Pith. sign in

REVIEW 3 major objections 7 minor 91 references

LiDAR scans can be completed to photogrammetric detail by growing Gaussian surfels only where the scan is ambiguous.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 18:51 UTC pith:FUDD4VWQ

load-bearing objection A practical, well-engineered LiDAR completion pipeline with a real blind spot: the ambiguity heuristic only fires when a missing structure leaves a density shadow, and the paper never tests the common perpendicular-thin-structure case. the 3 major comments →

arxiv 2512.03010 v2 pith:FUDD4VWQ submitted 2025-12-02 cs.CV cs.GRcs.RO

SurfFill: Completion of LiDAR Point Clouds via Gaussian Surfel Splatting

Svenja Strobel , Matthias Innmann , Bernhard Egger , Marc Stamminger , Linus Franke This is my paper
classification cs.CV cs.GRcs.RO
keywords LiDAR point cloud completionGaussian surfel splattingambiguity heuristicbeam divergence artifactsmixed pixelsphotometric reconstruction fusionlarge-scale scene reconstructionpoint density analysis
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to establish that LiDAR scanning and photo-based reconstruction complement each other well enough that a point cloud can be completed by deliberately growing geometry only into the places LiDAR missed. It argues that LiDAR errors concentrate at thin structures and edges because the beam diverges and mixes foreground and background returns, and that these failures leave a detectable signature: a transition zone of low, non-zero point density. SurfFill reads that signature with a per-point ambiguity score, focuses a Gaussian surfel reconstruction on those zones, and filters and samples the resulting surfels back into the scan. If the claims hold, LiDAR scans can gain photogrammetric detail on edges and dark objects while keeping LiDAR's precision on flat surfaces, in about 25 minutes per scene and with a chunking scheme for building-sized scans.

Core claim

The paper's core claim is that LiDAR incompleteness can be located without knowing what is missing. It defines ambiguity per point as the mean distance to k nearest neighbors scaled by a device-dependent spacing: low values mark well-scanned regions, high values mark transition zones, and these zones are exactly the borders of missing geometry. The paper then constrains a Gaussian surfel model to densify and optimize in those zones, guided by image-space normal uncertainty, an edge gradient loss, scale regularization, and opacity-aware filtering, and samples the surviving surfels back into the scan. The authors report a mean F1 of 0.9176 and a mean Chamfer distance of 0.0055 across five test

What carries the argument

The central object is the ambiguity score p(x), a per-point inverse-density estimate computed as the mean distance to k nearest neighbors divided by the typical LiDAR point spacing. Small p means dense, well-scanned; large p means transition toward a gap. This score drives non-uniform downsampling, controls the initial surfel scale, guides densification and pruning, informs training-view sampling, and later filters out low-ambiguity Gaussian surfels. The second mechanism is focused Gaussian surfel optimization: flat Gaussian primitives are grown only into flagged regions, with a max-gradient densification heuristic, positional jitter for exploration, a gradient edge loss, scale regularizatio

Load-bearing premise

The ambiguity heuristic assumes every missed structure sits next to a measurable transition zone of low but nonzero point density; a fully isolated missed object or a surface that returns no laser signal produces no ambiguity signal, so it would never be targeted.

What would settle it

Run SurfFill on a scene where a thin metal rod is visible in the RGB images but has been removed from the LiDAR scan, ensuring no residual points and no low-density transition zone remain around the rod; if the pipeline adds no points over the rod, the density-border premise fails. A simpler check is to compute the ambiguity score on a scan after deleting one isolated chair leg and verify whether the surrounding k-nearest-neighbor distance rises measurably at all.

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

If this is right

  • LiDAR and photometric reconstruction are complementary in exactly the way the artifact analysis suggests: the same regions where LiDAR is weak, such as thin structures, edges, and dark surfaces, are the regions where image-based reconstruction is strong.
  • Combining completed points with the original scan does not measurably degrade precision in flat, well-scanned regions, according to the reported precision and F1 scores.
  • Building-scale scans with tens of millions of points remain processable: chunked training runs in parallel and completes a six-room office scan in about 70 minutes on six GPUs.
  • The added heuristic and filtering stages cost little runtime; the total time is close to baseline splatting, around 25 minutes on the test scenes.
  • Direct learning-based point cloud completion methods trained on object-level benchmarks do not transfer to these LiDAR-scale scenes, as the paper reports far lower F1 scores for those methods.

Where Pith is reading between the lines

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

  • The ambiguity heuristic is effectively a saliency map for LiDAR failure; it could be reused outside Gaussian surfel splatting as a general prior for any photometric completion method, though the paper does not test that transfer.
  • Because the pipeline relies on RGB images, it inherits photometric failure modes: sparse camera views, dynamic objects, and lighting changes limit completion, which the paper acknowledges as a limitation.
  • A natural stress test would be completing a fully unobserved structure, such as a table leg with no residual neighboring points at all; in that case the density-based ambiguity signal would be absent and the heuristic would need image-only cues to fire.
  • The chunking extension introduces small artifacts at chunk borders, with about a 0.6% F1 drop on one small scene; this suggests that smarter inter-chunk blending is a promising refinement, as the paper notes.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. SurfFill proposes a pipeline for completing missing fine structures in LiDAR point clouds by combining a 3D point-density ambiguity heuristic with a 2D-Gaussian-Splatting optimization that is focused, via image-space uncertainty masks and adapted losses, on the detected ambiguous regions. After training, Gaussian surfels in these regions are filtered and sampled back into the LiDAR cloud. A divide-and-conquer chunking scheme extends the method to large building-scale scans. The paper reports quantitative results on five synthetic test scenes (ATTIC, KITCHEN, MUSEUM, MEETINGROOM, CATERPILLAR) with F1-score improvement from 0.8781 (2DGS) to 0.9176, lower Chamfer distance, and roughly 25 minutes runtime per scene, plus qualitative results on real-world scans. Ablations quantify the contribution of each component.

Significance. If the central claim holds, the paper provides a practical, efficient way to integrate photometric reconstruction into LiDAR scans without degrading their precision, which would be valuable for industrial scanning and scene reconstruction. The approach is well-motivated by a physically grounded analysis of LiDAR beam divergence and mixed pixels. The authors ship code and a project page, and the ablation study is fairly thorough. The method is not derived from its outputs and the KITCHEN removed-point recovery check is a genuinely held-out quantitative evaluation. However, the load-bearing evaluation is entirely synthetic, using the authors' own point-removal artifact model, and the ambiguity heuristic's coverage of real LiDAR failure modes is not quantitatively demonstrated.

major comments (3)
  1. [§5.1, Eq. (1); §11] The ambiguity heuristic p(x) = mean kNN distance / δ assumes that every region requiring completion is bordered by a transition zone of low, non-zero point density. This assumption fails for a common geometry: a thin structure rising perpendicularly from a dense, well-scanned surface, e.g., a chair leg or fence bar on a floor. For a floor point near the base of the missing structure, the k nearest neighbors are other floor points at distance ~δ; the absent structure contributes no neighbor, so p(x) remains near the well-scanned baseline. The region is not marked ambiguous, receives no focused densification bias, and may be filtered out. The synthetic test data (Fig. 11) is generated by deleting entire structures, which creates large voids and thus clear transition zones, so the evaluation is biased toward cases where the heuristic fires. The real-world qualitative results (e.g., fence ba
  2. [§10, Appendix E, Fig. 11] The quantitative evaluation is entirely synthetic: input LiDAR clouds are created by deleting points from ground truth scans using a point-removal scheme that mimics 'typical LiDAR artifacts' and in ATTIC 'the exact regions missing in the real dataset'. This protocol is reasonable, but it creates input clouds that are precisely those for which the ambiguity heuristic is designed to fire (large contiguous removed structures with surrounding transition zones). The paper does not report a quantitative experiment on real-world scans, where missing structures are often partially missing, fragmented, or accompanied by noise rather than cleanly removed whole structures. To strengthen the central claim, please add a robustness analysis with different deletion patterns (e.g., removing only portions of thin structures, or deleting points on dark/absorbent surfaces without creating a large void) an
  3. [§10.3 'Missing Points'] The removed-point recall reported for KITCHEN — 35% within 10 mm, 60% within 20 mm, 74% within 30 mm — indicates that a substantial fraction of the deleted structures is not recovered even in the synthetic setting that is most favorable to the heuristic. This is not necessarily a flaw, but it should be contextualized. Please report the same recall metric for the baselines (2DGS, 3DGS, GOF, COLMAP, Neuralangelo) and, ideally, for the other four test scenes. Without this, the reader cannot tell whether the F1-score improvement comes primarily from completing the deleted structures or from adding points near existing surfaces that slightly increase recall.
minor comments (7)
  1. [§7, Eq. (3)] The notation 'M(and the inverse M)' is ambiguous in the text; please define M and ar{M} clearly as the uncertainty mask and its complement. The equation itself is understandable but the surrounding text could be more explicit.
  2. [§8, first paragraph of 'Filtering'] The sentence 'Next, original Gaussians via the "was-densified" flag as well as Gaussians with remaining high ambiguity are removed' is confusing. It seems to say that both the original (non-densified) Gaussians and the high-ambiguity ones are discarded, but the goal of filtering is to keep splats that grew into missing areas. Please rewrite to state exactly which Gaussians are kept and which are removed, and how 'remaining high ambiguity' is computed.
  3. [§5.1, Eq. (1)] The parameter δ ('typical distance between points') is not given a precise computational definition. It is stated to depend on the LiDAR device and preprocessing, but the paper should specify how δ is estimated in practice (e.g., median kNN distance over a subsample of well-scanned points). This is important for reproducibility.
  4. [§2, Related Work] The phrase 'we present the first approach to evaluate artifacts and complete LiDAR scans using Gaussian Splatting' appears to be an overclaim given existing works that incorporate LiDAR into Gaussian splatting (e.g., Li-GS [31]) even if they do not target completion. Please soften the wording to 'the first approach to explicitly focus on LiDAR artifact completion with Gaussian surfels' or similar.
  5. [§10.3, Fig. 10] The initialization experiment (2DGS + ambiguity preprocessing vs. 2DGS + uniform downsampling) is informative, but the caption and text could clarify that the comparison uses the same total point budget after downsampling, so the effect is attributable to the non-uniform distribution rather than to different point counts.
  6. [Appendix F.3, Fig. 13] The pose robustness experiment reports only qualitative results for the ATTIC scene. A numeric F1/Chamfer comparison with and without pose noise would strengthen the claim that the method is 'less susceptible' to pose noise.
  7. [§10, Table 2] The COLMAP result for CATERPILLAR (CD=4.8968, F1=0.6338) is far worse than for other scenes, likely due to a failure case. The paper reports the mean including this outlier; a more robust statistic (e.g., median) or a breakdown discussion would be helpful.

Circularity Check

0 steps flagged

No significant circularity: heuristics and losses are method components; the held-out removal-recall test is not forced by construction.

full rationale

The claimed derivation chain is not circular. The ambiguity heuristic (Eq. 1) defines a point-wise density proxy; it is an input-dependent detector, not a quantity derived from the completion output. The completion itself is produced by Gaussian surfel optimization (Sec. 7) with external image-based losses, followed by filtering/sampling (Sec. 8). Eq. 4 selects Gaussians by distance to the input cloud but does not encode the ground-truth completed points; F1/Chamfer require spatially accurate placement, which is an empirical result rather than a logical consequence of the filter. The synthetic benchmarks (Appendix E) delete structures from ground truth to mimic LiDAR artifacts; the deletions are not chosen by Eq. 1, and the KITCHEN recall against removed points (Sec. 10.2) is a genuinely held-out measure. Real-world qualitative results on ScanNet++, Tanks & Temples, and the BRIDGE failure in Sec. F.8 provide independent evidence and an explicit admission of dependence on visual data quality. The paper's self-citations (e.g., VET, TRIPS) are related-work context and are not load-bearing for the central completion claim. Although Appendix B tunes the ambiguity threshold on the synthetic ATTIC scene and the synthetic removals are aligned with the paper's artifact model, this is an evaluation-validity caveat, not a reduction of the output to the input by construction.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 0 invented entities

The method rests on several hand-set thresholds and an untested spatial-transition assumption about LiDAR failure; it introduces no new physical entities. The key free parameters are τ, δ, the uncertainty threshold, and the filtering distances, most of which are tuned on one scene or set by hand.

free parameters (6)
  • point ambiguity threshold τ = 0.04
    Selected by ablation on ATTIC (Appendix B, Table 3); used in preprocessing (Sec. 6), focused splatting (Sec. 7), and filtering (Sec. 8). ATTIC is also reported in the main quantitative results, so this is tuning on a test scene.
  • uncertainty mask threshold = 0.2
    Selected by ablation on ATTIC (Appendix B, Table 3); used to mask the loss terms in Eq. 3.
  • δ (typical point distance)
    Eq. (1) normalizes the ambiguity score by δ, stated to depend on LiDAR device and preprocessing; no estimation procedure is given, so it is a free scale that directly controls which points count as ambiguous relative to τ.
  • k for KNN searches = 3 and 5
    Appendix A states 'k=3 and k=5 sufficed'; not ablated, but k affects both the ambiguity score and the distance filter in Eq. (4).
  • distance filter bounds t_min, t_max = 0.01 m, 3 m
    Sec. 8, Eq. (4): hand-set 'used for all test scenes'; determines which Gaussians are retained as completed points.
  • loss weights γ, ρ, α, β, λ = λ=0.4; others unspecified
    Eq. (3) defines the objective with weights γ, ρ, α, β; only λ is given in the text. These values affect the geometry and are not fully reproducible from the paper.
axioms (5)
  • domain assumption Regions that need completion are bordered by transition zones of low, non-zero point density.
    Sec. 5.1, Eq. (1): 'Transition regions... show a low, non-zero point density. We exploit this property.' A fully isolated missing structure or a dark surface with no neighboring returns breaks this assumption.
  • domain assumption LiDAR beam divergence is the main cause of missing thin structures and edges, and these are feature-rich regions recoverable from photos.
    Sec. 3.1 attributes artifacts to beam divergence and mixed pixels; the entire completion target depends on this physical model and on photometric reconstruction being able to recover those regions.
  • domain assumption Registered RGB images with sufficient overlap and photometric information are available for the ambiguous regions.
    The pipeline takes RGB images as input (Fig. 3). Without images covering the gaps, the focused splatting has no supervision; sparse or dynamic scenes fail (Sec. 11, Appendix F.8).
  • domain assumption Gaussian surfel optimization constrained as in Sec. 7 converges to geometry that fills the missing regions rather than hallucinating.
    The method assumes the focused 2DGS optimization will grow into the gap and stop at the true surface; failure modes are visible in the KITCHEN chair legs and BRIDGE results.
  • ad hoc to paper Uncertainty maps from Bae et al. normal estimation identify image regions where LiDAR is incomplete.
    Sec. 5.2 equates abrupt normal changes in image space with LiDAR ambiguity; no empirical or theoretical justification beyond the analogy to point-cloud curvature, and the masks are derived from these maps in Eq. (3).

pith-pipeline@v1.3.0-alltime-deepseek · 21809 in / 13985 out tokens · 120968 ms · 2026-08-03T18:51:39.250624+00:00 · methodology

0 comments
read the original abstract

LiDAR-captured point clouds are often considered the gold standard in active 3D reconstruction. While their accuracy is exceptional in flat regions, the capturing is susceptible to miss small geometric structures and may fail with dark, absorbent materials. Alternatively, capturing multiple photos of the scene and applying 3D photogrammetry can infer these details as they often represent feature-rich regions. However, the accuracy of LiDAR for featureless regions is rarely reached. Therefore, we suggest combining the strengths of LiDAR and camera-based capture by introducing SurfFill: a Gaussian surfel-based LiDAR completion scheme. We analyze LiDAR capturings and attribute LiDAR beam divergence as a main factor for artifacts, manifesting mostly at thin structures and edges. We use this insight to introduce an ambiguity heuristic for completed scans by evaluating the change in density in the point cloud. This allows us to identify points close to missed areas, which we can then use to grow additional points from to complete the scan. For this point growing, we constrain Gaussian surfel reconstruction to focus optimization and densification on these ambiguous areas. Finally, Gaussian primitives of the reconstruction in ambiguous areas are extracted and sampled for points to complete the point cloud. To address the challenges of large-scale reconstruction, we extend this pipeline with a divide-and-conquer scheme for building-sized point cloud completion. We evaluate on the task of LiDAR point cloud completion of synthetic and real-world scenes and find that our method outperforms previous reconstruction methods.

Figures

Figures reproduced from arXiv: 2512.03010 by Bernhard Egger, Linus Franke, Marc Stamminger, Matthias Innmann, Svenja Strobel.

Figure 1
Figure 1. Figure 1: For a given LiDAR scan, we complete the point cloud by carefully integrating Gaussian surfel-based 3D reconstruction. We [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: High-level functionality of LiDAR and formation of mixed pixels: a) General setup: The sensor emits and receives a light ray [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Our pipeline. We first downsample the LiDAR point cloud and generate uncertainty maps using our ambiguity heuristic. Next, [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Focused Gaussian surfel optimization. Our focused [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Our filtering strategy. We first remove outliers, nearly [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The handling of large scenes shown with an example [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Resulting point clouds for our test scenes trained using all evaluated methods. Our method is best able to reconstruct the small [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Resulting point clouds for real-world datasets trained with the evaluated methods: The scenes from ScanNet++ are scanned with [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison between our method and the direct shape [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Resulting point clouds for ATTIC using our method, 2DGS initialized with a uniformly downsampled LiDAR point cloud and 2DGS initialized with a LiDAR point cloud downsam￾pled with the ambiguity heuristic. outperforming other approaches that fail to do so or gen￾erate significant noise. In the three ScanNet++ scenes, our method excels at reconstructing small, missing elements, like the chair legs. Other met… view at source ↗
Figure 13
Figure 13. Figure 13: Resulting point clouds for the ATTIC scene without pose noise using 2DGS, 3DGS and our method compared to our method using pose noise. multi-room office building, where six rooms were captured. This scan encompasses 73M points as well as 2064 images, and as such is not processable by related Gaussian Splatting works. We can process it in six chunks and complete the point cloud, as seen in [PITH_FULL_IMAG… view at source ↗
Figure 12
Figure 12. Figure 12: Results of the improved point cloud of the large real [PITH_FULL_IMAGE:figures/full_fig_p017_12.png] view at source ↗
Figure 14
Figure 14. Figure 14: Comparison between our method, SnowflakeNet [ [PITH_FULL_IMAGE:figures/full_fig_p018_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Computation time of each step of our method for the [PITH_FULL_IMAGE:figures/full_fig_p019_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Resulting point clouds for the real-world datasets [PITH_FULL_IMAGE:figures/full_fig_p020_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Resulting Point Clouds for the ATTIC Test Scene using our method with 2x2 chunks with and w/o Point Extension: point extension reduces artifacts at the chunks’ borders [PITH_FULL_IMAGE:figures/full_fig_p020_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Massive 350m long BRIDGE dataset scanned with a NavVis VLX 3 with 2484 images. We train this scene with 18 chunks in 2.63 hours in parallel on 18 Nvidia A40 GPUs. Ground Truth LiDAR Input Ours [PITH_FULL_IMAGE:figures/full_fig_p020_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Resulting views of the improved point cloud of the mas [PITH_FULL_IMAGE:figures/full_fig_p020_19.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

91 extracted references · 9 linked inside Pith

  1. [1]

    Neural point-based graph- ics

    Kara-Ali Aliev, Artem Sevastopolsky, Maria Kolos, Dmitry Ulyanov, and Victor Lempitsky. Neural point-based graph- ics. InEuropean Conference on Computer Vision, pages 696–712. Springer, 2020. 3

    2020

  2. [2]

    Es- timating and exploiting the aleatoric uncertainty in surface normal estimation

    Gwangbin Bae, Ignas Budvytis, and Roberto Cipolla. Es- timating and exploiting the aleatoric uncertainty in surface normal estimation. InProceedings of the IEEE/CVF Interna- tional Conference on Computer Vision, pages 13137–13146,

  3. [3]

    Bagdasarian, Paul Knoll, Yi-Hsin Li, Florian Barthel, Anna Hilsmann, Peter Eisert, and Wieland Morgen- stern

    Milena T. Bagdasarian, Paul Knoll, Yi-Hsin Li, Florian Barthel, Anna Hilsmann, Peter Eisert, and Wieland Morgen- stern. 3dgs.zip: A survey on 3d gaussian splatting compres- sion methods, 2024. 3

    2024

  4. [4]

    MVSFormer: Multi-view stereo by learning robust image features and temperature-based depth.Trans

    Chenjie Cao, Xinlin Ren, and Yanwei Fu. MVSFormer: Multi-view stereo by learning robust image features and temperature-based depth.Trans. Mach. Learn. Res., 2022,

    2022

  5. [5]

    MVSFormer++: Revealing the devil in transformer’s details for multi-view stereo

    Chenjie Cao, Xinlin Ren, and Yanwei Fu. MVSFormer++: Revealing the devil in transformer’s details for multi-view stereo. InInternational Conference on Learning Represen- tations (ICLR), 2024. 2 11

    2024

  6. [6]

    Shapenet: An information-rich 3d model repository.arXiv preprint arXiv:1512.03012, 2015

    Angel X Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, et al. Shapenet: An information-rich 3d model repository.arXiv preprint arXiv:1512.03012, 2015. 2

    Pith/arXiv arXiv 2015

  7. [7]

    Silhouette-aware warping for image-based rendering

    Gaurav Chaurasia, Olga Sorkine, and George Drettakis. Silhouette-aware warping for image-based rendering. In Computer Graphics Forum, pages 1223–1232. Wiley Online Library, 2011. 3

    2011

  8. [8]

    Neusg: Neural im- plicit surface reconstruction with 3d gaussian splatting guid- ance.arXiv preprint arXiv:2312.00846, 2023

    Hanlin Chen, Chen Li, and Gim Hee Lee. Neusg: Neural im- plicit surface reconstruction with 3d gaussian splatting guid- ance.arXiv preprint arXiv:2312.00846, 2023. 3

    Pith/arXiv arXiv 2023

  9. [9]

    A space-sweep approach to true multi- image matching

    Robert T Collins. A space-sweep approach to true multi- image matching. InProceedings CVPR IEEE Computer So- ciety Conference on Computer Vision and Pattern Recogni- tion, pages 358–363. Ieee, 1996. 2

    1996

  10. [10]

    Letsgo: Large-scale garage modeling and rendering via lidar-assisted gaussian primitives.ACM Trans- actions on Graphics (TOG), 43(6):1–18, 2024

    Jiadi Cui, Junming Cao, Fuqiang Zhao, Zhipeng He, Yifan Chen, Yuhui Zhong, Lan Xu, Yujiao Shi, Yingliang Zhang, and Jingyi Yu. Letsgo: Large-scale garage modeling and rendering via lidar-assisted gaussian primitives.ACM Trans- actions on Graphics (TOG), 43(6):1–18, 2024. 3, 16

    2024

  11. [11]

    High-quality surface recon- struction using gaussian surfels

    Pinxuan Dai, Jiamin Xu, Wenxiang Xie, Xinguo Liu, Huamin Wang, and Weiwei Xu. High-quality surface recon- struction using gaussian surfels. InACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024. 3, 6

    2024

  12. [12]

    Floating textures

    Martin Eisemann, Bert De Decker, Marcus Magnor, Philippe Bekaert, Edilson De Aguiar, Naveed Ahmed, Christian Theobalt, and Anita Sellent. Floating textures. pages 409–

  13. [13]

    Comprehensive review of deep learning-based 3d point cloud completion processing and analysis.IEEE Transactions on Intelligent Transportation Systems, 23(12):22862–22883, 2022

    Ben Fei, Weidong Yang, Wen-Ming Chen, Zhijun Li, Yikang Li, Tao Ma, Xing Hu, and Lipeng Ma. Comprehensive review of deep learning-based 3d point cloud completion processing and analysis.IEEE Transactions on Intelligent Transportation Systems, 23(12):22862–22883, 2022. 2

    2022

  14. [14]

    VET: visual error tomography for point cloud completion and high-quality neural render- ing

    Linus Franke, Darius R ¨uckert, Laura Fink, Matthias Inn- mann, and Marc Stamminger. VET: visual error tomography for point cloud completion and high-quality neural render- ing. InSIGGRAPH Asia Conference Papers, New York, NY , USA, 2023. ACM. 3

    2023

  15. [15]

    TRIPS: Trilinear Point Splatting for Real-Time Ra- diance Field Rendering.Computer Graphics Forum, 43(2),

    Linus Franke, Darius R ¨uckert, Laura Fink, and Marc Stam- minger. TRIPS: Trilinear Point Splatting for Real-Time Ra- diance Field Rendering.Computer Graphics Forum, 43(2),

  16. [16]

    Vr- splatting: Foveated radiance field rendering via 3d gaussian splatting and neural points.Proc

    Linus Franke, Laura Fink, and Marc Stamminger. Vr- splatting: Foveated radiance field rendering via 3d gaussian splatting and neural points.Proc. ACM Comput. Graph. In- teract. Tech., 8(1), 2025. 3

    2025

  17. [17]

    Efros, and Xiaolong Wang

    Yang Fu, Sifei Liu, Amey Kulkarni, Jan Kautz, Alexei A. Efros, and Xiaolong Wang. Colmap-free 3d gaussian splat- ting. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 20796– 20805, 2024. 11

    2024

  18. [18]

    Multi-view stereo for commu- nity photo collections

    Michael Goesele, Noah Snavely, Brian Curless, Hugues Hoppe, and Steven M Seitz. Multi-view stereo for commu- nity photo collections. In2007 IEEE 11th International Con- ference on Computer Vision, pages 1–8. IEEE, 2007. 2

    2007

  19. [19]

    Eggs: Edge guided gaussian splatting for ra- diance fields.International Conference on 3D Technologies for the World Wide Web, 2024

    Yuanhao Gong. Eggs: Edge guided gaussian splatting for ra- diance fields.International Conference on 3D Technologies for the World Wide Web, 2024. 6

    2024

  20. [20]

    Sugar: Surface- aligned gaussian splatting for efficient 3d mesh reconstruc- tion and high-quality mesh rendering

    Antoine Gu ´edon and Vincent Lepetit. Sugar: Surface- aligned gaussian splatting for efficient 3d mesh reconstruc- tion and high-quality mesh rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5354–5363, 2024. 3

    2024

  21. [21]

    Efficient perspective-correct 3d gaussian splatting using hybrid trans- parency.arXiv, 2024

    Florian Hahlbohm, Fabian Friederichs, Tim Weyrich, Li- nus Franke, Moritz Kappel, Susana Castillo, Marc Stam- minger, Martin Eisemann, and Marcus Magnor. Efficient perspective-correct 3d gaussian splatting using hybrid trans- parency.arXiv, 2024. 3

    2024

  22. [22]

    INPC: Implicit neural point clouds for radiance field rendering

    Florian Hahlbohm, Linus Franke, Moritz Kappel, Susana Castillo, Martin Eisemann, Marc Stamminger, and Marcus Magnor. INPC: Implicit neural point clouds for radiance field rendering. InInternational Conference on 3D Vision,

  23. [23]

    Inovis: Instant novel-view syn- thesis

    Mathias Harrer, Linus Franke, Laura Fink, Marc Stam- minger, and Tim Weyrich. Inovis: Instant novel-view syn- thesis. InSIGGRAPH Asia 2023 Conference Papers, pages 1–12, New York, NY , USA, 2023. ACM. 3

    2023

  24. [24]

    Cambridge university press,

    Richard Hartley and Andrew Zisserman.Multiple View Ge- ometry in Computer Vision. Cambridge university press,

  25. [25]

    Scalable inside-out image-based rendering

    Peter Hedman, Tobias Ritschel, George Drettakis, and Gabriel Brostow. Scalable inside-out image-based rendering. ACM Transactions on Graphics (TOG), 35(6):1–11, 2016. 3

    2016

  26. [26]

    Deep blending for free-viewpoint image-based rendering.ACM Transactions on Graphics (TOG), 37(6):1–15, 2018

    Peter Hedman, Julien Philip, True Price, Jan-Michael Frahm, George Drettakis, and Gabriel Brostow. Deep blending for free-viewpoint image-based rendering.ACM Transactions on Graphics (TOG), 37(6):1–15, 2018. 3

    2018

  27. [27]

    Splatad: Real-time li- dar and camera rendering with 3d gaussian splatting for au- tonomous driving.arXiv preprint arXiv:2411.16816, 2024

    Georg Hess, Carl Lindstr ¨om, Maryam Fatemi, Christoffer Petersson, and Lennart Svensson. Splatad: Real-time li- dar and camera rendering with 3d gaussian splatting for au- tonomous driving.arXiv preprint arXiv:2411.16816, 2024. 3

    Pith/arXiv arXiv 2024

  28. [28]

    A comparison of outdoor 3d reconstruction between visual slam and lidar slam

    Yi-Tian Hong and Han-Pang Huang. A comparison of outdoor 3d reconstruction between visual slam and lidar slam. In2023 International Automatic Control Conference (CACS), pages 1–6. IEEE, 2023. 1

    2023

  29. [29]

    2d gaussian splatting for geometrically ac- curate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically ac- curate radiance fields. InACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024. 1, 2, 3, 4, 5, 6, 10, 16

    2024

  30. [30]

    Vegs: View extrapolation of urban scenes in 3d gaussian splatting using learned priors

    Sungwon Hwang, Min-Jung Kim, Taewoong Kang, Jayeon Kang, and Jaegul Choo. Vegs: View extrapolation of urban scenes in 3d gaussian splatting using learned priors. InEuro- pean Conference on Computer Vision, pages 1–18. Springer,

  31. [31]

    Li-gs: Gaussian splatting with lidar incorporated for accurate large-scale reconstruction.arXiv preprint arXiv:2409.12899, 2024

    Changjian Jiang, Ruilan Gao, Kele Shao, Yue Wang, Rong Xiong, and Yu Zhang. Li-gs: Gaussian splatting with lidar incorporated for accurate large-scale reconstruction.arXiv preprint arXiv:2409.12899, 2024. 3

    Pith/arXiv arXiv 2024

  32. [32]

    Percep- tual losses for real-time style transfer and super-resolution

    Justin Johnson, Alexandre Alahi, and Li Fei-Fei. Percep- tual losses for real-time style transfer and super-resolution. CoRR, abs/1603.08155, 2016. 16 12

    Pith/arXiv arXiv 2016

  33. [33]

    3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics, 42 (4), 2023

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics, 42 (4), 2023. 3, 5, 6, 10, 16

    2023

  34. [34]

    A hierarchical 3d gaussian representation for real-time ren- dering of very large datasets.ACM Transactions on Graphics (TOG), 43(4):1–15, 2024

    Bernhard Kerbl, Andreas Meuleman, Georgios Kopanas, Michael Wimmer, Alexandre Lanvin, and George Drettakis. A hierarchical 3d gaussian representation for real-time ren- dering of very large datasets.ACM Transactions on Graphics (TOG), 43(4):1–15, 2024. 3, 5, 7, 19

    2024

  35. [35]

    3d gaussian splat- ting as markov chain monte carlo

    Shakiba Kheradmand, Daniel Rebain, Gopal Sharma, Wei- wei Sun, Yang-Che Tseng, Hossam Isack, Abhishek Kar, Andrea Tagliasacchi, and Kwang Moo Yi. 3d gaussian splat- ting as markov chain monte carlo. InAdvances in Neural Information Processing Systems (NeurIPS), 2024. Spotlight Presentation. 3, 5, 7, 16

    2024

  36. [36]

    Tanks and temples: Benchmarking large-scale scene reconstruction.ACM Transactions on Graphics, 36(4), 2017

    Arno Knapitsch, Jaesik Park, Qian-Yi Zhou, and Vladlen Koltun. Tanks and temples: Benchmarking large-scale scene reconstruction.ACM Transactions on Graphics, 36(4), 2017. 1, 4, 7

    2017

  37. [37]

    Point-based neural rendering with per- view optimization.Computer Graphics Forum, 2021

    Georgios Kopanas, Julien Philip, Thomas Leimk ¨uhler, and George Drettakis. Point-based neural rendering with per- view optimization.Computer Graphics Forum, 2021. 3

    2021

  38. [38]

    Neural point cata- caustics for novel-view synthesis of reflections.ACM Trans- actions on Graphics (TOG), 41(6):1–15, 2022

    Georgios Kopanas, Thomas Leimk ¨uhler, Gilles Rainer, Cl´ement Jambon, and George Drettakis. Neural point cata- caustics for novel-view synthesis of reflections.ACM Trans- actions on Graphics (TOG), 41(6):1–15, 2022. 3

    2022

  39. [39]

    Gaussian-lic: Photo-realistic lidar-inertial-camera slam with 3d gaussian splatting.arXiv preprint arXiv:2404.06926, 2024

    Xiaolei Lang, Laijian Li, Hang Zhang, Feng Xiong, Mu Xu, Yong Liu, Xingxing Zuo, and Jiajun Lv. Gaussian-lic: Photo-realistic lidar-inertial-camera slam with 3d gaussian splatting.arXiv preprint arXiv:2404.06926, 2024. 3

    Pith/arXiv arXiv 2024

  40. [40]

    Pu-gan: a point cloud upsampling ad- versarial network

    Ruihui Li, Xianzhi Li, Chi-Wing Fu, Daniel Cohen-Or, and Pheng-Ann Heng. Pu-gan: a point cloud upsampling ad- versarial network. InProceedings of the IEEE/CVF inter- national conference on computer vision, pages 7203–7212,

  41. [41]

    Neuralangelo: High-fidelity neural surface reconstruction

    Zhaoshuo Li, Thomas M ¨uller, Alex Evans, Russell H Tay- lor, Mathias Unberath, Ming-Yu Liu, and Chen-Hsuan Lin. Neuralangelo: High-fidelity neural surface reconstruction. In IEEE Conference on Computer Vision and Pattern Recogni- tion (CVPR), 2023. 2, 3, 10

    2023

  42. [42]

    KITTI-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d.Pattern Analysis and Machine Intelligence (PAMI),

    Yiyi Liao, Jun Xie, and Andreas Geiger. KITTI-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d.Pattern Analysis and Machine Intelligence (PAMI),

  43. [43]

    Vastgaussian: Vast 3d gaus- sians for large scene reconstruction

    Jiaqi Lin, Zhihao Li, Xiao Tang, Jianzhuang Liu, Shiyong Liu, Jiayue Liu, Yangdi Lu, Xiaofei Wu, Songcen Xu, You- liang Yan, and Wenming Yang. Vastgaussian: Vast 3d gaus- sians for large scene reconstruction. InCVPR, 2024. 3, 7, 19

    2024

  44. [44]

    Dynamic 3d gaussians: Tracking by per- sistent dynamic view synthesis

    Jonathon Luiten, Georgios Kopanas, Bastian Leibe, and Deva Ramanan. Dynamic 3d gaussians: Tracking by per- sistent dynamic view synthesis. In3DV, 2024. 3

    2024

  45. [45]

    McManamon.LiDAR Technologies and Systems

    Paul F. McManamon.LiDAR Technologies and Systems. SPIE, 2019. 3

    2019

  46. [46]

    Nerf: Representing scenes as neural radiance fields for view syn- thesis.Communications of the ACM, 65(1):99–106, 2021

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis.Communications of the ACM, 65(1):99–106, 2021. 3

    2021

  47. [47]

    Compressed 3d gaussian splatting for accelerated novel view synthesis

    Simon Niedermayr, Josef Stumpfegger, and R ¨udiger West- ermann. Compressed 3d gaussian splatting for accelerated novel view synthesis. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10349–10358, 2024. 3

    2024

  48. [48]

    Radsplat: Radiance field-informed gaussian splat- ting for robust real-time rendering with 900+ fps.arXiv preprint arXiv:2403.13806, 2024

    Michael Niemeyer, Fabian Manhardt, Marie-Julie Rakoto- saona, Michael Oechsle, Daniel Duckworth, Rama Gosula, Keisuke Tateno, John Bates, Dominik Kaeser, and Federico Tombari. Radsplat: Radiance field-informed gaussian splat- ting for robust real-time rendering with 900+ fps.arXiv preprint arXiv:2403.13806, 2024. 3

    Pith/arXiv arXiv 2024

  49. [49]

    Unisurf: Unifying neural implicit surfaces and radiance fields for multi-view reconstruction

    Michael Oechsle, Songyou Peng, and Andreas Geiger. Unisurf: Unifying neural implicit surfaces and radiance fields for multi-view reconstruction. InInternational Con- ference on Computer Vision (ICCV), 2021. 3

    2021

  50. [50]

    Pointnet: Deep learning on point sets for 3d classification and segmentation

    Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 652–660,

  51. [51]

    StopThePop: Sorted Gaussian Splatting for View-Consistent Real-time Rendering.ACM Trans

    Lukas Radl, Michael Steiner, Mathias Parger, Alexan- der Weinrauch, Bernhard Kerbl, and Markus Steinberger. StopThePop: Sorted Gaussian Splatting for View-Consistent Real-time Rendering.ACM Trans. Gr., 4(43), 2024. 3

    2024

  52. [52]

    Stable view synthesis

    Gernot Riegler and Vladlen Koltun. Stable view synthesis. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12216–12225, 2021. 3

    2021

  53. [53]

    Adop: Approximate differentiable one-pixel point rendering.ACM Transactions on Graphics (TOG), 41(4):1–14, 2022

    Darius R ¨uckert, Linus Franke, and Marc Stamminger. Adop: Approximate differentiable one-pixel point rendering.ACM Transactions on Graphics (TOG), 41(4):1–14, 2022. 3

    2022

  54. [54]

    Structure- from-motion revisited

    Johannes L Sch ¨onberger and Jan-Michael Frahm. Structure- from-motion revisited. InProceedings of the IEEE con- ference on computer vision and pattern recognition, pages 4104–4113, 2016. 2

    2016

  55. [55]

    Pixelwise view selection for un- structured multi-view stereo

    Johannes Lutz Sch ¨onberger, Enliang Zheng, Marc Pollefeys, and Jan-Michael Frahm. Pixelwise view selection for un- structured multi-view stereo. InEuropean Conference on Computer Vision (ECCV), 2016. 2, 10

    2016

  56. [56]

    Review of image-based rendering techniques

    Harry Shum and Sing Bing Kang. Review of image-based rendering techniques. InVisual Communications and Image Processing 2000, pages 2–13. SPIE, 2000. 3

    2000

  57. [57]

    Photo tourism: exploring photo collections in 3d

    Noah Snavely, Steven M Seitz, and Richard Szeliski. Photo tourism: exploring photo collections in 3d. InACM siggraph 2006 papers, pages 835–846. 2006. 2

    2006

  58. [58]

    A compar- ative analysis of depth-discontinuity and mixed-pixel detec- tion algorithms

    Pingbo Tang, Daniel Huber, and Burcu Akinci. A compar- ative analysis of depth-discontinuity and mixed-pixel detec- tion algorithms. InSixth International Conference on 3-D Digital Imaging and Modeling (3DIM 2007), pages 29–38. IEEE, 2007. 4

    2007

  59. [59]

    Tchapmi, Vineet Kosaraju, Hamid Rezatofighi, Ian Reid, and Silvio Savarese

    Lyne P. Tchapmi, Vineet Kosaraju, Hamid Rezatofighi, Ian Reid, and Silvio Savarese. Topnet: Structural point cloud decoder. In2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 383–392, 2019. 10, 18 13

    2019

  60. [60]

    Velodyne LiDAR Inc., 2019

    Velodyne2019.VLP-16 User Manual. Velodyne LiDAR Inc., 2019. 4

    2019

  61. [61]

    Pointattn: You only need attention for point cloud completion.Proceedings of the AAAI Con- ference on Artificial Intelligence, 38(6):5472–5480, 2024

    Jun Wang, Ying Cui, Dongyan Guo, Junxia Li, Qingshan Liu, and Chunhua Shen. Pointattn: You only need attention for point cloud completion.Proceedings of the AAAI Con- ference on Artificial Intelligence, 38(6):5472–5480, 2024. 2, 10, 17, 18

    2024

  62. [62]

    Neus: Learning neural im- plicit surfaces by volume rendering for multi-view recon- struction.Advances in Neural Information Processing Sys- tems, 34:27171–27183, 2021

    Peng Wang, Lingjie Liu, Yuan Liu, Christian Theobalt, Taku Komura, and Wenping Wang. Neus: Learning neural im- plicit surfaces by volume rendering for multi-view recon- struction.Advances in Neural Information Processing Sys- tems, 34:27171–27183, 2021. 3

    2021

  63. [63]

    Neus2: Fast learning of neural implicit surfaces for multi-view recon- struction

    Yiming Wang, Qin Han, Marc Habermann, Kostas Dani- ilidis, Christian Theobalt, and Lingjie Liu. Neus2: Fast learning of neural implicit surfaces for multi-view recon- struction. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 3295–3306, 2023. 2, 3

    2023

  64. [64]

    4d gaussian splatting for real-time dynamic scene rendering

    Guanjun Wu, Taoran Yi, Jiemin Fang, Lingxi Xie, Xiaopeng Zhang, Wei Wei, Wenyu Liu, Qi Tian, and Xinggang Wang. 4d gaussian splatting for real-time dynamic scene rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20310–20320, 2024. 3

    2024

  65. [65]

    Recent advances in 3d gaussian splatting.Computational Visual Media, 10(4):613– 642, 2024

    Tong Wu, Yu-Jie Yuan, Ling-Xiao Zhang, Jie Yang, Yan- Pei Cao, Ling-Qi Yan, and Lin Gao. Recent advances in 3d gaussian splatting.Computational Visual Media, 10(4):613– 642, 2024. 3 [66]lixel l2. XGRIDS, 2024. https://www.xgrids.com/lixell2. 7

    2024

  66. [67]

    Gaussian- room: Improving 3d gaussian splatting with sdf guidance and monocular cues for indoor scene reconstruction.arXiv preprint arXiv:2405.19671, 2024

    Haodong Xiang, Xinghui Li, Xiansong Lai, Wanting Zhang, Zhichao Liao, Kai Cheng, and Xueping Liu. Gaussian- room: Improving 3d gaussian splatting with sdf guidance and monocular cues for indoor scene reconstruction.arXiv preprint arXiv:2405.19671, 2024. 5

    Pith/arXiv arXiv 2024

  67. [68]

    SnowflakeNet: Point cloud completion by snowflake point deconvolution with skip-transformer

    Peng Xiang, Xin Wen, Yu-Shen Liu, Yan-Pei Cao, Pengfei Wan, Wen Zheng, and Zhizhong Han. SnowflakeNet: Point cloud completion by snowflake point deconvolution with skip-transformer. InProceedings of the IEEE International Conference on Computer Vision (ICCV), 2021. 2, 10, 17, 18

    2021

  68. [69]

    Learning compact representations for lidar comple- tion and generation

    Yuwen Xiong, Wei-Chiu Ma, Jingkang Wang, and Raquel Urtasun. Learning compact representations for lidar comple- tion and generation. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 1074–1083, 2023. 2

    2023

  69. [70]

    Multi-scale geometric consistency guided multi-view stereo

    Qingshan Xu and Wenbing Tao. Multi-scale geometric consistency guided multi-view stereo. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5483–5492, 2019. 2

    2019

  70. [71]

    Multi-scale geometric consistency guided and planar prior assisted multi-view stereo.IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 45(4):4945–4963,

    Qingshan Xu, Weihang Kong, Wenbing Tao, and Marc Polle- feys. Multi-scale geometric consistency guided and planar prior assisted multi-view stereo.IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 45(4):4945–4963,

  71. [72]

    Point-nerf: Point-based neural radiance fields

    Qiangeng Xu, Zexiang Xu, Julien Philip, Sai Bi, Zhixin Shu, Kalyan Sunkavalli, and Ulrich Neumann. Point-nerf: Point-based neural radiance fields. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5438–5448, 2022. 3

    2022

  72. [73]

    Depth completion from sparse li- dar data with depth-normal constraints

    Yan Xu, Xinge Zhu, Jianping Shi, Guofeng Zhang, Hujun Bao, and Hongsheng Li. Depth completion from sparse li- dar data with depth-normal constraints. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 2811–2820, 2019. 2

    2019

  73. [74]

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

    Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything: Unleashing the power of large-scale unlabeled data. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10371–10381, 2024. 16

    2024

  74. [75]

    Mvsnet: Depth inference for unstructured multi-view stereo

    Yao Yao, Zixin Luo, Shiwei Li, Tian Fang, and Long Quan. Mvsnet: Depth inference for unstructured multi-view stereo. InProceedings of the European Conference on Computer Vi- sion (ECCV), pages 767–783, 2018. 2

    2018

  75. [76]

    V ol- ume rendering of neural implicit surfaces.Advances in Neu- ral Information Processing Systems, 34:4805–4815, 2021

    Lior Yariv, Jiatao Gu, Yoni Kasten, and Yaron Lipman. V ol- ume rendering of neural implicit surfaces.Advances in Neu- ral Information Processing Systems, 34:4805–4815, 2021. 3

    2021

  76. [77]

    Bakedsdf: Meshing neural sdfs for real- time view synthesis

    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. InACM SIGGRAPH 2023 Conference Proceedings, pages 1–9, 2023. 3

    2023

  77. [78]

    Scannet++: A high-fidelity dataset of 3d indoor scenes

    Chandan Yeshwanth, Yueh-Cheng Liu, Matthias Nießner, and Angela Dai. Scannet++: A high-fidelity dataset of 3d indoor scenes. InProceedings of the International Confer- ence on Computer Vision (ICCV), 2023. 1, 4, 7, 16

    2023

  78. [79]

    Up-sampling method for low-resolution lidar point cloud to enhance 3d object detec- tion in an autonomous driving environment.Sensors, 23(1): 322, 2022

    Jihwan You and Young-Keun Kim. Up-sampling method for low-resolution lidar point cloud to enhance 3d object detec- tion in an autonomous driving environment.Sensors, 23(1): 322, 2022. 4

    2022

  79. [80]

    Pu-net: Point cloud upsampling network

    Lequan Yu, Xianzhi Li, Chi-Wing Fu, Daniel Cohen-Or, and Pheng-Ann Heng. Pu-net: Point cloud upsampling network. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 2790–2799, 2018. 2

    2018

  80. [81]

    Gsdf: 3dgs meets sdf for improved rendering and reconstruction.arXiv preprint arXiv:2403.16964, 2024

    Mulin Yu, Tao Lu, Linning Xu, Lihan Jiang, Yuanbo Xiangli, and Bo Dai. Gsdf: 3dgs meets sdf for improved rendering and reconstruction.arXiv preprint arXiv:2403.16964, 2024. 3

    Pith/arXiv arXiv 2024

Showing first 80 references.