxes Dotiwalla, Aurelien Chouard, Vijaysai Patnaik, Paul Tsui, James Guo, Yin Zhou, Yuning Chai, Benjamin Caine, Vijay Vasudevan, Wei Han, Jiquan Ngiam, Hang Zhao, Aleksei Timofeev, Scott Ettinger, Maxim Krivokon, Amy Gao, Aditya Joshi, Yu Zhang,...

    2020

  28. [36]

    Lidar-nerf: Novel lidar view synthesis via neural radiance fields

    Tang Tao, Longfei Gao, Guangrun Wang, Peng Chen, Dayang Hao, Xiaodan Liang, Mathieu Salzmann, and Kaicheng Yu. Lidar-nerf: Novel lidar view synthesis via neural radiance fields. arXiv preprint, 2023

    2023

  29. [37]

    Alignmif: Geometry-aligned multimodal implicit field for lidar-camera joint synthesis

    Tang Tao, Guangrun Wang, Yixing Lao, Peng Chen, Jie Liu, Liang Lin, Kaicheng Yu, and Xiaodan Liang. Alignmif: Geometry-aligned multimodal implicit field for lidar-camera joint synthesis. In CVPR, 2024

    2024

  30. [38]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE TIP, 2004

    2004

  31. [39]

    Dynamic lidar re-simulation using compositional neural fields

    Hanfeng Wu, Xingxing Zuo, Stefan Leutenegger, Or Litany, Konrad Schindler, and Shengyu Huang. Dynamic lidar re-simulation using compositional neural fields. In CVPR, 2024

    2024

  32. [40]

    Physgaussian: Physics-integrated 3d gaussians for generative dynamics

    Tianyi Xie, Zeshun Zong, Yuxing Qiu, Xuan Li, Yutao Feng, Yin Yang, and Chenfanfu Jiang. Physgaussian: Physics-integrated 3d gaussians for generative dynamics. arXiv preprint, 2023 a

    2023

  33. [41]

    S-nerf: Neural radiance fields for street views

    Ziyang Xie, Junge Zhang, Wenye Li, Feihu Zhang, and Li Zhang. S-nerf: Neural radiance fields for street views. In ICLR, 2023 b

    2023

  34. [42]

    Geonlf: Geometry guided pose-free neural lidar fields

    Weiyi Xue, Zehan Zheng, Fan Lu, Haiyun Wei, Guang Chen, and Changjun Jiang. Geonlf: Geometry guided pose-free neural lidar fields. arXiv preprint, 2024

    2024

  35. [43]

    Street gaussians for modeling dynamic urban scenes

    Yunzhi Yan, Haotong Lin, Chenxu Zhou, Weijie Wang, Haiyang Sun, Kun Zhan, Xianpeng Lang, Xiaowei Zhou, and Sida Peng. Street gaussians for modeling dynamic urban scenes. In ECCV, 2024

    2024

  36. [44]

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

    Jiawei Yang, Boris Ivanovic, Or Litany, Xinshuo Weng, Seung Wook Kim, Boyi Li, Tong Che, Danfei Xu, Sanja Fidler, Marco Pavone, et al. Emernerf: Emergent spatial-temporal scene decomposition via self-supervision. In ICLR, 2024 a

    2024

  37. [45]

    Real-time photorealistic dynamic scene representation and rendering with 4d gaussian splatting

    Zeyu Yang, Hongye Yang, Zijie Pan, Xiatian Zhu, and Li Zhang. Real-time photorealistic dynamic scene representation and rendering with 4d gaussian splatting. In ICLR, 2024 b

    2024

  38. [46]

    Srinivasan, Richard Szeliski, Jonathan T

    Lior Yariv, Peter Hedman, Christian Reiser, Dor Verbin, Pratul P. Srinivasan, Richard Szeliski, Jonathan T. Barron, and Ben Mildenhall. Bakedsdf: Meshing neural sdfs for real-time view synthesis. arXiv preprint, 2023

    2023

  39. [47]

    PlenOctrees for real-time rendering of neural radiance fields

    Alex Yu, Ruilong Li, Matthew Tancik, Hao Li, Ren Ng, and Angjoo Kanazawa. PlenOctrees for real-time rendering of neural radiance fields. In ICCV, 2021

    2021

  40. [48]

    Nerf-lidar: Generating realistic lidar point clouds with neural radiance fields

    Junge Zhang, Feihu Zhang, Shaochen Kuang, and Li Zhang. Nerf-lidar: Generating realistic lidar point clouds with neural radiance fields. In AAAI, 2024

    2024

  41. [49]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In CVPR, 2018

    2018

  42. [50]

    Lidar4d: Dynamic neural fields for novel space-time view lidar synthesis

    Zehan Zheng, Fan Lu, Weiyi Xue, Guang Chen, and Changjun Jiang. Lidar4d: Dynamic neural fields for novel space-time view lidar synthesis. In CVPR, 2024

    2024

  43. [51]

    Drivable 3d gaussian avatars

    Wojciech Zielonka, Timur Bagautdinov, Shunsuke Saito, Michael Zollhöfer, Justus Thies, and Javier Romero. Drivable 3d gaussian avatars. arXiv preprint, 2023

    2023

  44. [52]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  45. [53]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  46. [54]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

    1976

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.