pith. sign in

arxiv: 2606.10541 · v1 · pith:7ZA5RPYAnew · submitted 2026-06-09 · 💻 cs.CV

GRAR: Glass-induced Reflection Artifact Removal in LiDAR Point Clouds

Wanpeng Shao , Zeyi Guo , Bo Zhang , Yifei Xue , Tie Ji , Yizhen Lao This is my paper

Pith reviewed 2026-06-27 13:49 UTC · model grok-4.3

classification 💻 cs.CV
keywords glass-induced reflection artifactsLiDAR point cloudsterrestrial laser scanningartifact removalgeometric descriptorvision foundation modelreflection geometrypoint cloud processing
0
0 comments X

The pith

A two-stage framework detects glass regions with a vision foundation model and removes reflection artifacts using a physics-driven geometric descriptor.

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

Terrestrial laser scanning point clouds captured in cities often include false points created when laser beams reflect off glass surfaces. The paper presents GRAR, a unified method with a first stage that generates initial glass masks from a multi-modal vision foundation model, refines those masks using geometric cues, and completes missing glass areas. A second stage then applies the RE-LGGS descriptor, which measures local-global geometric similarity grounded in actual laser reflection physics to identify and eliminate the artifact points. Prior approaches relied on ideal reflection symmetry but were limited by poor glass estimation; this method addresses that gap directly. If the framework works as described, it would produce cleaner point clouds suitable for downstream urban mapping and analysis tasks.

Core claim

The central claim is that a unified two-stage framework removes glass-induced reflection artifacts from TLS point clouds: the first stage uses a multi-modal vision foundation model to produce initial glass masks refined by geometric cues and completed for no-return regions; the second stage introduces the Reflection-aware Local-Global Geometric Similarity (RE-LGGS) descriptor grounded in laser reflection geometry that jointly encodes multi-scale structures and orientation consistency via PCA-based representations, leading to consistent outperformance over state-of-the-art methods on multiple public TLS datasets.

What carries the argument

The Reflection-aware Local-Global Geometric Similarity (RE-LGGS) descriptor, which encodes multi-scale geometric structures and orientation consistency using PCA-based local shape representations based on actual laser reflection geometry.

If this is right

  • Higher-precision glass region detection directly improves the identification and removal of spurious reflection points.
  • The physics-based RE-LGGS descriptor provides robustness to imperfect observations that break ideal symmetry assumptions.
  • Glass completion recovers missing scene parts that would otherwise be lost to transparent surfaces.
  • Consistent gains across multiple public TLS datasets indicate the approach generalizes within urban scanning settings.

Where Pith is reading between the lines

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

  • Cleaned point clouds from this method could improve reliability of 3D models used in autonomous vehicle mapping through glass-heavy environments.
  • The two-stage separation of detection and geometric cleaning might transfer to removing similar reflection artifacts in mobile or aerial LiDAR systems.
  • Replacing the foundation model component with domain-specific glass detectors could test whether the geometric stage alone suffices for certain datasets.

Load-bearing premise

The multi-modal vision foundation model produces initial glass masks accurate enough that geometric refinement and completion can support effective downstream artifact removal.

What would settle it

Applying the full pipeline to a held-out TLS dataset where the vision model yields glass masks with large errors and finding no measurable improvement in artifact removal over existing methods would disprove the central claim.

Figures

Figures reproduced from arXiv: 2606.10541 by Bo Zhang, Tie Ji, Wanpeng Shao, Yifei Xue, Yizhen Lao, Zeyi Guo.

Figure 2
Figure 2. Figure 2: Challenges in reflection artifact removal (glass and virtual points are colored in yellow and red, respectively). (a) "Glass void" in TLS measurement. (b) TLS measurements with partial, distorted reflected virtual points. to substantial performance degradation. Therefore, two crit￾ical physical challenges inherent to this paradigm remain inadequately addressed: 1. The "Measurement Void" problem in TLS mea￾… view at source ↗
Figure 1
Figure 1. Figure 1: Reflection artifact in TLS point clouds. (a) The principle of reflection in TLS measurement. The laser beam hits the glass surface, producing a glass point (𝑃glass), a virtual point reflected from a building by the glass (𝑃virtual), and a light point inside the building (𝑃light). (b) The real building scene captured by TLS where the glass planes are shown in yellow and green, and virtual points are shown i… view at source ↗
Figure 3
Figure 3. Figure 3: Glass-induced Reflection Artifact Removal in LiDAR Point Clouds (GRAR). Given input point clouds, we first project LiDAR data onto a spherical panoramic form. The RGB, intensity, and multi-count maps are then fed into a vision foundation model (e.g., NanoBanana) to generate accurate glass masks. These masks are back-projected into the original 3D space for extraction, refinement, and completion. Finally, t… view at source ↗
Figure 4
Figure 4. Figure 4: Spherical projection of TLS point clouds to produce intensity and multi-count map. plane to generate a multi-count map and an intensity map, where each pixel records the echo count and the first-return intensity, respectively. Subsequently, intensity map, multi-count map together the synchro￾nized panoramic RGB image, are jointly fed into a vision foundation model to extract a high-quality 2D semantic glas… view at source ↗
Figure 5
Figure 5. Figure 5: Overview of the proposed glass mask generation strategy. An initial glass mask is inferred from RGB imagery, LiDAR intensity maps, and multi-count maps using a vision foundation model, and is subsequently refined and completed through geometric constraints to obtain high-completeness glass surfaces. that closely resemble real objects, often leading to am￾biguous semantic interpretations (Lin et al., 2021; … view at source ↗
Figure 6
Figure 6. Figure 6: Affected area detection in real scene. Line in blue color is the transmission path from the scan pose to the outline of each glass region. Point fall into the red area are reflection￾affected points. Finally, we perform a geometry-based completion for points that are recognized as belonging to glass regions but are sparsely sampled. Specifically, for each point 𝑝𝑚 that falls within the es￾timated glass mas… view at source ↗
Figure 8
Figure 8. Figure 8: Overview of the proposed RE-LGGS descriptor for imperfect reflection observations. (a) Visualization of reflection observations between direct real-point correspondence and the searched real points. (b) Illustration of the PCA-based geometric descriptor with orientation consistency constraints at a single neighborhood scale. The descriptor is computed in the same manner at multiple scales. substantial loss… view at source ↗
Figure 9
Figure 9. Figure 9: Overview of virtual points removal process. (a) Reflection-affected points. (b) Point-level final scores of affected points where points differ in the same shape. (c) Segmentation results. (d) Segment-level final scores of affected points where points in the same shape show the same score. (e) Reflections removal results. By adjusting the parameters 𝛽1 and 𝛽2 in Eqs. (5) and (14), respectively, the final s… view at source ↗
Figure 10
Figure 10. Figure 10: Comparision of the glass region estimation results on TLS data with multiple small glass planes. (a) The panoramic images of input TLS point clouds: The red rectangles are glass objects with curtains drawn behind; green circles indicate glass component with no virtual points. (b) Intensity map; the color from dark blue to bright yellow represents the intensity values, ranging from low to high. (c) Glass r… view at source ↗
Figure 11
Figure 11. Figure 11: Comparison of the dominant glass region estimation. First row: panoramic images of input point clouds. Second row: multi-return method (Yun and Sim, 2018). Third row: segment-level multi-count method (Shao et al., 2026). Bottom row: our proposed method. Scenes from left to right: (a) “Architecture building”, (b) “International hall”, (c) “Botanical garden”, (d) “Terrace”, (e) “Engineering building”, (f) “… view at source ↗
Figure 12
Figure 12. Figure 12: Comparison of glass detection results in several TLS data scenes with severe glass points missing. Detected glass points are shown in yellow in 3D form. It is noted that points close to the glass are not shown to improve the visualization of glass points. (a) Input glass points. (b) Segment-level multi-count results (Shao et al., 2026). (c) Our proposed method (d) Target glass points. Results are shown fr… view at source ↗
Figure 13
Figure 13. Figure 13: Comparison of the virtual point detection results on UNIST building dataset and a multiple-glass dataset. (a) Input ground truth points. (b) Yun and Sim (2018). (c) GRASS (Shao et al., 2026). (d) Our proposed method. Scenes from top to bottom: “Architecture building”, “Botanical garden”, “Engineering building”, “Natural science building”, “Terrace”, “Office building”. : Preprint submitted to Elsevier Page… view at source ↗
Figure 14
Figure 14. Figure 14: Result of proposed method on two "street view" 3DRN dataset. The red points denote virtual points. (a) Input ground truth points. (b) Our implementation of the method from Fang et al. (2025) (provided for reference). (c) GRASS (Shao et al., 2026). (d) Our proposed method. Scenes from top to bottom: "Scan 04", "Scan 05". (a) (b) [PITH_FULL_IMAGE:figures/full_fig_p016_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Reflection artifact removal in a subway scene (Dong et al., 2020b), where virtual points appear on the tracks. (a) Manually annotated reference data; glass points are shown in yellow and virtual points in red. (b) Virtual point removal results produced by the proposed method. In addition, an adaptive orientation consistency constraint is introduced to distinguish structures with similar ge￾ometric statist… view at source ↗
read the original abstract

Terrestrial Laser Scanning (TLS) point clouds captured in urban environments frequently suffer from glass-induced reflection artifacts, severely degrading downstream applications. Existing reflection artifact removal methods generally rely on ideal reflection symmetry assumptions, yet their performance is limited by inaccurate glass estimation and insufficient geometric representations. To address these issues, we propose a novel unified framework aimed at robust reflection artifact removal: In the first stage, we leverage a multi-modal vision foundation model to produce initial glass masks, which are then refined using geometric cues to achieve high-precision glass regions, followed by glass completion to recover missing regions caused by no-return measurements on transparent surfaces; In the second stage, we propose a physics-driven descriptor, termed Reflection-aware Local-Global Geometric Similarity (RE-LGGS), which is grounded in actual laser reflection geometry and jointly encodes multi-scale geometric structures and orientation consistency using PCA-based local shape representations, thereby significantly improving robustness against imperfect observations. Extensive experiments on multiple public TLS datasets demonstrate that our framework consistently outperforms state-of-the-art methods in reflection artifacts removal.

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 / 1 minor

Summary. The manuscript presents GRAR, a two-stage framework for removing glass-induced reflection artifacts from TLS point clouds. Stage 1 uses a multi-modal vision foundation model to generate initial glass masks, which are refined via geometric cues and completed to handle no-return regions on transparent surfaces. Stage 2 introduces the RE-LGGS descriptor, a physics-driven measure grounded in laser reflection geometry that jointly encodes multi-scale structures and orientation consistency via PCA-based local shape representations. The paper claims that extensive experiments on multiple public TLS datasets demonstrate consistent outperformance over state-of-the-art reflection artifact removal methods.

Significance. If the quantitative claims hold, the work could improve reliability of TLS data in urban scenes for downstream tasks such as 3D reconstruction and semantic segmentation. The explicit grounding of the similarity measure in reflection physics and the two-stage separation of mask generation from geometric removal are potentially useful design choices.

major comments (3)
  1. [Abstract] Abstract: the central claim that the framework 'consistently outperforms state-of-the-art methods' is unsupported by any quantitative metrics, error bars, dataset statistics, or ablation results, rendering the headline result impossible to evaluate from the provided text.
  2. [Abstract] Abstract (first-stage description): the entire pipeline is load-bearing on the assumption that the multi-modal vision foundation model supplies initial glass masks accurate enough for geometric refinement to succeed; no mask-quality metric (IoU, precision-recall on glass regions) or domain-shift analysis is referenced, leaving this prerequisite unanchored.
  3. [Abstract] Abstract (second-stage description): the RE-LGGS descriptor is asserted to be 'grounded in actual laser reflection geometry,' yet the abstract supplies neither the explicit geometric derivation nor any equation showing how the PCA-based local shape representations enforce orientation consistency under imperfect observations.
minor comments (1)
  1. [Abstract] The acronym RE-LGGS is introduced without expansion on first use.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive comments on the abstract. We agree that the abstract can be made more self-contained and will revise it accordingly while preserving its concise nature. We address each major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the framework 'consistently outperforms state-of-the-art methods' is unsupported by any quantitative metrics, error bars, dataset statistics, or ablation results, rendering the headline result impossible to evaluate from the provided text.

    Authors: The abstract serves as a high-level summary; the full manuscript contains the supporting quantitative results, metrics, error bars, dataset statistics, and ablations on multiple public TLS datasets. To strengthen the abstract, we will incorporate key performance highlights and dataset details in the revision. revision: yes

  2. Referee: [Abstract] Abstract (first-stage description): the entire pipeline is load-bearing on the assumption that the multi-modal vision foundation model supplies initial glass masks accurate enough for geometric refinement to succeed; no mask-quality metric (IoU, precision-recall on glass regions) or domain-shift analysis is referenced, leaving this prerequisite unanchored.

    Authors: We agree the abstract does not reference mask-quality metrics. The manuscript evaluates the initial glass mask generation, geometric refinement, and completion stages using metrics such as IoU and precision-recall. We will revise the abstract to reference these metrics and note the evaluation of the first stage. revision: yes

  3. Referee: [Abstract] Abstract (second-stage description): the RE-LGGS descriptor is asserted to be 'grounded in actual laser reflection geometry,' yet the abstract supplies neither the explicit geometric derivation nor any equation showing how the PCA-based local shape representations enforce orientation consistency under imperfect observations.

    Authors: The abstract summarizes the descriptor; the full manuscript provides the physics-based derivation and the equations detailing how the PCA-based local shape representations capture orientation consistency. We will update the abstract to include a concise reference to the geometric grounding and the relevant equation. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The provided abstract and method description introduce a two-stage pipeline that invokes an external multi-modal vision foundation model for initial masks, followed by geometric refinement and a new RE-LGGS descriptor grounded in laser reflection geometry. No equations, fitted parameters, or self-citations are quoted that reduce any claimed prediction or result to a definition or input by construction. The central claim of outperformance rests on experimental results rather than tautological steps, satisfying the criteria for a self-contained derivation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Review based solely on abstract; no explicit free parameters, axioms, or invented entities beyond the named descriptor are described.

invented entities (1)
  • RE-LGGS descriptor no independent evidence
    purpose: Encodes multi-scale geometric structures and orientation consistency grounded in laser reflection geometry
    Introduced as a new physics-driven descriptor in the second stage

pith-pipeline@v0.9.1-grok · 5719 in / 911 out tokens · 20144 ms · 2026-06-27T13:49:14.122708+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

61 extracted references · 7 canonical work pages

  1. [1]

    Ying and P

    S. Ying and P. Van Oosterom and H. Fan , title =. J. Geovis. Spat. Anal. , volume =. 2023 , doi =

    2023

  2. [2]

    Snavely and S

    N. Snavely and S. M. Seitz and R. Szeliski , title =. ACM Trans. Graph. , volume =. 2006 , doi =

    2006

  3. [3]

    Furukawa and J

    Y. Furukawa and J. Ponce , title =. IEEE Trans. Pattern Anal. Mach. Intell. , volume =

  4. [4]

    Kerbl and G

    B. Kerbl and G. Kopanas and T. Leimk. 3D Gaussian Splatting for Real-Time Radiance Field Rendering , journal =. 2023 , doi =

    2023

  5. [5]

    T. S. Fong and W. Y. Yan , title =. Autom. Constr. , volume =. 2025 , doi =

    2025

  6. [6]

    ISPRS Journal of Photogrammetry and Remote Sensing , volume=

    Registration of large-scale terrestrial laser scanner point clouds: A review and benchmark , author=. ISPRS Journal of Photogrammetry and Remote Sensing , volume=. 2020 , publisher=

    2020

  7. [7]

    Xiong and Y

    B. Xiong and Y. Jin and F. Li and Y. Chen and Y. Zou and Z. Zhou , title =. Autom. Constr. , volume =. 2023 , doi =

    2023

  8. [8]

    Robust Multiview Point Cloud Registration Using Algebraic Connectivity and Spatial Compatibility , year=

    Fang, Li and Li, Tianyu and Zhou, Shudong and Lin, Yanghong , journal=. Robust Multiview Point Cloud Registration Using Algebraic Connectivity and Spatial Compatibility , year=

  9. [9]

    Ambrosino and A

    A. Ambrosino and A. Di Benedetto and M. Fiani , title =. Remote Sens. , volume =. 2024 , doi =

    2024

  10. [10]

    Lao , title =

    Y. Lao , title =. 2019 , address =

    2019

  11. [11]

    Srinivasan, Matthew Tancik, Jonathan T

    Mildenhall, Ben and Srinivasan, Pratul P. and Tancik, Matthew and Barron, Jonathan T. and Ramamoorthi, Ravi and Ng, Ren , title =. 2021 , issue_date =. doi:10.1145/3503250 , journal =

    work page doi:10.1145/3503250 2021

  12. [12]

    2024 , issn =

    WHU-Urban3D: An urban scene LiDAR point cloud dataset for semantic instance segmentation , journal =. 2024 , issn =. doi:https://doi.org/10.1016/j.isprsjprs.2024.02.007 , author =

    work page doi:10.1016/j.isprsjprs.2024.02.007 2024

  13. [13]

    2026 , title =

    Geist, Louis and Landrieu, Loic and Robert, Damien , journal =. 2026 , title =

    2026

  14. [14]

    Proceedings of the 35th International Conference on Neural Information Processing Systems , articleno =

    Wang, Peng and Liu, Lingjie and Liu, Yuan and Theobalt, Christian and Komura, Taku and Wang, Wenping , title =. Proceedings of the 35th International Conference on Neural Information Processing Systems , articleno =. 2021 , isbn =

    2021

  15. [15]

    and Frahm, Jan-Michael , booktitle=

    Schönberger, Johannes L. and Frahm, Jan-Michael , booktitle=. Structure-from-Motion Revisited , year=

  16. [16]

    Gonizzi Barsanti and M

    S. Gonizzi Barsanti and M. R. Marini and S. G. Malatesta and A. Rossi , title =. Remote Sens. , volume =. 2024 , doi =

    2024

  17. [17]

    Zheng and B

    Z. Zheng and B. Zha and Y. Zhou and J. Huang and Y. Xuchen and H. Zhang , title =. Remote Sens. , volume =. 2022 , doi =

    2022

  18. [18]

    Wang and Y

    L. Wang and Y. Chen and H. Xu , title =. Remote Sens. , volume =. 2024 , doi =

    2024

  19. [19]

    J. -S. Yun and J. -Y. Sim , title =. Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , pages =. 2018 , address =

    2018

  20. [20]

    J. -S. Yun and J. -Y. Sim , title =. IEEE Trans. Pattern Anal. Mach. Intell. , volume =. 2021 , doi =

    2021

  21. [21]

    J. -S. Yun and J. -Y. Sim , title =. Proc. IEEE Int. Conf. Image Process. (ICIP) , pages =. 2019 , address =

    2019

  22. [22]

    Shao and K

    W. Shao and K. Kakizaki and S. Araki and T. Mukai , title =. Proc. IEEE Annu. Comput. Softw. Appl. Conf. (COMPSAC) , pages =. 2023 , address =

    2023

  23. [23]

    Lee and K

    O. Lee and K. Joo and J. -Y. Sim , title =. IEEE Robot. Autom. Lett. , volume =. 2023 , doi =

    2023

  24. [24]

    Fang and T

    L. Fang and T. Li and Y. Lin and S. Zhou and W. Yao , title =. ISPRS J. Photogramm. Remote Sens. , volume =. 2025 , doi =

    2025

  25. [25]

    Shao and Y

    W. Shao and Y. Zhang and Y. Xue and T. Ji and Y. Lao , title =. Remote Sens. , volume =. 2026 , doi =

    2026

  26. [26]

    2025 , note =

    Nano Banana (Gemini 2.5 Flash Image) , howpublished =. 2025 , note =

    2025

  27. [27]

    R. B. Rusu and N. Blodow and M. Beetz , title =. Proc. IEEE Int. Conf. Robot. Autom. (ICRA) , pages =. 2009 , address =

    2009

  28. [28]

    Advances in Civil Engineering , volume =

    Hosamo, Haidar Hosamo and Hosamo, Mohsen Hosamo , title =. Advances in Civil Engineering , volume =. doi:https://doi.org/10.1155/2022/2194949 , year =

    work page doi:10.1155/2022/2194949 2022

  29. [29]

    Liu and P

    C. Liu and P. Zhang and X. Xu , title =. J. Infrastruct. Intell. Resil. , volume =. 2023 , doi =

    2023

  30. [30]

    Gao and J

    R. Gao and J. Park and X. Hu and S. Yang and K. Cho , title =. Remote Sens. , volume =. 2021 , doi =

    2021

  31. [31]

    Gao and M

    R. Gao and M. Li and S. -J. Yang and K. Cho , title =. Remote Sens. , volume =. 2022 , doi =

    2022

  32. [32]

    Koch and S

    R. Koch and S. May and P. Koch and M. K. Detection of Specular Reflections in Range Measurements for Faultless Robotic SLAM , booktitle =. 2016 , address =

    2016

  33. [33]

    Koch and S

    R. Koch and S. May and P. Murmann and A. N. Identification of Transparent and Specular Reflective Material in Laser Scans to Discriminate Affected Measurements for Faultless Robotic SLAM , journal =. 2017 , doi =

    2017

  34. [34]

    Koch and S

    R. Koch and S. May and A. N. Detection and Purging of Specular Reflective and Transparent Object Influences in 3D Range Measurements , booktitle =. 2017 , address =

    2017

  35. [35]

    Zhao and Z

    X. Zhao and Z. Yang and S. Schwertfeger , title =. Proc. IEEE Int. Symp. Safety, Secur. Rescue Robot. (SSRR) , pages =. 2020 , address =

    2020

  36. [36]

    Li and X

    Y. Li and X. Zhao and S. Schwertfeger , title =. Sensors , volume =. 2024 , doi =

    2024

  37. [37]

    2015 , issn =

    A Review of LIDAR Radiometric Processing: From Ad Hoc Intensity Correction to Rigorous Radiometric Calibration , journal =. 2015 , issn =. doi:https://doi.org/10.3390/s151128099 , author =

    work page doi:10.3390/s151128099 2015

  38. [38]

    Vosselman , title =

    G. Vosselman , title =. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. , year =

  39. [39]

    Vosselman and B

    G. Vosselman and B. G. H. Gorte and G. Sithole and T. Rabbani , title =. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. , pages =. 2004 , volume =

    2004

  40. [40]

    Mei and X

    H. Mei and X. Yang and Y. Wang and Y. Liu and S. He and Q. Zhang and X. Wei and R. W. Lau , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , address =

  41. [41]

    He and X

    H. He and X. Li and G. Cheng and J. Shi and Y. Tong and G. Meng and V. Prinet and L. Weng , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages =

  42. [42]

    Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence,

    Ke Fan and Changan Wang and Yabiao Wang and Chengjie Wang and Ran Yi and Lizhuang Ma , title =. Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence,. 2023 , doi =

    2023

  43. [43]

    Qi and X

    F. Qi and X. Tan and Z. Zhang and M. Chen and Y. Xie and L. Ma , title =. IEEE Transactions on Industrial Informatics , volume =

  44. [44]

    Lin and Z

    J. Lin and Z. He and R. W. Lau , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

  45. [45]

    Liu and Y

    F. Liu and Y. Liu and J. Lin and K. Xu and R. W. Lau , title =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =

  46. [46]

    Lin and Y.-H

    J. Lin and Y.-H. Yeung and R. Lau , title =. Advances in Neural Information Processing Systems , volume =

  47. [47]

    and Shi, Boxin , journal=

    Hong, Yuchen and Zheng, Qian and Zhao, Lingran and Jiang, Xudong and Kot, Alex C. and Shi, Boxin , journal=. PAR2Net: End-to-End Panoramic Image Reflection Removal , year=

  48. [48]

    MLLM - Tool : A Multimodal Large Language Model for Tool Agent Learning

    Tan, Tianlong and Chen, Bin and Cao, Hongliang and Yan, Chenggang and Ma, Yike and Dai, Feng , booktitle =. 2025 , volume =. doi:10.1109/WACV61041.2025.00852 , url =

    work page doi:10.1109/wacv61041.2025.00852 2025

  49. [49]

    Zhang, Jiaming and Yang, Kailun and Shi, Hao and Reiß, Simon and Peng, Kunyu and Ma, Chaoxiang and Fu, Haodong and Torr, Philip H. S. and Wang, Kaiwei and Stiefelhagen, Rainer , journal=. Behind Every Domain There is a Shift: Adapting Distortion-Aware Vision Transformers for Panoramic Semantic Segmentation , year=

  50. [50]

    Pairwise coarse registration of point clouds in urban scenes using voxel-based 4-planes congruent sets , year=

    Xu, Yusheng and Boerner, Richard and Yao, Wei and Hoegner, Ludwig and Stilla, Uwe , journal=. Pairwise coarse registration of point clouds in urban scenes using voxel-based 4-planes congruent sets , year=

  51. [51]

    Landrieu and G

    L. Landrieu and G. Obozinski , title =. SIAM J. Imaging Sci. , volume =. 2017 , month =

    2017

  52. [52]

    2014 , note =

    RIEGL VZ-400 3D Terrestrial Laser Scanner , howpublished =. 2014 , note =

    2014

  53. [53]

    2024 , note =

    RIEGL VZ-2000i Long Range 3D Laser Scanning System , howpublished =. 2024 , note =

    2024

  54. [54]

    , title =

    Householder, Alston S. , title =. 1958 , issue_date =. doi:10.1145/320941.320947 , journal =

    work page doi:10.1145/320941.320947 1958

  55. [55]

    Thomas and C

    H. Thomas and C. R. Qi and J. -E. Deschaud and B. Marcotegui and F. Goulette and L. Guibas , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages =. 2019 , address =

    2019

  56. [56]

    2020 , issn =

    Registration of large-scale terrestrial laser scanner point clouds: A review and benchmark , journal =. 2020 , issn =. doi:https://doi.org/10.1016/j.isprsjprs.2020.03.013 , url =

    work page doi:10.1016/j.isprsjprs.2020.03.013 2020

  57. [57]

    2025 , eprint=

    Is Nano Banana Pro a Low-Level Vision All-Rounder? A Comprehensive Evaluation on 14 Tasks and 40 Datasets , author=. 2025 , eprint=

    2025

  58. [58]

    High-Resolution Image Synthesis with Latent Diffusion Models , year=

    Rombach, Robin and Blattmann, Andreas and Lorenz, Dominik and Esser, Patrick and Ommer, Björn , booktitle=. High-Resolution Image Synthesis with Latent Diffusion Models , year=

  59. [59]

    2025 , eprint=

    Pico-Banana-400K: A Large-Scale Dataset for Text-Guided Image Editing , author=. 2025 , eprint=

    2025

  60. [60]

    How Multimodal

    Zhuoran Yu and Yong Jae Lee , booktitle=. How Multimodal. 2025 , url=

    2025

  61. [61]

    and Landrieu, L

    Guinard, S. and Landrieu, L. , TITLE =. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences , VOLUME =. 2017 , PAGES =

    2017