Pith. sign in

REVIEW 5 major objections 4 minor 27 references

ICP-3DGS: SfM-free 3D Gaussian Splatting for Large-scale Unbounded Scenes

T0 review · 5 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper proposes an SfM-free 3D Gaussian Splatting pipeline that uses monocular depth and Generalized ICP to recover camera poses and reconstruct large unbounded outdoor scenes without structure-from-motion.

desk verdict Useful incremental SfM-free 3DGS pipeline with real gains, but the near-perfect pose errors and missing depth-scale analysis make the headline claim overreach as written. read the letter →

arxiv 2506.21629 v1 pith:55GQ6SQ6 submitted 2025-06-24 cs.GR cs.CV

Chenhao Zhang , Yezhi Shen , Fengqing Zhu This is my paper · ORCID
classification cs.GRcs.CV
keywords 3DGaussianSplattingSfM-freeposeestimationIterativeClosestPointmonoculardepthnovelviewsynthesisunboundedscenesvoxeldensification
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

The paper proposes a structure-from-motion-free 3D Gaussian Splatting pipeline that estimates camera poses and reconstructs a 3D scene from raw video, aimed at large-scale unbounded outdoor scenes where SfM preprocessing is brittle and expensive. The method uses monocular metric depth to lift each frame into a point cloud, aligns consecutive clouds with Generalized ICP to initialize the relative camera pose, and then refines that pose by optimizing Gaussians against the next frame; a sky mask removes unreliable infinite-depth regions, and a voxel-based densification step grows the Gaussian scene only where the current coverage is sparse. On KITTI-360 street sequences the authors report 23.32 dB PSNR against 13.96 dB for CF-3DGS and 17.60 dB for Nope-NeRF, with far lower trajectory error, and a smaller but consistent improvement on the bounded Tanks and Temples scenes. If those results stand, neural rendering from raw outdoor video becomes feasible without SfM, which matters because SfM is the conventional bottleneck in large-scale scene reconstruction.

What carries the argument

The load-bearing machinery is a camera-pose estimator built from monocular depth plus Generalized ICP, and a voxel-based scene-growth rule. For each adjacent frame pair, Metric3DV2 depth is lifted into two point clouds; G-ICP finds a coarse rigid transform between them, which seeds an optimization of a learnable SE(3) warp applied to the previous frame's Gaussians, supervised by an L1 rendering loss against the next frame. A sky mask defined by maximum depth excludes the unreliable infinite-depth background from this optimization. On the reconstruction side, the current frame's points are transformed into the scene coordinate frame and voxelized alongside the existing Gaussian centers, and new points are inserted only into voxels where the incoming density ratio exceeds a threshold, giving the scene a way to grow into areas that the first frame never covered.

What would settle it

A decisive check would be to split KITTI-360 frames into high- and low-depth-error groups using the dataset's LiDAR ground truth, then measure the method's per-group ATE and PSNR; if the high-error group does not show substantially worse pose and rendering quality, the claimed dependence on monocular depth quality is cast into doubt, and if it does, the central assumption is confirmed and the method's range of applicability is precisely bounded.

Watch

Extended reading notes

Core claim

The central claim is that a 3D Gaussian Splatting scene can be built and its camera poses recovered from an unposed video sequence, even with large inter-frame motion and unbounded spatial extent, by pairing monocular metric depth with Generalized ICP for pose initialization and a voxel-driven growth rule for the Gaussian representation. In the proposed ICP-3DGS, each adjacent frame pair is registered by lifting both depth maps to point clouds and applying G-ICP to get a coarse relative transform; the transform then initializes a learnable SE(3) warp of the previous frame's Gaussians, refined with an L1 photometric loss against the next frame. Skipping sky pixels and comparing voxel occupancy between the new frame's points and the existing Gaussian centers lets the reconstruction expand into uncovered regions rather than relying on gradient-based splitting that stalls in empty space. The authors present this as an SfM-free 3DGS formulation that scales to street-level scenes, evidenced by 9 dB PSNR improvements over CF-3DGS on KITTI-360 and accurate recovered trajectories (average ATE of 0.001 after scale alignment).

Load-bearing premise

The pipeline assumes the pretrained monocular depth model Metric3DV2 produces sufficiently accurate, scale-consistent metric depth at every frame, because the depth lifts both the pose-initialization point clouds and the densification input; if that depth estimate drifts or fails, the pose initialization and scene growth collapse, and the reported gains evaporate.

Editorial extensions

If this is right

  • Adopting the pipeline would let practitioners reconstruct street-scale scenes directly from raw vehicle-mounted video, bypassing the feature-matching failure modes of SfM on low-texture or fast-moving footage.
  • The recovered camera trajectory is reported at least an order of magnitude more accurate than prior pose-free baselines (average ATE of 0.001 on KITTI-360 after alignment), which could make the pose output usable as a calibration or pseudo-ground-truth source in its own right.
  • The voxel-based densification means the method can grow into previously unseen scene regions, removing the assumption that the first frame's point cloud covers the whole reconstruction volume.
  • Because pose estimation and scene growth take roughly 1.5 hours on KITTI-360, comparable to CF-3DGS and far less than Nope-NeRF's 60 hours, the approach could make pose-free 3DGS the default for large-scale outdoor capture rather than a research novelty.

Reading between the lines

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

  • We infer that a direct extension would be to feed the recovered poses and per-frame depth back into Metric3DV2 or a depth-refinement network, closing the loop between pose and depth; the ablations show a strong dependence on depth quality, so a depth-consistency loss across frames might push the pose error even lower.
  • We infer that the voxel density-ratio criterion could be made adaptive to local scene complexity, for example using finer voxels near the camera and coarser ones far away, which would plausibly improve the unbounded-scene results further than the fixed voxel size used here.
  • We infer that if the ATE numbers are reproduced, the output trajectories could plausibly replace ground-truth poses for training other view-synthesis or pose-estimation models in outdoor settings, a use the paper does not discuss.
  • We infer that the method inherits Metric3DV2's failure modes: on scenes with reflective or translucent surfaces, where monocular depth is notoriously unreliable, the pose initialization would be expected to degrade, a regime the paper's evaluations do not cover.
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 / 4 minor

Summary. The paper proposes ICP-3DGS, an SfM-free pipeline for 3D Gaussian Splatting in large-scale unbounded scenes. For each adjacent frame pair, the method computes metric depth with the pretrained Metric3DV2 model, lifts the depth maps to point clouds, aligns them with G-ICP to obtain a coarse relative pose, and then refines that pose by optimizing an SE(3) transform on a per-frame overfitted Gaussian under an L1 photometric loss. A sky mask defined by the maximum depth value is used to exclude unreliable sky regions. The estimated relative poses are composed sequentially along the trajectory. Reconstruction uses a voxel-based densification strategy that adds points from new frames only in voxels where the current Gaussian point density is low relative to the new point cloud. Experiments on five selected KITTI-360 sequences and on Tanks and Temples report large PSNR gains over Nope-NeRF and CF-3DGS, low pose errors on KITTI-360, and an ablation showing contributions from the sky mask, G-ICP, and voxel densification.

Significance. If the reported results are correct, the paper makes a useful contribution: it is one of the first SfM-free 3DGS methods aimed at large outdoor scenes, it releases source code, and its ablation indicates that each proposed module contributes to the final rendering quality. The use of external benchmarks with fixed test frames and the comparison against two prior pose-free methods are strengths. However, the central claim depends on an unvalidated assumption about the metric-scale consistency of the pretrained depth model, and the evaluation lacks error bars, omits recent baseline methods, and reports pose errors that are implausibly small without a precise statement of the alignment protocol. These issues currently limit confidence in the magnitude of the reported gains and in the attribution of those gains to the proposed modules rather than to the depth prior.

major comments (5)
  1. [§2.2, Eq. (3)] The entire pose-estimation pipeline rests on the assumption that Metric3DV2 produces scale-consistent metric depth across frames. G-ICP between two depth-lifted point clouds recovers a rigid transform only if the clouds are in the same metric scale; any per-frame scale drift biases the recovered translation, and because Eq. (3) composes pairwise transforms, the bias accumulates along the trajectory. The paper provides no depth-error or scale-drift analysis on KITTI-360, where LiDAR ground truth is available, and Table 5 contains no ablation condition that replaces Metric3DV2 with ground-truth depth. Please add (i) per-frame depth MAE and scale error relative to LiDAR, (ii) pose and rendering results when the initialization uses ground-truth depth, and (iii) a controlled scale-perturbation experiment. Without these, the attribution of the 9 dB rendering gain and the low pose errors to the proposed ICP and voxel modules rather than to the depth prior is not established.
  2. [§3.2, Table 3] The reported pose errors are implausibly small: an average ATE of roughly 0.001 m over trajectories exceeding 100 m is difficult to reconcile with a sequential monocular estimator. The text says that 'we align the scale and trajectory for Nope-NeRF and CF-3DGS before computing the camera pose error,' but it does not state whether the same alignment is applied to the proposed method or what exact alignment (e.g., Sim(3) Umeyama alignment) is used. Please specify the full evaluation protocol for all methods, report errors both before and after alignment, and include trajectory plots. As written, the pose claim is not falsifiable.
  3. [§3.1, Tables 1-3] All quantitative results are single-run point estimates with no error bars, standard deviations, or statistical significance tests. The KITTI-360 evaluation uses only five sequences selected for mainly static traffic agents, which is a narrow subset of urban driving and does not exercise the method under moving vehicles, pedestrians, or changing lighting. Moreover, the introduction cites Free-SurGS and IncEventGS as relevant pose-free 3DGS methods, but neither is included in any comparison. Please report multiple runs with variance, evaluate on additional KITTI-360 sequences or the standard split, and compare against these recent baselines where applicable.
  4. [§2.3, §3.1] Reproducibility-critical hyperparameters are missing. The voxel size and the density ratio threshold in §2.3 are not reported anywhere, the photometric loss weight λ in Eq. (5) is not specified, and the optimizer settings for the refinement in Eq. (6) are not described. A reader cannot reproduce Tables 1-3 or transfer the method to a new scene without guessing these values. Please report all hyperparameters and the training schedule.
  5. [§2.2, Eq. (7)] The sky mask is defined as D_t = max(D_t), which assumes that the global maximum depth in every frame is always sky. In outdoor scenes, distant buildings, vegetation, or clipped depth values can also attain the maximum, and Metric3DV2's depth clamping may assign sky and other far regions the same maximum value. The paper does not quantify how well this mask agrees with true sky or compare it with a semantic sky segmenter. Please include a mask-quality evaluation and an ablation with a semantic sky mask to justify this design choice.
minor comments (4)
  1. [Throughout] There are inconsistent spellings: 'SfM' in the title and abstract but 'SFM' in the body, 'out door' in the abstract, and 'Tanks and Temple' in Section 3.2. Please unify terminology and fix typos.
  2. [§3.3] The text says the Tanks and Temples qualitative results are 'presented in Figure 3', but the correct reference is Figure 4. The captions for Figures 3 and 4 are also nearly identical and should be made distinct.
  3. [§3.2] The sentence 'we achieve a significant 5.7 dB PSNR improvement on average' is ambiguous: 5.7 dB is the average improvement over Nope-NeRF on KITTI-360 (Table 1), not the improvement over CF-3DGS, while the surrounding text refers to both baselines. Please state the comparison explicitly.
  4. [§2.2, Eq. (2)] In Eq. (2), the view transform matrix W is used without first defining the coordinate frame convention. A brief definition of W and J would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the pose and rendering predictions are checked against external ground truth, and the self-cited depth model is independently validated rather than fitted to the target.

full rationale

The paper's derivation chain (Sec. 2.2) is a standard estimator: it computes per-frame depth maps from a pretrained monocular model (Metric3DV2), lifts them to point clouds, registers adjacent clouds with G-ICP, refines the transform via photometric loss (Eq. 6), and composes the relative poses (Eq. 3). This is not a case where the claimed 'prediction' is defined as the fitted parameter: the target quantities (camera poses and rendered novel views) are evaluated against KITTI-360 LiDAR/GPS ground truth and held-out test views, which are external to the fitted pipeline. The cited Metric3DV2 (ref. [13]) is co-authored by the first author but is a publicly released, independently trained depth foundation model with its own benchmarks; its output (depth) is not the target of this paper and is not re-derived from the paper's fitting. The ablation in Table 5 controls for the depth prior by using Metric3DV2 in the base setup, isolating the contribution of the proposed G-ICP and voxel densification. No equation reduces to its input by construction; Eq. (3) is the standard composition of relative poses, and Eq. (6) optimizes a free SE(3) transform against an image. Therefore, there is no significant circularity.

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

The central claim rests on the accuracy of a pretrained monocular depth model, a max-depth sky heuristic, and drift-prone pose composition, but no new physical entities are introduced. The undisclosed densification hyperparameters are free parameters that affect the reported results.

free parameters (3)
  • voxel_size = not reported
    Used in the voxel-based densification to compare density of scene points and existing Gaussians; the value is not given in the paper.
  • density_ratio_threshold = not reported
    Determines when a voxel is under-covered and new points should be added; the threshold is not reported.
  • photometric_loss_weight lambda = not reported
    Balances L1 and D-SSIM in the photometric loss in Eq. 5; the value is not given.
assumptions (3)
  • domain assumption Metric3DV2 provides metric depth accurate enough for point cloud alignment and Gaussian initialization
    Invoked in Section 2.2 when lifting frames to 3D points for G-ICP; no depth-error analysis is provided.
  • ad hoc to paper Sky corresponds to the maximum depth value in the depth map
    Eq. 7 defines the sky mask as M_t = I(D_t = max(D_t)); this is a heuristic with no semantic segmentation validation.
  • domain assumption Composing pairwise relative poses without global bundle adjustment is drift-free enough for the tested sequences
    Eq. 3 accumulates relative poses; the paper does not report drift analysis or loop closure.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ICP-3DGS: SfM-free 3D Gaussian Splatting for Large-scale Unbounded Scenes." pith.science (2026). https://pith.science/paper/55GQ6SQ6

@misc{pith2026250621629,
  author       = {Pith},
  title        = {Pith review of: ICP-3DGS: SfM-free 3D Gaussian Splatting for Large-scale Unbounded Scenes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/55GQ6SQ6}},
  note         = {Machine review of arXiv:2506.21629}
}
read the original abstract

In recent years, neural rendering methods such as NeRFs and 3D Gaussian Splatting (3DGS) have made significant progress in scene reconstruction and novel view synthesis. However, they heavily rely on preprocessed camera poses and 3D structural priors from structure-from-motion (SfM), which are challenging to obtain in outdoor scenarios. To address this challenge, we propose to incorporate Iterative Closest Point (ICP) with optimization-based refinement to achieve accurate camera pose estimation under large camera movements. Additionally, we introduce a voxel-based scene densification approach to guide the reconstruction in large-scale scenes. Experiments demonstrate that our approach ICP-3DGS outperforms existing methods in both camera pose estimation and novel view synthesis across indoor and outdoor scenes of various scales. Source code is available at https://github.com/Chenhao-Z/ICP-3DGS.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 24 canonical work pages

  1. [1]

    ICP-3DGS: SfM-free 3D Gaussian Splatting for Large-scale Unbounded Scenes

    INTRODUCTION Neural rendering has witnessed significant progress recently, with models like Neural Radiance Fields (NeRF) [1] and 3D *These authors contributed equally. Gaussian Splatting (3DGS) [2] emerging as leading tech- niques for photorealistic view synthesis and 3D reconstruc- tion. Despite their success, NeRF and 3DGS share a critical limitation: ...

    work page Pith review arXiv 2025

  2. [2]

    Preliminary Problem formulation.Given a sequence ofNunposed RGB images frames, we denote it as{Ii}N i=1, whereI i ∈R H×W×3 and K is the camera intrinsics

    METHODOLOGY 2.1. Preliminary Problem formulation.Given a sequence ofNunposed RGB images frames, we denote it as{Ii}N i=1, whereI i ∈R H×W×3 and K is the camera intrinsics. Our objective is to recover the camera poseP i for each of the imagesI i and reconstruct a 3D Gaussian splatting (3DGS) representationGfor the whole scene simultaneously. The output is ...

    work page

  3. [3]

    EXPERIMENT 3.1. Experimental Setting CF-3DGS Nope-NeRF Ours PSNR↑/ SSIM↑/ LPIPS↓PSNR↑/ SSIM↑/ LPIPS↓PSNR↑/ SSIM↑/ LPIPS↓ Seq 1 13.59 / 0.59 / 0.45 17.01 / 0.65 / 0.5323.77/0.84/0.19 Seq 2 14.85 / 0.48 / 0.49 16.69 / 0.50 / 0.6322.81/0.78/0.22 Seq 3 13.52 / 0.40 / 0.56 18.02 / 0.50 / 0.6223.76/0.77/0.21 Seq 4 12.14 / 0.41 / 0.54 16.92 / 0.51 / 0.6121.49/0....

    work page

  4. [4]

    We propose an SFM-free 3D Gaussian method for large-scale unbounded scenes

    CONCLUSION. We propose an SFM-free 3D Gaussian method for large-scale unbounded scenes. To address the challenges of large camera movements and inconsistent depth in outdoor scenes, we in- troduce a sky mask to separate the background and utilize G- ICP in a two-step optimization process to achieve robust cam- era pose estimation. Furthermore, we propose ...

    work page

  5. [5]

    Barf: Bundle-adjusting neural radiance fields,

    C.-H. Lin, W.-C. Ma, A. Torralba, and S. Lucey, “Barf: Bundle-adjusting neural radiance fields,”2021 IEEE/CVF International Conference on Computer Vi- sion (ICCV), pp. 5721–5731, 2021

    work page 2021

  6. [6]

    Gaussian activated neural radiance fields for high fi- delity reconstruction and pose estimation,

    S.-F. Chng, S. Ramasinghe, J. Sherrah, and S. Lucey, “Gaussian activated neural radiance fields for high fi- delity reconstruction and pose estimation,”Computer Vision – ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part XXXIII, p. 264–280, 2022

    work page 2022

  7. [7]

    to construct the training and testing frames sets. For each scene in the datasets, one in every eight frames is selected for testing, with the remaining frames used for training, ex- cept for the ‘Family’ scene in the Tanks and Temples dataset, where every other frame is selected for training and testing. Nope-NeRF CF-3DGS Ours RPEt↓/ RPEr↓/ ATE↓RPEt↓/ RP...

    work page

  8. [8]

    S mask” denotes the sky mask, and “VB-Den

    on both novel view synthesis and camera pose estima- tion. Following the idea of NeRF−−[4], we freeze our trained 3DGS model and minimize the photometric error between the synthesized images and the test views to obtain the testing camera poses. We apply this procedure to all methods during evaluation to ensure a fair comparison. The quantitative comparis...

    work page

Show all 27 references
  1. [9]

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

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

    2021

  2. [10]

    3d gaussian splatting for real-time radiance field ren- dering,

    B. Kerbl, G. Kopanas, T. Leimkuehler, and G. Drettakis, “3d gaussian splatting for real-time radiance field ren- dering,”ACM Trans. Graph., vol. 42, no. 4, July 2023

    2023

  3. [11]

    Structure-from- motion revisited,

    J. L. Sch ¨onberger and J.-M. Frahm, “Structure-from- motion revisited,”2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4104–4113, 2016

    2016

  4. [12]

    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, vol. arXiv:2102.07064, 2022

    2022 arXiv

  5. [13]

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

    W. Bian, Z. Wang, K. Li, and J.-W. Bian, “Nope-nerf: Optimising neural radiance field with no pose prior,” 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4160–4169, 2023

    2023

  6. [14]

    Colmap-free 3d gaussian splatting,

    Y . Fu, X. Wang, S. Liu, A. Kulkarni, J. Kautz, and A. A. Efros, “Colmap-free 3d gaussian splatting,”2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 20796–20805, 2024

    2024

  7. [15]

    Free-SurGS: SfM-Free 3D Gaussian Splatting for Surgical Scene Reconstruction ,

    J. Guo, J. Wang, D. Kang, W. Dong, W. Wang, and Y .-h. Liu, “ Free-SurGS: SfM-Free 3D Gaussian Splatting for Surgical Scene Reconstruction ,”proceedings of Medi- cal Image Computing and Computer Assisted Interven- tion – MICCAI 2024, vol. LNCS 15007, October 2024

    2024

  8. [16]

    Inceventgs: Pose-free gaussian splatting from a single event camera,

    J. Huang, C. Dong, and P. Liu, “Inceventgs: Pose-free gaussian splatting from a single event camera,”arXiv preprint, vol. arXiv:2410.08107, 2024

    2024 arXiv

  9. [17]

    6img-to-3d: Few-image large-scale outdoor driving scene reconstruction,

    T. Gieruc, M. K ¨astingsch¨afer, S. Bernhard, and M. Salz- mann, “6img-to-3d: Few-image large-scale outdoor driving scene reconstruction,”arXiv preprint, vol. arXiv:2404.12378, 2024

    2024 arXiv

  10. [18]

    Generalized-icp.,

    A. Segal, D. H ¨ahnel, and S. Thrun, “Generalized-icp.,” Robotics: Science and Systems, vol. 2, pp. 435, 2009

    2009

  11. [19]

    Metric3d v2: A versatile monocular geometric foundation model for zero-shot metric depth and surface normal estimation,

    M. Hu, W. Yin, C. Zhang, Z. Cai, X. Long, H. Chen, K. Wang, G. Yu, C. Shen, and S. Shen, “Metric3d v2: A versatile monocular geometric foundation model for zero-shot metric depth and surface normal estimation,” IEEE Transactions on Pattern Analysis and Machine In- telligence, ...

    2024

  12. [20]

    Tangent space backpropagation for 3d transformation groups,

    Z. Teed and J. Deng, “Tangent space backpropagation for 3d transformation groups,”2021 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR), pp. 10333–10342, 2021

    2021

  13. [21]

    A method for registration of 3-d shapes,

    P. Besl and N. D. McKay, “A method for registration of 3-d shapes,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 14, no. 2, pp. 239–256, 1992

    1992

  14. [22]

    Revis- ing densification in gaussian splatting,

    S. Rota Bul `o, L. Porzi, and P. Kontschieder, “Revis- ing densification in gaussian splatting,”Computer Vi- sion – ECCV 2024: 18th European Conference, Milan, Italy, September 29–October 4, 2024, Proceedings, Part LXIII, p. 347–362, 2024, Milan, Italy

    2024

  15. [23]

    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 Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 3, pp. 3292–3310, 2023

    2023

  16. [24]

    Tanks and temples: benchmarking large-scale scene re- construction,

    A. Knapitsch, J. Park, Q.-Y . Zhou, and V . Koltun, “Tanks and temples: benchmarking large-scale scene re- construction,”ACM Trans. Graph., vol. 36, no. 4, July 2017

    2017

  17. [25]

    Lightning nerf: Ef- ficient hybrid scene representation for autonomous driv- ing,

    J. Cao, Z. Li, N. Wang, and C. Ma, “Lightning nerf: Ef- ficient hybrid scene representation for autonomous driv- ing,”2024 IEEE International Conference on Robotics and Automation (ICRA), pp. 16803–16809, 2024

    2024

  18. [26]

    Im- age quality assessment: from error visibility to struc- tural similarity,

    Z. Wang, A. Bovik, H. Sheikh, and E. Simoncelli, “Im- age quality assessment: from error visibility to struc- tural similarity,”IEEE Transactions on Image Process- ing, vol. 13, no. 4, pp. 600–612, 2004

    2004

  19. [27]

    The unreasonable effectiveness of deep fea- tures as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep fea- tures as a perceptual metric,”2018 IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, pp. 586–595, 2018

    2018

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.