Pith. sign in

REVIEW 4 major objections 5 minor 52 references

Rethinking 3D Segmentation from Individual LiDAR Scans: Incidence-Aware Sampling on the SIP Benchmark

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Downsampling individual LiDAR scans should be an active part of 3D segmentation, and resampling by incidence and range improves thin-object segmentation such as ladders.

desk verdict Useful sampling front-end for single-scan LiDAR with solid repeated-run evidence, but the mechanism is under-tested and key hyperparameters are missing. read the letter →

arxiv 2608.07757 v1 pith:IJYDQ6PV submitted 2026-08-07 cs.CV

Seongyong Kim , Jingdao Chen , Yong Kwon Cho This is my paper · ORCID
classification cs.CV
keywords 3Dsemanticsegmentationsingle-scanLiDARpointclouddownsamplingincidence-awaresamplingconstructionsitesceneslong-tailedclassesmanifoldvoxelizationSIPbenchmark
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's central claim is that when individual LiDAR scans are downsampled, the sampling step itself is a source of representation bias, and that this bias can be corrected before learning by reweighting point allocation according to how each surface was seen by the sensor. It proposes an incidence-aware manifold sampling scheme that moves each point by $\sqrt{\gamma}(r_0/r)^\beta$ in a normalized coordinate space, voxelizes there, and then keeps the original Euclidean coordinates of the selected points. On the SIP construction benchmark, the scheme improves resolution-averaged segmentation for two hierarchical point-based backbones, with the largest gains for ladders and other non-planar elements, and reduces sensitivity to the choice of sampling resolution. If the claim holds, point allocation should be treated as an active component of single-scan 3D segmentation rather than as a neutral preprocessing step.

What carries the argument

The load-bearing object is the incidence-aware manifold mapping $T(x)=x\sqrt{\gamma(x)}(r_0/r(x))^\beta$, combined with voxel-based cell selection in the transformed space. It converts sensor-centered density bias into a coordinate-space correction: because the local area element in the transformed space satisfies $dA'=w(x)dA$ with $w=\gamma(r_0/r)^2$, a standard voxel grid in the transformed space gives each occupied cell a more uniform share of surface support. The original coordinates are never replaced, so downstream geometric context is preserved.

What would settle it

Replace the scanner-provided normals in the SIP experiments with normals estimated from local neighborhoods; if the reported ladder IoU gains of roughly 16 points disappear or reverse, the dependence on reliable normals would be demonstrated.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that acquisition geometry, measured by the incidence term $\gamma(x)=|n(x)^\top v(x)|$ and range $r(x)$, can be folded into the sampling coordinate system so that voxel cells correspond more evenly to observed surface area. The proposed mapping is $x'=x\sqrt{\gamma(x)}(r_0/r(x))^\beta$; voxelization in the transformed space selects points, while the original Euclidean coordinates are fed to the network. Under the smooth-surface assumption, the Jacobian on the tangent plane is approximately $c(x)I_2$ with $c(x)=\sqrt{\gamma(x)}r_0/r(x)$, so the area element rescales by $w(x)=\gamma(x)(r_0/r(x))^2$, which is the mechanism claimed to counter the over-representation of grazing and near-range points. In the SIP experiments at five resolutions, manifold sampling raises resolution-averaged mIoU from 64.0 to 68.7 for Point Transformer and from 41.6 to 44.7 for PointNeXt, with ladder IoU gains of 16.3 and 16.9 points respectively.

Load-bearing premise

The load-bearing premise is that surface normals are trustworthy and that the surface is locally smooth enough for incidence and range to vary slowly across each neighborhood; the places where the method gains the most, such as ladders, are exactly where single-scan normals are noisiest and surfaces are most fragmented.

Editorial extensions

If this is right

  • Sampling becomes an input-design decision: under a fixed per-fragment point budget, incidence-aware allocation raises resolution-averaged mIoU by 4.7 points for Point Transformer and 3.1 points for PointNeXt.
  • Thin and sparsely represented elements benefit disproportionately: resolution-averaged ladder IoU rises by 16.3 points for Point Transformer and 16.9 points for PointNeXt, and non-planar NP-IoU by 8.1 and 5.5 points.
  • The method stabilizes performance across sampling resolutions; the clearest case is the coarsest tested setting, where Point Transformer mIoU at 0.18 m improves from 53.5 with grid sampling to 65.9 with manifold sampling.
  • Because the front end needs only point coordinates and normals and keeps the original Euclidean coordinates of selected points, it can be inserted into existing point-based segmentation pipelines without changing the backbone.
  • The gain is not architecture-independent: with the sparse-convolution model SpUNet, pure manifold sampling lowers mIoU, while a mixed variant that retains grid samples roughly restores the baseline, so voxel-lattice backbones need a lattice-preserving sampling variant.

Reading between the lines

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

  • Beyond the paper: the same incidence-and-range reweighting could be applied to other sparse-sensor tasks such as 3D object detection or instance segmentation, where thin structures are also underrepresented.
  • Beyond the paper: the parameters $\beta$ and $r_0$ are hand-set, so a data-driven or sensor-calibrated choice of these values could adapt the method to other scanner placements; the paper does not test this.
  • Beyond the paper: the reported confusions between stair railings, pipes, and ladders suggest that sampling alone cannot separate classes with similar local geometry, so combining incidence-aware allocation with height or contextual cues is a natural next step.
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 an incidence-aware sampling method for single-scan LiDAR point clouds. Points are mapped to a geometry-normalized space via a radial scaling that depends on the incidence term gamma(x) = |n(x)^T v(x)| and range r(x), then standard voxel-based selection is performed in that transformed space while the original Euclidean coordinates are retained for downstream segmentation. The method is evaluated on the SIP benchmark (an indoor construction-site single-scan LiDAR dataset) using Point Transformer and PointNeXt under a fixed per-fragment point budget across five sampling resolutions. The central empirical claim is that the proposed sampling improves resolution-averaged mIoU (by 4.7 points for PT and 3.1 points for PNxt) and NP-IoU, with the largest gains on ladders, and that it reduces sensitivity to sampling resolution. The paper also compares against farthest-point sampling and a lattice-preserving variant, and observes that pure manifold sampling hurts a sparse-convolution backbone (SpUNet).

Significance. If the empirical claims hold, the paper makes a useful case that point-allocation strategy is an active component of single-scan 3D segmentation rather than a neutral preprocessing step. The strengths include a controlled fixed-budget evaluation protocol, repeated runs with mean and standard deviation, two point-based backbones, a publicly available dataset and code, and a lightweight front-end method that requires no backbone modification. The resolution-averaged gains, particularly for non-planar classes, are non-trivial. However, the significance is tempered by the lack of an ablation isolating the incidence term from the radial term, the absence of reported values for the method's free parameters (beta and r0), and the small number of test scans underlying the headline ladder gains.

major comments (4)
  1. [Section 3.2 and Table 6] The Proposition in Section 3.2 is essentially a restatement of the mapping definition: it derives the area-element scaling w(x) = gamma(x)(r0/r)^2 from the local Jacobian approximation, but it does not establish the claimed 'more consistent correspondence between occupied cells and represented surface support.' The proof shows only that the area element transforms; the segmentation benefit is an empirical hypothesis. More importantly, the experiments do not isolate the incidence term from the radial term. The mapping in Algorithm 1 includes both gamma and (r0/r)^beta, and the radial term alone could reproduce the density redistribution that drives the gains. An ablation with gamma set to 1 (or with beta = 0) is needed to support the 'incidence-aware' attribution and to explain the ladder results in Table 6.
  2. [Section 3.3, Algorithm 1, and Table 3] The values of the free parameters beta and r0 are never reported in the experimental section. Algorithm 1 takes both as inputs, and Section 3.3 states that beta controls the strength of radial normalization and that h is interpreted at reference range r0, but the manuscript does not say what values were used for beta and r0 in the five-resolution experiments. Without these values, the results cannot be reproduced, and the sensitivity of the method to these parameters is unknown. Please report the chosen values and, ideally, a sensitivity study over beta.
  3. [Section 3.1 and Algorithm 1] The method critically depends on surface normals, because gamma = |n^T v| is computed from unit normals. The paper never states how normals are obtained for the SIP dataset: whether they are provided by the sensor, estimated from neighborhoods, or computed with a specific estimator and radius. Since the incidence term is the core of the proposed mechanism, the normal estimation procedure (and any preprocessing such as orientation consistency) must be described. A normal-ablation study would also address the concern that the ladder gains are driven by noisy normals on thin structures.
  4. [Section 4.2 and Tables 2, 6] The paper states that a validation set is not feasible due to rare-class coverage, but this means the hyperparameters of the proposed method (beta, r0, and any design choices) are not selected on held-out data. The test set contains only 6 scans, and the ladder class appears in only 3 test scans (Table 2). The headline resolution-averaged ladder IoU gains of +16.3 and +16.9 points in Table 6 are therefore based on a very small number of test instances. Please report per-scan ladder IoU or a bootstrap confidence interval for the resolution-averaged ladder gain, and clarify whether the reported setting corresponds to any hyperparameter search on the test split.
minor comments (5)
  1. [Section 5.2] The notation 'r04 to r09' is used in the text but never defined; please define r04 = 0.04 m, r06 = 0.06 m, etc., at the first use.
  2. [Tables 4 and 5] The table headers use 'mIOU' while the text uses 'mIoU'; please standardize the capitalization throughout.
  3. [Figure 6] The caption refers to 'Manifold (ours)', but the method is not explicitly named in the text; consider introducing a short name (e.g., 'incidence-aware sampling, IAS') and using it consistently in captions and tables.
  4. [Table 7(a)] The footnote states that FPS is evaluated only on sampled points while other methods are evaluated after inverse mapping to the original scan. This makes the comparison not apples-to-apples; please explain why FPS cannot be inverse-mapped and discuss how this affects the comparison.
  5. [Algorithm 1] Line 7 uses spherical bin indices B_r, B_theta, B_phi and a hash K, but these symbols are not defined anywhere. Please clarify the hashing/voxelization step or refer to a standard grid hashing formulation.

Circularity Check

1 steps flagged · score 2.0 of 10

Minor definitional circularity in the Section 3.2 Proposition; the central empirical claim is independently tested and not circular.

  1. self definitional [Section 3.2 (Manifold Mapping under Incidence-Induced Density Scaling), definition of T(x), w(x), and the Proposition/Proof]
    "Let 𝑀 ⊂ ℝ3 denote the observed surface manifold underlying a single-scan LiDAR point cloud. We define a geometry-normalized space 𝑀′ through the mapping 𝑇: 𝑀 → 𝑀′, 𝑥′ = 𝑇(𝑥) = 𝑥√𝛾(𝑥) 𝑟0/𝑟(𝑥) ... Define 𝑤(𝑥) = 𝛾(𝑥) ( 𝑟0/𝑟(𝑥))2. ... Proof. ... 𝑑𝐴′ ≈∣ det(𝐽𝑇 ∣𝑇𝑥𝑀) ∣ 𝑑𝐴 ≈ 𝑐(𝑥)2𝑑𝐴 = 𝑤(𝑥) 𝑑𝐴."

    The Proposition's claimed area reweighting is obtained by squaring the scaling factor c(x) = sqrt(gamma(x)) r0/r(x) that was already used to define the transformation T. The proof is therefore a Jacobian restatement of the definition of w, not an independent geometric result. The 'consistent correspondence between occupied cells and surface support' is built into the choice of T by construction. This is a minor definitional circularity because the paper presents the Proposition as a formal justification, but the actual support for the method is the empirical segmentation comparison in Tables 4-6, which does not depend on this proposition.

full rationale

The only load-bearing derivation that reduces to its own input is the Section 3.2 Proposition, which restates the consequences of the chosen mapping T(x) = x sqrt(gamma(x)) r0/r(x). The area element identity dA' = w dA follows directly from the definition of w, so the proposition is a mathematical consequence of the construction rather than independent evidence. However, the paper's central claim is the resolution-averaged segmentation improvement observed on the SIP benchmark, which is tested empirically against Euclidean grid sampling and a farthest-point-sampling reference under a fixed per-fragment point budget. That comparison is self-contained and not derived from the proposition. The SIP benchmark is cited from the authors' prior work, but it is an externally available dataset with published data and code access, so this self-citation is not load-bearing in the argument. Concerns about noisy normals on thin structures and the smoothness assumptions failing for ladders are legitimate correctness risks, but they are limitations of the method's applicability, not circularity. Overall, the derivation chain contains one minor by-construction step, but the main empirical findings stand on independent experimental evidence.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim depends on two unreported hyperparameters, beta and r0, plus assumptions about normal quality and local surface smoothness. No new physical entities are introduced.

free parameters (2)
  • beta = not reported
    Radial normalization exponent introduced in Section 3.3. Algorithm 1 requires beta, but the paper never states the value used in any experiment or configuration table.
  • r0 = not reported
    Reference range in the manifold mapping defined in Section 3.2 and used in Algorithm 1. The value is never reported, yet it changes the transformed coordinates and therefore the voxel selection.
assumptions (4)
  • domain assumption Local smoothness of the observed surface manifold M, and smooth variation of gamma and r over the tangent plane
    The Proposition in Section 3.2 assumes gamma and r are approximately constant on the local tangent plane so the Jacobian is c(x) I2. Near sharp edges, depth discontinuities, and thin structures this assumption fails.
  • domain assumption Unit surface normals are available and reliable
    Algorithm 1 step 2 computes gamma = |N^T V| from normals. On sparse, thin, single-scan elements, normal estimation is noisy, which directly affects the transformation.
  • standard math Standard Jacobian and area-element calculus on smooth manifolds
    Used in Section 3.2 to derive dA' = w dA. This is a routine mathematical result and is not contentious.
  • domain assumption Point density in single-scan LiDAR decays approximately with the square of range
    Motivates the radial factor (r0/r)^2 in the weight w. This is a standard sensor model but is only an approximation in cluttered, occluded construction scenes.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Rethinking 3D Segmentation from Individual LiDAR Scans: Incidence-Aware Sampling on the SIP Benchmark." pith.science (2026). https://pith.science/paper/IJYDQ6PV

@misc{pith2026260807757,
  author       = {Pith},
  title        = {Pith review of: Rethinking 3D Segmentation from Individual LiDAR Scans: Incidence-Aware Sampling on the SIP Benchmark},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IJYDQ6PV}},
  note         = {Machine review of arXiv:2608.07757}
}
read the original abstract

3D scene understanding is increasingly important in construction, yet most methods are developed on curated datasets that do not fully reflect real site sensing conditions. In many workflows, individual LiDAR scans provide rapid local updates rather than complete scene representations, producing limited surface coverage, acquisition-driven density variation, and severe imbalance between dominant planar surfaces and sparse construction elements. Because large point clouds must be downsampled, sampling resolution and point allocation directly affect the balance between geometric detail and spatial context. This study evaluates these effects under a fixed per-fragment point budget and introduces an incidence-aware sampling strategy for individual LiDAR scans. The method maps points to a geometry-normalized manifold space for voxel-based selection while preserving original Euclidean coordinates for downstream learning. It requires only point coordinates and normals and no backbone modification. Using the Site in Pieces (SIP) benchmark, experiments with Point Transformer and PointNeXt show improved resolution-averaged segmentation performance, especially for non-planar elements and ladders, while reducing sensitivity to sampling resolution. The results show that acquisition-aware sampling can provide a more stable geometric representation and should be treated as an active component of individual-scan 3D segmentation rather than generic preprocessing.

Figures

Figures reproduced from arXiv: 2608.07757 by the authors.

Figure 1
Figure 1. Incidence-aware sampling for 3D training from single-scan LiDAR. Points are sampled to promote consistent coverage over the surface manifold, improving the balance of geometric cues. Compared to standard voxelization, planar regions are reduced while structurally thin regions are preserved. Color encodes normalized incidence, with red indicating grazing angles and blue indicating near-orthogonal surfaces. 2. Related… view at source ↗
Figure 2
Figure 2. Training and inference pipeline for SIP. Underlined text indicates training-only components. 5. Experimental Results [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 45 canonical work pages

  1. [1]

    Big Data in the construction industry: A review of present status, opportunities, and future trends,

    M. Bilal et al., “Big Data in the construction industry: A review of present status, opportunities, and future trends,” Advanced engineering informatics, vol. 30, no. 3, pp. 500–521, 2016

    work page 2016

  2. [2]

    Classical strategies include random sampling, farthest-point sampling, and voxel- or grid-based sampling, which remain widely used for their simplicity and efficiency

    Related Work 2.1 Point Cloud Sampling Methods 5 Point cloud sampling is a fundamental component of 3D processing pipelines, serving to reduce computational cost while retaining useful geometric information. Classical strategies include random sampling, farthest-point sampling, and voxel- or grid-based sampling, which remain widely used for their simplicit...

    work page

  3. [3]

    Geometry Intuition: Sensor-Centered Sampling Bias Single-scan LiDAR point clouds do not represent volumetric occupancy

    Incidence-Aware Sampling via Surface Manifolds 3.1. Geometry Intuition: Sensor-Centered Sampling Bias Single-scan LiDAR point clouds do not represent volumetric occupancy. Instead, they consist only of points sampled from surfaces visible to the sensor, making the resulting distribution inherently sensor -dependent and non -uniform. This non -uniformity i...

    work page

  4. [4]

    Benchmark Protocol on SIP 4.1 Dataset Characteristics The SIP dataset [13] is a construction -focused benchmark of single -station terrestrial LiDAR scans designed for 3D segmentation, from which the indoor subset is used in this study. SIP preserves the raw characteristics of single-scan acquisition in active construction sites: • Geometry-driven samplin...

    work page

  5. [5]

    Design considerations The maximum number of points that can be processed in a scene fragment defines the primary computational constraint in large -scale 3D segmentation

    Experimental Results 15 5.1. Design considerations The maximum number of points that can be processed in a scene fragment defines the primary computational constraint in large -scale 3D segmentation. Accordingly, all configurations are evaluated under the same per-fragment point budget. Sampling resolution and point allocation are not independently normal...

    work page

  6. [6]

    Reference comparison with FPS Figure 9 compares the retained-point distributions produced by grid sampling, farthest point sampling (FPS), manifold sampling, and manifold+

    DISCUSSION 6.1. Reference comparison with FPS Figure 9 compares the retained-point distributions produced by grid sampling, farthest point sampling (FPS), manifold sampling, and manifold+. Manifold+ combines Euclidean grid samples with manifold -selected points to preserve lattice coverage while incorp orating incidence-aware allocation. FPS promotes broa...

    work page

  7. [7]

    CONCLUSION This study reframed individual construction LiDAR scans as local site evidence for scene understanding and examined the effects of sampling design under realistic sensing and input constraints. To address acquisition-driven density bias, this study introduced incidence -aware manifold sampling as a lightweight front-end normalization strategy t...

    work page

  8. [8]

    BIM information integration based VR modeling in digital twins in industry 5.0,

    W. Wang et al., “BIM information integration based VR modeling in digital twins in industry 5.0,” J. Ind. Inf. Integr., vol. 28, p. 100351, 2022. 25

    work page 2022

Show all 52 references
  1. [9]

    Towards big data driven construction industry,

    F. Li et al., “Towards big data driven construction industry,” J. Ind. Inf. Integr., vol. 35, p. 100483, 2023

    2023

  2. [10]

    The future of construction automation: Technological disruption and the upcoming ubiquity of robotics,

    T. Bock, “The future of construction automation: Technological disruption and the upcoming ubiquity of robotics,” Autom. Constr., vol. 59, pp. 113–121, 2015

    2015

  3. [11]

    Future of robotics and automation in construction,

    B. G. de Soto and M. J. Skibniewski, “Future of robotics and automation in construction,” in Construction 4.0, Routledge, 2020, pp. 289–306

    2020

  4. [12]

    Review of image-based 3D reconstruction of building for automated construction progress monitoring,

    J. Xue, X. Hou, and Y. Zeng, “Review of image-based 3D reconstruction of building for automated construction progress monitoring,” Applied Sciences, vol. 11, no. 17, p. 7840, 2021

    2021

  5. [13]

    Automated continuous construction progress monitoring using multiple workplace real time 3D scans,

    Z. Pučko, N. Šuman, and D. Rebolj, “Automated continuous construction progress monitoring using multiple workplace real time 3D scans,” Advanced Engineering Informatics, vol. 38, pp. 27– 40, 2018

    2018

  6. [14]

    A framework for dimensional and surface quality assessment of precast concrete elements using BIM and 3D laser scanning,

    M.-K. Kim, J. C. P. Cheng, H. Sohn, and C.-C. Chang, “A framework for dimensional and surface quality assessment of precast concrete elements using BIM and 3D laser scanning,” Autom. Constr., vol. 49, pp. 225–238, 2015

    2015

  7. [15]

    An adaptive down-sampling method of laser scan data for scan-to-BIM,

    Q. Qiu, M. Wang, J. Guo, Z. Liu, and Q. Wang, “An adaptive down-sampling method of laser scan data for scan-to-BIM,” Autom. Constr., vol. 135, p. 104135, 2022

    2022

  8. [16]

    Construction quality assessment using 3D as-built models generated with Project Tango,

    T. S. Kalyan, P. A. Zadeh, S. Staub-French, and T. M. Froese, “Construction quality assessment using 3D as-built models generated with Project Tango,” Procedia Eng., vol. 145, pp. 1416–1423, 2016

    2016

  9. [18]

    Planning for terrestrial laser scanning in construction: A review,

    A. Aryan, F. Bosché, and P. Tang, “Planning for terrestrial laser scanning in construction: A review,” Autom. Constr., vol. 125, p. 103551, 2021

    2021

  10. [19]

    A review of point cloud segmentation for understanding 3D indoor scenes,

    Y. Sun, X. Zhang, and Y. Miao, “A review of point cloud segmentation for understanding 3D indoor scenes,” Visual Intelligence, vol. 2, no. 1, p. 14, 2024

    2024

  11. [20]

    SIP: Site in Pieces-A Dataset of Disaggregated Construction-Phase 3D Scans for Semantic Segmentation and Scene Understanding,

    S. Kim and Y. K. Cho, “SIP: Site in Pieces-A Dataset of Disaggregated Construction-Phase 3D Scans for Semantic Segmentation and Scene Understanding,” arXiv preprint arXiv:2512.09062, 2025

    2025 arXiv

  12. [21]

    Dynamic downsampling algorithm for 3D point cloud map based on voxel filtering,

    W. Lyu, W. Ke, H. Sheng, X. Ma, and H. Zhang, “Dynamic downsampling algorithm for 3D point cloud map based on voxel filtering,” Applied Sciences, vol. 14, no. 8, p. 3160, 2024

    2024

  13. [22]

    Scannet: Richly- annotated 3d reconstructions of indoor scenes,

    A. Dai, A. X. Chang, M. Savva, M. Halber, T. Funkhouser, and M. Nießner, “Scannet: Richly- annotated 3d reconstructions of indoor scenes,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 5828–5839

    2017

  14. [23]

    Adaptive hierarchical down-sampling for point cloud classification,

    E. Nezhadarya, E. Taghavi, R. Razani, B. Liu, and J. Luo, “Adaptive hierarchical down-sampling for point cloud classification,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 12956–12964

    2020

  15. [24]

    Deep learning for 3d point clouds: A survey,

    Y. Guo, H. Wang, Q. Hu, H. Liu, L. Liu, and M. Bennamoun, “Deep learning for 3d point clouds: A survey,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 43, no. 12, pp. 4338–4364, 2020

    2020

  16. [25]

    Samplenet: Differentiable point cloud sampling,

    I. Lang, A. Manor, and S. Avidan, “Samplenet: Differentiable point cloud sampling,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 7578–7588

    2020

  17. [26]

    Pointasnl: Robust point clouds processing using nonlocal neural networks with adaptive sampling,

    X. Yan, C. Zheng, Z. Li, S. Wang, and S. Cui, “Pointasnl: Robust point clouds processing using nonlocal neural networks with adaptive sampling,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 5589–5598

    2020

  18. [27]

    Lsnet: Learned sampling network for 3d object detection from point clouds,

    M. Wang, Q. Chen, and Z. Fu, “Lsnet: Learned sampling network for 3d object detection from point clouds,” Remote Sens. (Basel)., vol. 14, no. 7, p. 1539, 2022

    2022

  19. [28]

    3d semantic parsing of large-scale indoor spaces,

    I. Armeni et al., “3d semantic parsing of large-scale indoor spaces,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 1534–1543

    2016

  20. [29]

    Scalability in perception for autonomous driving: Waymo open dataset,

    P. Sun et al., “Scalability in perception for autonomous driving: Waymo open dataset,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 2446–2454. 26

    2020

  21. [30]

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

    C. Yeshwanth, Y.-C. Liu, M. Nießner, and A. Dai, “Scannet++: A high-fidelity dataset of 3d indoor scenes,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 12–22

    2023

  22. [31]

    Matterport3d: Learning from rgb-d data in indoor environments,

    A. Chang et al., “Matterport3d: Learning from rgb-d data in indoor environments,” arXiv preprint arXiv:1709.06158, 2017

    2017 arXiv

  23. [32]

    Structured3d: A large photo-realistic dataset for structured 3d modeling,

    J. Zheng, J. Zhang, J. Li, R. Tang, S. Gao, and Z. Zhou, “Structured3d: A large photo-realistic dataset for structured 3d modeling,” in European Conference on Computer Vision, Springer, 2020, pp. 519–535

    2020

  24. [33]

    Arkitscenes: A diverse real-world dataset for 3d indoor scene understanding using mobile rgb-d data,

    G. Baruch et al., “Arkitscenes: A diverse real-world dataset for 3d indoor scene understanding using mobile rgb-d data,” arXiv preprint arXiv:2111.08897, 2021

    2021 arXiv

  25. [34]

    Semantickitti: A dataset for semantic scene understanding of lidar sequences,

    J. Behley et al., “Semantickitti: A dataset for semantic scene understanding of lidar sequences,” in Proceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 9297–9307

    2019

  26. [35]

    nuscenes: A multimodal dataset for autonomous driving,

    H. Caesar et al., “nuscenes: A multimodal dataset for autonomous driving,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 11621–11631

    2020

  27. [36]

    Stratified transformer for 3d point cloud segmentation,

    X. Lai et al., “Stratified transformer for 3d point cloud segmentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 8500–8509

    2022

  28. [37]

    Nothing stands still: A spatiotemporal benchmark on 3d point cloud registration under large geometric and temporal change,

    T. Sun et al., “Nothing stands still: A spatiotemporal benchmark on 3d point cloud registration under large geometric and temporal change,” arXiv preprint arXiv:2311.09346, 2023

    2023 arXiv

  29. [38]

    4d spatio-temporal convnets: Minkowski convolutional neural networks,

    C. Choy, J. Gwak, and S. Savarese, “4d spatio-temporal convnets: Minkowski convolutional neural networks,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 3075–3084

    2019

  30. [39]

    Searching efficient 3d architectures with sparse point-voxel convolution,

    H. Tang et al., “Searching efficient 3d architectures with sparse point-voxel convolution,” in European conference on computer vision, Springer, 2020, pp. 685–702

    2020

  31. [40]

    Point transformer,

    H. Zhao, L. Jiang, J. Jia, P. H. S. Torr, and V. Koltun, “Point transformer,” in Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 16259–16268

    2021

  32. [42]

    Swin3d: A pretrained transformer backbone for 3d indoor scene understanding,

    Y.-Q. Yang et al., “Swin3d: A pretrained transformer backbone for 3d indoor scene understanding,” Comput. Vis. Media (Beijing)., vol. 11, no. 1, pp. 83–101, 2025

    2025

  33. [43]

    Pointnext: Revisiting pointnet++ with improved training and scaling strategies,

    G. Qian et al., “Pointnext: Revisiting pointnet++ with improved training and scaling strategies,” Adv. Neural Inf. Process. Syst., vol. 35, pp. 23192–23204, 2022

    2022

  34. [44]

    Sonata: Self-supervised learning of reliable point representations,

    X. Wu et al., “Sonata: Self-supervised learning of reliable point representations,” in Proceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 22193–22204

    2025

  35. [45]

    Concerto: Joint 2d-3d self-supervised learning emerges spatial representations,

    Y. Zhang et al., “Concerto: Joint 2d-3d self-supervised learning emerges spatial representations,” arXiv preprint arXiv:2510.23607, 2025

    2025

  36. [46]

    Together, the two backbones allow the sampling effect to be examined across different feature-aggregation mechanisms within the hierarchical point-based model family

    is adopted to reduce interaction between the backbone and the input sampling scheme. Together, the two backbones allow the sampling effect to be examined across different feature-aggregation mechanisms within the hierarchical point-based model family. 5.2. Resolution-wise Samp...

  37. [47]

    Point-bert: Pre-training 3d point cloud transformers with masked point modeling,

    X. Yu, L. Tang, Y. Rao, T. Huang, J. Zhou, and J. Lu, “Point-bert: Pre-training 3d point cloud transformers with masked point modeling,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 19313–19322

    2022

  38. [48]

    Utonia: Toward One Encoder for All Point Clouds,

    Y. Zhang et al., “Utonia: Toward One Encoder for All Point Clouds,” arXiv preprint arXiv:2603.03283, 2026

    2026 arXiv

  39. [49]

    Towards large-scale 3d representation learning with multi-dataset point prompt training,

    X. Wu et al., “Towards large-scale 3d representation learning with multi-dataset point prompt training,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 19551–19562

    2024

  40. [50]

    The lovász-softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks,

    M. Berman, A. R. Triki, and M. B. Blaschko, “The lovász-softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 4413–4421

    2018

  41. [51]

    Pointcept: A codebase for point cloud perception research

    Pointcept Contributors, “Pointcept: A codebase for point cloud perception research.” Accessed: Dec. 07, 2025. [Online]. Available: https://github. com/Pointcept/Pointcept

    2025

  42. [52]

    Point transformer v2: Grouped vector attention and partition-based pooling,

    X. Wu, Y. Lao, L. Jiang, X. Liu, and H. Zhao, “Point transformer v2: Grouped vector attention and partition-based pooling,” Adv. Neural Inf. Process. Syst., vol. 35, pp. 33330–33342, 2022

    2022

  43. [53]

    Point transformer v3: Simpler faster stronger,

    X. Wu et al., “Point transformer v3: Simpler faster stronger,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 4840–4851

    2024

  44. [54]

    3d semantic segmentation with submanifold sparse convolutional networks,

    B. Graham, M. Engelcke, and L. Van Der Maaten, “3d semantic segmentation with submanifold sparse convolutional networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 9224–9232

    2018

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.