pith. sign in

arxiv: 2604.21356 · v1 · submitted 2026-04-23 · 💻 cs.CV

SparseGF: A Height-Aware Sparse Segmentation Framework with Context Compression for Robust Ground Filtering Across Urban to Natural Scenes

Nannan Qin , Pengjie Tao , Haiyan Guan , Zhizhong Kang , Lingfei Ma , Xiangyun Hu , Jonathan Li This is my paper

Pith reviewed 2026-05-09 22:56 UTC · model grok-4.3

classification 💻 cs.CV
keywords ground filteringpoint cloud segmentationairborne laser scanningcontext compressionheight-aware losssparse voxel networkdigital terrain modelscross-scene generalization
0
0 comments X

The pith

SparseGF combines context compression with height-aware training to separate ground from non-ground points more reliably in airborne laser scans across cities, mixed areas, and steep forests.

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

The paper develops SparseGF to solve two problems in deep-learning ground filtering of large ALS point clouds: the inability to handle wide contexts without losing detail due to compute limits, and the tendency to randomly mislabel tall objects. It introduces a convex-mirror-inspired module that shrinks broad surroundings into compact forms while keeping central geometry intact, a hybrid sparse voxel-point network to read those compressed forms, and a loss term that adds explicit elevation rules during training. If these pieces work together, the result is ground filtering that stays accurate without scene-by-scene retraining. A reader would care because better terrain models from laser data support flood mapping, infrastructure planning, and forest monitoring that currently break when methods trained on one landscape meet another.

Core claim

SparseGF is a height-aware sparse segmentation framework that addresses the context-detail dilemma and random tall-object misclassifications by condensing large contexts via a convex-mirror-inspired compression module while preserving central details, interpreting the results with a hybrid sparse voxel-point network that limits geometric distortion, and applying a height-aware loss that enforces topographic elevation priors, yielding leading performance on complex urban scenes, competitive results on mixed terrains, and moderate but non-catastrophic accuracy in densely forested steep areas across two large-scale ALS benchmarks.

What carries the argument

The convex-mirror-inspired context compression module that condenses expansive contexts into compact representations while preserving central details, combined with the height-aware loss that enforces topographic elevation priors.

If this is right

  • Large-scale ALS processing can proceed with wider context windows without proportional increases in memory or compute.
  • Digital terrain models become more consistent when moving between urban, mixed, and natural landscapes without retraining.
  • Random misclassification of buildings, trees, and other tall objects decreases through explicit use of elevation priors.
  • The same compression-plus-height approach can be tested on other point-cloud segmentation tasks that face scale and elevation issues.
  • Operational geospatial pipelines can reduce reliance on manual post-processing or scene-specific tuning.

Where Pith is reading between the lines

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

  • The compression technique could transfer to other remote-sensing segmentation problems where full-context processing is currently impossible.
  • Adding height priors might improve performance in related 3D tasks such as building extraction or vegetation classification.
  • If the method scales, it could support real-time terrain updates from repeated airborne or drone surveys.
  • Wider adoption might lower the barrier to creating unified terrain databases spanning multiple ecosystems.

Load-bearing premise

The context compression preserves central geometric details without introducing unrecoverable distortion, and the height-aware loss generalizes to suppress misclassifications across entirely unseen scenes without any scene-specific adjustments.

What would settle it

Running SparseGF on a fresh ALS dataset from a steep, densely forested region and observing a sharp drop in ground-point accuracy or a rise in tall-object misclassifications relative to prior methods would show the claimed cross-scene robustness does not hold.

Figures

Figures reproduced from arXiv: 2604.21356 by Haiyan Guan, Jonathan Li, Lingfei Ma, Nannan Qin, Pengjie Tao, Xiangyun Hu, Zhizhong Kang.

Figure 1
Figure 1. Figure 1: Overview of the proposed SparseGF framework. [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Schematic of the proposed context compression module (2D view): (a) underlying principle and (b) concrete [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the proposed hybrid sparse voxel-point network architecture. [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of the test samples from the OpenGF and adapted DFC2019 datasets. [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison between SparseGF and its backbone baseline on the OpenGF dataset. [PITH_FULL_IMAGE:figures/full_fig_p019_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison between SparseGF and its backbone baseline in four local areas. [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Investigation of pronounced DTM errors through profile analyses. [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative ablation results on Test II of the OpenGF dataset. “CC” denotes the context compression module, [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative HAG estimation results by SparseGF on the OpenGF dataset. [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Absolute DTM residuals for SparseGF on Test III: (a) spatial distribution and (b) histogram. [PITH_FULL_IMAGE:figures/full_fig_p025_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Qualitative comparison between SparseGF and its backbone baseline on the adapted DFC2019 dataset. [PITH_FULL_IMAGE:figures/full_fig_p029_11.png] view at source ↗
read the original abstract

High-quality digital terrain models derived from airborne laser scanning (ALS) data are essential for a wide range of geospatial analyses, and their generation typically relies on robust ground filtering (GF) to separate point clouds across diverse landscapes into ground and non-ground parts. Although current deep-learning-based GF methods have demonstrated impressive performance, especially in specific challenging terrains, their cross-scene generalization remains limited by two persistent issues: the context-detail dilemma in large-scale processing due to limited computational resources, and the random misclassification of tall objects arising from classification-only optimization. To overcome these limitations, we propose SparseGF, a height-aware sparse segmentation framework enhanced with context compression. It is built upon three key innovations: (1) a convex-mirror-inspired context compression module that condenses expansive contexts into compact representations while preserving central details; (2) a hybrid sparse voxel-point network architecture that effectively interprets compressed representations while mitigating compression-induced geometric distortion; and (3) a height-aware loss function that explicitly enforces topographic elevation priors during training to suppress random misclassification of tall objects. Extensive evaluations on two large-scale ALS benchmark datasets demonstrate that SparseGF delivers robust GF across urban to natural terrains, achieving leading performance in complex urban scenes, competitive results on mixed terrains, and moderate yet non-catastrophic accuracy in densely forested steep areas. This work offers new insights into deep-learning-based GF research and encourages further exploration toward truly cross-scene generalization for large-scale environmental monitoring.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes SparseGF, a deep-learning framework for ground filtering (GF) of airborne laser scanning (ALS) point clouds to separate ground and non-ground points across diverse landscapes. It introduces three innovations: a convex-mirror-inspired context compression module to condense large contexts while preserving central details, a hybrid sparse voxel-point network to interpret the compressed representations and mitigate geometric distortion, and a height-aware loss that incorporates topographic elevation priors to reduce random misclassifications of tall objects. Evaluations on two large-scale ALS benchmarks are reported to demonstrate robust cross-scene performance, with leading results in complex urban scenes, competitive results on mixed terrains, and moderate accuracy in densely forested steep areas.

Significance. If the central claims hold with proper validation, the work could meaningfully advance deep-learning-based ground filtering by addressing the context-detail trade-off and classification-only optimization issues that limit generalization. The proposed modules offer a concrete approach to efficient large-scale point-cloud processing and height-prior enforcement, which could improve digital terrain model generation for environmental monitoring; the emphasis on cross-scene robustness without scene-specific retraining would be a notable contribution if substantiated by appropriate splits and distortion metrics.

major comments (3)
  1. [§4 (Experiments)] §4 (Experiments): The cross-scene generalization claim requires explicit confirmation that training/test splits enforce terrain separation (e.g., train exclusively on urban data and test on natural/forested scenes). The current description does not provide this protocol detail, so the reported leading/competitive/moderate numbers cannot yet be attributed to the proposed modules rather than possible dataset overlap or mixed-distribution testing.
  2. [§3.1 (Context Compression Module)] §3.1 (Context Compression Module): No quantitative check of compression-induced geometric distortion is presented (e.g., pre/post-compression point fidelity, Chamfer distance, or normal error on recovered surfaces). Without such metrics, the claim that the hybrid voxel-point network effectively mitigates distortion remains unverified and load-bearing for the overall robustness argument.
  3. [§4.3 (Ablation Studies)] §4.3 (Ablation Studies): The abstract and results summary provide no ablation tables isolating the contribution of the context-compression module, the hybrid architecture, and the height-aware loss. This omission prevents assessment of whether each component is necessary for the reported performance gains.
minor comments (2)
  1. [Abstract] Abstract: The qualitative descriptors 'leading performance', 'competitive results', and 'moderate yet non-catastrophic accuracy' should be replaced or supplemented with concrete metrics (e.g., IoU, F1-score, or accuracy percentages with standard deviations) to enable direct comparison with prior work.
  2. [§3.3 (Height-Aware Loss)] Notation: The mathematical definition of the height-aware loss would benefit from an explicit equation showing how elevation priors are weighted relative to the standard segmentation loss, to avoid ambiguity in implementation.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and will revise the manuscript to strengthen the presentation of our claims and supporting evidence.

read point-by-point responses

  1. Referee: [§4 (Experiments)] The cross-scene generalization claim requires explicit confirmation that training/test splits enforce terrain separation (e.g., train exclusively on urban data and test on natural/forested scenes). The current description does not provide this protocol detail, so the reported leading/competitive/moderate numbers cannot yet be attributed to the proposed modules rather than possible dataset overlap or mixed-distribution testing.

    Authors: We agree that the data-split protocol must be stated explicitly to support the cross-scene generalization claims. The experiments were performed with strict terrain separation (urban scenes for training, natural/forested scenes for testing), but the manuscript description was insufficiently detailed. In the revision we will add a dedicated paragraph in §4 that specifies the exact partitioning procedure, confirms the absence of terrain overlap, and lists the scene identifiers used in each split. revision: yes

  2. Referee: [§3.1 (Context Compression Module)] No quantitative check of compression-induced geometric distortion is presented (e.g., pre/post-compression point fidelity, Chamfer distance, or normal error on recovered surfaces). Without such metrics, the claim that the hybrid voxel-point network effectively mitigates distortion remains unverified and load-bearing for the overall robustness argument.

    Authors: The referee is correct that quantitative distortion metrics would provide stronger verification of the hybrid network’s mitigation effect. Although the architecture was designed with this goal in mind, we did not report pre/post-compression fidelity measures in the initial submission. We will add these analyses—specifically Chamfer distance and surface-normal error computed on representative compressed and recovered point sets—to §3.1 and the supplementary material. revision: yes

  3. Referee: [§4.3 (Ablation Studies)] The abstract and results summary provide no ablation tables isolating the contribution of the context-compression module, the hybrid architecture, and the height-aware loss. This omission prevents assessment of whether each component is necessary for the reported performance gains.

    Authors: We acknowledge that comprehensive ablation tables are required to isolate the contribution of each proposed module. The original manuscript contained limited component-wise comparisons but lacked systematic ablation tables. We will expand §4.3 with new tables that successively disable the context-compression module, the hybrid voxel-point architecture, and the height-aware loss, reporting the resulting changes in accuracy, precision, and recall on both benchmarks. revision: yes

Circularity Check

0 steps flagged

No circularity: claims rest on novel modules and external benchmark evaluations

full rationale

The paper's derivation introduces three explicit architectural components (convex-mirror context compression, hybrid voxel-point network, height-aware loss) whose design is described independently of the target performance metrics. Performance assertions are tied to reported results on two named ALS benchmarks rather than any fitted parameter being relabeled as a prediction or any self-citation chain substituting for a proof. No equation or module definition reduces to its own output by construction, and the cross-scene claims are presented as empirical outcomes rather than tautological restatements of the inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based solely on the abstract; no explicit free parameters, axioms, or invented entities are detailed beyond high-level module descriptions. The height-aware loss may implicitly involve weighting parameters, but none are specified.

pith-pipeline@v0.9.0 · 5584 in / 1183 out tokens · 36267 ms · 2026-05-09T22:56:44.729400+00:00 · methodology

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

107 extracted references · 107 canonical work pages

  1. [1]

    Estimating wildfire fuel consumption with multitemporal airborne laser scanning data and demonstrating linkage with

    McCarley, T Ryan and Hudak, Andrew T and Sparks, Aaron M and Vaillant, Nicole M and Meddens, Arjan JH and Trader, Laura and Mauro, Francisco and Kreitler, Jason and Boschetti, Luigi , journal=. Estimating wildfire fuel consumption with multitemporal airborne laser scanning data and demonstrating linkage with

    work page

  2. [2]

    Remote Sens

    Automated forest inventory: Analysis of high-density airborne LiDAR point clouds with. Remote Sens. Environ. , volume =. 2024 , issn =

    work page 2024

  3. [3]

    The use of

    Muhadi, Nur Atirah and Abdullah, Ahmad Fikri and Bejo, Siti Khairunniza and Mahadi, Muhammad Razif and Mijic, Ana , journal=. The use of

    work page

  4. [4]

    Alin Mihu-Pintilie and Andrei Urzică and Cristian Constantin Stoleriu and Claudiu Ionut Pricop , title =. Geomat. Nat. Haz. RISK , volume =

    work page

  5. [5]

    Characterization and modeling of power line corridor elements from. ISPRS J. Photogramm. Remote Sens. , volume =. 2019 , author =

    work page 2019

  6. [6]

    2024 , issn =

    A review and future directions of techniques for extracting powerlines and pylons from LiDAR point clouds , journal =. 2024 , issn =

    work page 2024

  7. [7]

    2022 , author =

    Beyond single receptive field: A receptive field fusion-and-stratification network for airborne laser scanning point cloud classification , journal =. 2022 , author =

    work page 2022

  8. [8]

    Photogramm

    ISPRS J. Photogramm. Remote Sens. , volume =. 2024 , issn =

    work page 2024

  9. [9]

    Capture And evaluation of Airborne Laser Scanner Data , author=. Int. Arch. Photogramm. Remote Sens. , year=

    work page

  10. [10]

    An improved simple morphological filter for the terrain classification of airborne

    Pingel, Thomas J and Clarke, Keith C and McBride, William A , journal=. An improved simple morphological filter for the terrain classification of airborne

    work page

  11. [11]

    Slope based filtering of laser altimetry data , author=. Int. Arch. Photogramm. Remote Sens. , volume=

    work page

  12. [12]

    Filtering of laser altimetry data using a slope adaptive filter , author=. Int. Arch. Photogramm. Remote Sens. , volume=

    work page

  13. [13]

    Adaptive slope filtering of airborne

    Susaki, Junichi , journal=. Adaptive slope filtering of airborne

    work page

  14. [14]

    ISPRS WG III/3, III/4, V/3 Workshop ``Laserscanning 2005" , volume=

    Filtering of airborne laser scanner data based on segmented point clouds , author=. ISPRS WG III/3, III/4, V/3 Workshop ``Laserscanning 2005" , volume=

    work page 2005

  15. [15]

    and Caccetta, P

    Hingee, K. and Caccetta, P. and Caccetta, L. and Wu, X. and Devereaux, D. , TITLE =. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. , VOLUME =. 2016 , PAGES =

    work page 2016

  16. [16]

    Charles Beumier and Mahamadou Idrissa , title =. Int. J. Remote Sens. , volume =

    work page

  17. [17]

    Bartels, Marc and Wei, Hong and Mason, David C , booktitle=

    work page

  18. [18]

    IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. , volume=

    work page

  19. [19]

    Wei Su and Zhongping Sun and Ruofei Zhong and Jianxi Huang and Menglin Li and Jingguo Zhu and Keshu Zhang and Honggan Wu and Dehai Zhu , title =. Int. J. Remote Sens. , volume =

    work page

  20. [20]

    Improved progressive. ISPRS J. Photogramm. Remote Sens. , volume =. 2016 , issn =

    work page 2016

  21. [21]

    A hybrid conditional random field for estimating the underlying ground surface from airborne

    Lu, Wei-Lwun and Murphy, Kevin P and Little, James J and Sheffer, Alla and Fu, Hongbo , journal=. A hybrid conditional random field for estimating the underlying ground surface from airborne

    work page

  22. [22]

    A novel filtering algorithm for bare-earth extraction from airborne laser scanning data using an artificial neural network , author=. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. , volume=

    work page

  23. [23]

    Deep-learning-based classification for

    Hu, Xiangyun and Yuan, Yi , journal=. Deep-learning-based classification for

    work page

  24. [24]

    Remote Sens

    Rizaldy, Aldino and Persello, Claudio and Gevaert, Caroline and Oude Elberink, Sander and Vosselman, George , TITLE =. Remote Sens. , VOLUME =. 2018 , NUMBER =

    work page 2018

  25. [25]

    A deep learning approach to

    Gevaert, Caroline and Persello, Claudio and Nex, Francesco and Vosselman, George , journal=. A deep learning approach to

    work page

  26. [26]

    and Sakamoto, M

    Yotsumata, T. and Sakamoto, M. and Satoh, T. , TITLE =. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. , VOLUME =. 2020 , PAGES =

    work page 2020

  27. [27]

    Remote Sens

    Fareed, Nadeem and Flores, Joao Paulo and Das, Anup Kumar , TITLE =. Remote Sens. , VOLUME =. 2023 , NUMBER =

    work page 2023

  28. [28]

    A Point-Based Fully Convolutional Neural Network for Airborne

    Jin, Shichao and Su, Yanjun and Zhao, Xiaoqian and Hu, Tianyu and Guo, Qinghua , journal=. A Point-Based Fully Convolutional Neural Network for Airborne. 2020 , volume=

    work page 2020

  29. [29]

    2023 , issn =

    Towards intelligent ground filtering of large-scale topographic point clouds: A comprehensive survey , journal =. 2023 , issn =

    work page 2023

  30. [30]

    Choy, Christopher and Gwak, JunYoung and Savarese, Silvio , booktitle=

    work page

  31. [31]

    Qin, Nannan and Tan, Weikai and Ma, Lingfei and Zhang, Dedong and Li, Jonathan , booktitle=

    work page

  32. [32]

    2019 , volume=

    Le Saux, Bertrand and Yokoya, Naoto and Haensch, Ronny and Brown, Myron , journal=. 2019 , volume=

    work page 2019

  33. [33]

    , journal=

    Axelsson, P. , journal=

    work page

  34. [34]

    A progressive morphological filter for removing nonground measurements from airborne

    Zhang, Keqi and Chen, Shu-Ching and Whitman, Dean and Shyu, Mei-Ling and Yan, Jianhua and Zhang, Chengcui , journal=. A progressive morphological filter for removing nonground measurements from airborne

    work page

  35. [35]

    and Hudak, Andrew T

    Evans, Jeffrey S. and Hudak, Andrew T. , journal=. A Multiscale Curvature Algorithm for Classifying Discrete Return. 2007 , volume=

    work page 2007

  36. [36]

    An easy-to-use airborne

    Zhang, Wuming and Qi, Jianbo and Wan, Peng and Wang, Hongtao and Xie, Donghui and Wang, Xiaoyan and Yan, Guangjian , journal=. An easy-to-use airborne

    work page

  37. [37]

    Qi, Charles Ruizhongtai and Yi, Li and Su, Hao and Guibas, Leonidas J , journal=. Point

    work page

  38. [38]

    Thomas, Hugues and Qi, Charles R and Deschaud, Jean-Emmanuel and Marcotegui, Beatriz and Goulette, Fran. Proc. IEEE Int. Conf. Comput. Vis. , pages=

    work page

  39. [39]

    Hu, Qingyong and Yang, Bo and Xie, Linhai and Rosa, Stefano and Guo, Yulan and Wang, Zhihua and Trigoni, Niki and Markham, Andrew , booktitle=

    work page

  40. [40]

    Fan, Siqi and Dong, Qiulei and Zhu, Fenghua and Lv, Yisheng and Ye, Peijun and Wang, Fei-Yue , title =. Proc. IEEE Conf. Comput. Vis. Pattern Recog. , month =. 2021 , pages =

    work page 2021

  41. [41]

    Remote Sens

    Dai, Hengming and Hu, Xiangyun and Shu, Zhen and Qin, Nannan and Zhang, Jinming , TITLE =. Remote Sens. , VOLUME =. 2023 , NUMBER =

    work page 2023

  42. [42]

    Deep learning for filtering the ground from

    Qin, Nannan and Tan, Weikai and Ma, Lingfei and Zhang, Dedong and Guan, Haiyan and Li, Jonathan , journal=. Deep learning for filtering the ground from

    work page

  43. [43]

    2015 , volume=

    Girshick, Ross , booktitle=. 2015 , volume=

    work page 2015

  44. [44]

    Remote Sens

    Hamed. Remote Sens. Environ. , volume=

    work page

  45. [45]

    The Lovasz-Softmax Loss: A Tractable Surrogate for the Optimization of the Intersection-Over-Union Measure in Neural Networks , author=. Proc. IEEE Conf. Comput. Vis. Pattern Recog. , year=

    work page

  46. [46]

    and Briese, Christian and Rieger, W

    Pfeifer, Norbert and Reiter, T. and Briese, Christian and Rieger, W. , year =. Interpolation of high quality ground models from laser scanner data in forested areas , volume =

    work page

  47. [47]

    IEEE Geosci

    Semantic Labeling of ALS Point Cloud via Learning Voxel and Pixel Representations , author=. IEEE Geosci. Remote Sens. Lett. , volume=

    work page

  48. [48]

    Large-Scale ALS Point Cloud Segmentation via Projection-Based Context Embedding , year=

    Dai, Hengming and Hu, Xiangyun and Zhang, Jinming and Shu, Zhen and Xu, Jiabo and Du, Juan , journal=. Large-Scale ALS Point Cloud Segmentation via Projection-Based Context Embedding , year=

    work page

  49. [49]

    2018 , author =

    A multi-scale fully convolutional network for semantic labeling of 3D point clouds , journal =. 2018 , author =

    work page 2018

  50. [50]

    2019 , author =

    Addressing overfitting on point cloud classification using Atrous XCRF , journal =. 2019 , author =

    work page 2019

  51. [51]

    Point-SCT: A Multiscale Spatial Convolution-Swin Transformer Network for Point Cloud Ground Filtering in Complex Mountainous Terrains , year=

    Li, Jingxiang and Tang, Fuquan and Ma, Lingfei and Zhu, Chao and Gong, Zheng and Ruhaiyem, Nur Intan Raihana and Li, Jonathan , journal=. Point-SCT: A Multiscale Spatial Convolution-Swin Transformer Network for Point Cloud Ground Filtering in Complex Mountainous Terrains , year=

    work page

  52. [52]

    Behnaz Bigdeli and Hamed Amini Amirkolaee and Parham Pahlavani , journal=

    work page

  53. [53]

    Remote Sens

    A rapid high-resolution multi-sensory urban flood mapping framework via DEM upscaling. Remote Sens. Environ. , year=

    work page

  54. [54]

    Remote Sens

    Forested landslide detection using. Remote Sens. Environ. , volume =. 2014 , author =

    work page 2014

  55. [55]

    2011 , author =

    Airborne laser scanning of forested landslides characterization: Terrain model quality and visualization , journal =. 2011 , author =

    work page 2011

  56. [56]

    End-to-End High-Quality DEM Generation and Ground Filtering From ALS Point Clouds , year=

    Xu, Jiabo and Dai, Hengming and Su, Bo and Wang, Pengfei and Hu, Xiangyun and Ke, Tao , journal=. End-to-End High-Quality DEM Generation and Ground Filtering From ALS Point Clouds , year=

    work page

  57. [57]

    , author Arefi, H

    author Amini Amirkolaee , H. , author Arefi, H. , author Ahmadlou, M. , author Raikwar, V. , year 2022 . title DTM extraction from DSM using a multi-scale DTM fusion strategy based on deep learning . journal Remote Sens. Environ. volume 274 , pages 113014

    work page 2022

  58. [58]

    , author Indahl, U.G

    author Arief, H.A. , author Indahl, U.G. , author Strand, G.H. , author Tveite, H. , year 2019 . title Addressing overfitting on point cloud classification using atrous xcrf . journal ISPRS J. Photogramm. Remote Sens. volume 155 , pages 90--101

    work page 2019

  59. [59]

    , year 2000

    author Axelsson, P. , year 2000 . title DEM generation from laser scanner data using adaptive TIN models . journal Int. Arch. Photogramm. Remote Sens. volume XXXIII, Part B4 , pages 110--117

    work page 2000

  60. [60]

    , author Wei, H

    author Bartels, M. , author Wei, H. , author Mason, D.C. , year 2006 . title DTM generation from LiDAR data using skewness balancing , in: booktitle Proc. Int. Conf. on Pattern Recog. , pp. pages 566--569

    work page 2006

  61. [61]

    , author Triki, A.R

    author Berman, M. , author Triki, A.R. , author Blaschko, M.B. , year 2018 . title The lovasz-softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks , in: booktitle Proc. IEEE Conf. Comput. Vis. Pattern Recog. , pp. pages 4413--4421

    work page 2018

  62. [62]

    , author Idrissa, M

    author Beumier, C. , author Idrissa, M. , year 2016 . title Digital terrain models derived from digital surface model uniform regions in urban areas . journal Int. J. Remote Sens. volume 37 , pages 3477--3493

    work page 2016

  63. [63]

    , author Amirkolaee, H.A

    author Bigdeli, B. , author Amirkolaee, H.A. , author Pahlavani, P. , year 2018 . title DTM extraction under forest canopy using LiDAR data and a modified invasive weed optimization algorithm . journal Remote Sens. Environ. volume 216 , pages 289--300

    work page 2018

  64. [64]

    , author Li, X

    author Chen, W. , author Li, X. , author Wang, Y. , author Chen, G. , author Liu, S. , year 2014 . title Forested landslide detection using LiDAR data and the random forest algorithm: A case study of the Three Gorges, China . journal Remote Sens. Environ. volume 152 , pages 291--301

    work page 2014

  65. [65]

    , author Gwak, J

    author Choy, C. , author Gwak, J. , author Savarese, S. , year 2019 . title 4D spatio-temporal convnets: Minkowski convolutional neural networks , in: booktitle Proc. IEEE Conf. Comput. Vis. Pattern Recog. , pp. pages 3075--3084

    work page 2019

  66. [66]

    , author Hu, X

    author Dai, H. , author Hu, X. , author Shu, Z. , author Qin, N. , author Zhang, J. , year 2023 . title Deep ground filtering of large-scale ALS point clouds via iterative sequential ground prediction . journal Remote Sens. volume 15 , pages 961

    work page 2023

  67. [67]

    , author Hu, X

    author Dai, H. , author Hu, X. , author Zhang, J. , author Shu, Z. , author Xu, J. , author Du, J. , year 2024 . title Large-scale als point cloud segmentation via projection-based context embedding . journal IEEE Trans. Geosci. Remote Sens. volume 62 , pages 1--16

    work page 2024

  68. [68]

    , author Hudak, A.T

    author Evans, J.S. , author Hudak, A.T. , year 2007 . title A multiscale curvature algorithm for classifying discrete return LiDAR in forested environments . journal IEEE Trans. Geosci. Remote Sens. volume 45 , pages 1029--1038

    work page 2007

  69. [69]

    , author Dong, Q

    author Fan, S. , author Dong, Q. , author Zhu, F. , author Lv, Y. , author Ye, P. , author Wang, F.Y. , year 2021 . title SCF-Net : Learning spatial contextual features for large-scale point cloud segmentation , in: booktitle Proc. IEEE Conf. Comput. Vis. Pattern Recog. , pp. pages 14504--14513

    work page 2021

  70. [70]

    , author Flores, J.P

    author Fareed, N. , author Flores, J.P. , author Das, A.K. , year 2023 . title Analysis of UAS-LiDAR ground points classification in agricultural fields using traditional algorithms and PointCNN . journal Remote Sens. volume 15 , pages 483

    work page 2023

  71. [71]

    , author Persello, C

    author Gevaert, C. , author Persello, C. , author Nex, F. , author Vosselman, G. , year 2018 . title A deep learning approach to DTM extraction from imagery using rule-based training labels . journal ISPRS J. Photogramm. Remote Sens. volume 142 , pages 106--123

    work page 2018

  72. [72]

    , year 2015

    author Girshick, R. , year 2015 . title Fast R-CNN , in: booktitle Proc. IEEE Int. Conf. Comput. Vis. , pp. pages 1440--1448

    work page 2015

  73. [73]

    , author Caccetta, P

    author Hingee, K. , author Caccetta, P. , author Caccetta, L. , author Wu, X. , author Devereaux, D. , year 2016 . title Digital terrain from a two-step segmentation and outlier-based algorithm . journal Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. volume XLI-B3 , pages 233--239

    work page 2016

  74. [74]

    , author Yang, B

    author Hu, Q. , author Yang, B. , author Xie, L. , author Rosa, S. , author Guo, Y. , author Wang, Z. , author Trigoni, N. , author Markham, A. , year 2020 . title RandLA-Net : Efficient semantic segmentation of large-scale point clouds , in: booktitle Proc. IEEE Conf. Comput. Vis. Pattern Recog. , pp. pages 11105--11114

    work page 2020

  75. [75]

    , author Yuan, Y

    author Hu, X. , author Yuan, Y. , year 2016 . title Deep-learning-based classification for DTM extraction from ALS point cloud . journal Remote sens. volume 8 , pages 730

    work page 2016

  76. [76]

    , author Zoej, M.J.V

    author Jahromi, A.B. , author Zoej, M.J.V. , author Mohammadzadeh, A. , author Sadeghian, S. , year 2011 . title A novel filtering algorithm for bare-earth extraction from airborne laser scanning data using an artificial neural network . journal IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. volume 4 , pages 836--843

    work page 2011

  77. [77]

    , author Su, Y

    author Jin, S. , author Su, Y. , author Zhao, X. , author Hu, T. , author Guo, Q. , year 2020 . title A point-based fully convolutional neural network for airborne LiDAR ground point filtering in forested environments . journal IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. volume 13 , pages 3958--3974

    work page 2020

  78. [78]

    , author Haala, N

    author Kilian, J. , author Haala, N. , author Englich, M. , author Iii, C. , year 1996 . title Capture and evaluation of airborne laser scanner data . journal Int. Arch. Photogramm. Remote Sens. volume XXXI , pages 383--388

    work page 1996

  79. [79]

    , author Yokoya, N

    author Le Saux, B. , author Yokoya, N. , author Haensch, R. , author Brown, M. , year 2019 . title 2019 IEEE GRSS Data Fusion Contest : Large-scale semantic 3d reconstruction [technical committees] . journal IEEE Geosci. Remote Sens. Mag. volume 7 , pages 33--36

    work page 2019

  80. [80]

    , author Tang, F

    author Li, J. , author Tang, F. , author Ma, L. , author Zhu, C. , author Gong, Z. , author Ruhaiyem, N.I.R. , author Li, J. , year 2025 . title Point-sct: A multiscale spatial convolution-swin transformer network for point cloud ground filtering in complex mountainous terrains . journal IEEE Trans. Geosci. Remote Sens. volume 63 , pages 1--18

    work page 2025

Showing first 80 references.