pith. sign in

arxiv: 2606.09123 · v1 · pith:NALVVAYInew · submitted 2026-06-08 · 💻 cs.CV · cs.AI

An Enhanced Geometric-Spectral Feature Learning Framework for Airborne Multispectral Point Cloud Classification

Xian Li , Yanfeng Gu , Aleksandra Pi\v{z}urica This is my paper

Pith reviewed 2026-06-27 17:18 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords multispectral point cloudfeature fusionattention mechanismpoint cloud classificationgeometric-spectral featuresjoint lossairborne remote sensing
0
0 comments X

The pith

A two-stream attention fusion model with joint loss improves classification of airborne multispectral point clouds.

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

The paper seeks to address the challenges of high-dimensional heterogeneous data, class imbalance, and spectral similarity in multispectral point clouds for land-cover classification. It introduces a framework with parallel streams that separately capture global spectral features via self-attention and geometric features via multikernel convolution plus aggregation attention, then merges the most useful parts through a residual attention block. A joint loss term is added to boost learning on difficult samples. If the approach works as described, it would yield higher accuracy on real airborne datasets than prior methods by more effectively combining 3D position and spectral information.

Core claim

The model extracts position-encoded global spectral features in one stream using fusion self-attention and spectral-guided geometric features in the second stream using multikernel point convolution and feature aggregation attention; a residual attention fusion block then integrates the informative parts from both streams, while a joint loss function strengthens performance on unbalanced and inter-class similar samples.

What carries the argument

Two-stream attention-based geometric-spectral feature fusion via residual attention block, paired with a joint loss function.

If this is right

  • Higher classification accuracy on airborne MPC data with inter-class spectral overlap.
  • More effective use of both geometric structure and spectral signatures in a single model.
  • Improved handling of unbalanced class distributions common in remote-sensing point clouds.

Where Pith is reading between the lines

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

  • The same two-stream fusion pattern could be tested on other multimodal 3D datasets that combine geometry with per-point attributes.
  • The joint loss design might transfer to other remote-sensing tasks that suffer from class imbalance and feature similarity.
  • Public release of the two datasets would allow direct comparison of future fusion methods against this baseline.

Load-bearing premise

Performance gains on the two datasets come from the attention fusion and joint loss rather than from dataset construction choices, preprocessing, or unstated tuning.

What would settle it

Training the same network architecture on the two datasets but removing the attention mechanisms and joint loss term, then checking whether accuracy falls to levels comparable with earlier methods.

Figures

Figures reproduced from arXiv: 2606.09123 by Aleksandra Pi\v{z}urica, Xian Li, Yanfeng Gu.

Figure 5
Figure 5. Figure 5: Classification maps on the SKM dataset obtained by (a) Ground truth, (b) RandLA-Net, (c) CGANet, (d) RFFS-Net, (e) EyeNet, (f) MAMPC, (g) GeoSegNet, (h) EMPC, (i) Ours [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
read the original abstract

Multispectral point cloud (MPC) is composed of 3D spatial-spectral information, which holds tremendous potential for accurate land-cover classification. However, the representation power of classification models is limited by inherent high-dimensional and heterogeneous spatial-spectral information, unbalanced sample distribution, and inter-class spectral similarity of airborne MPCs. We build two MPC datasets and propose an enhanced geometric-spectral feature learning framework based on attentions for airborne MPC classification. A key component in our model is a two-stream feature fusion method with attention mechanisms, which enhances the representation capability of spatial-spectral features from high-dimensional heterogeneous MPCs. The first stream aims to extract position-encoded global spectral features with fusion self-attention, and the second stream comprises a multikernel point convolution and feature aggregation attention to extract spectral-guided geometric features. We then develop a residual attention fusion block to integrate the most informative geometric-spectral features from the two parallel streams. Another important contribution of this work is a joint loss function to improve the learning ability on unbalanced and interclass similar samples. Experimental results on two airborne MPC datasets demonstrate the effectiveness of the proposed method compared with the state-of-the-art methods. Furthermore, the codes and datasets used in this paper will be made available freely at https://github.com/HITlixian/TGRS_GSFF.

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

2 major / 1 minor

Summary. The manuscript introduces two new airborne multispectral point cloud (MPC) datasets and proposes an enhanced geometric-spectral feature learning framework. Key components include a two-stream attention fusion method (position-encoded spectral self-attention in one stream and multikernel point convolution with feature aggregation attention in the other), a residual attention fusion block to integrate features, and a joint loss function to handle unbalanced samples and inter-class spectral similarity. The central claim is that this framework yields superior land-cover classification performance compared to state-of-the-art methods on the two introduced datasets, with code and data to be released.

Significance. If the performance improvements are robustly attributable to the proposed attention mechanisms and joint loss, the work could advance handling of high-dimensional heterogeneous spatial-spectral data in remote sensing. The commitment to release datasets and code is a clear strength that supports reproducibility and further research in the field.

major comments (2)
  1. [Experimental results] Experimental results section: The claim that the two-stream attention fusion and joint loss produce the reported gains is load-bearing but unsecured, as the manuscript simultaneously introduces both the method and the new datasets without ablation studies isolating the contribution of each proposed component (attention fusion, residual block, joint loss) from dataset-specific tuning, preprocessing, or implementation choices.
  2. [Abstract] Abstract and results: No quantitative metrics, error bars, ablation tables, or statistical tests are referenced to support the superiority claim over SOTA methods, leaving the magnitude and reliability of gains unverified in the provided summary.
minor comments (1)
  1. Ensure that upon publication the promised code and dataset release at the GitHub link is completed and includes the exact preprocessing and training configurations used for the reported results.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help strengthen the validation of our claims. We address each major point below and commit to revisions that provide the requested isolation of contributions and quantitative support.

read point-by-point responses

  1. Referee: [Experimental results] Experimental results section: The claim that the two-stream attention fusion and joint loss produce the reported gains is load-bearing but unsecured, as the manuscript simultaneously introduces both the method and the new datasets without ablation studies isolating the contribution of each proposed component (attention fusion, residual block, joint loss) from dataset-specific tuning, preprocessing, or implementation choices.

    Authors: We agree that the absence of component-wise ablations leaves the attribution of gains partially unsecured, particularly given the introduction of new datasets. In the revised manuscript we will add a dedicated ablation section that systematically removes or replaces each element (position-encoded spectral self-attention, multikernel convolution with feature aggregation attention, residual attention fusion block, and joint loss) while keeping all other implementation and preprocessing choices fixed. Results will be reported on both datasets with the same training protocol, thereby isolating the contribution of the proposed modules from dataset-specific factors. revision: yes

  2. Referee: [Abstract] Abstract and results: No quantitative metrics, error bars, ablation tables, or statistical tests are referenced to support the superiority claim over SOTA methods, leaving the magnitude and reliability of gains unverified in the provided summary.

    Authors: The current abstract states only qualitative superiority. We will revise the abstract to report concrete metrics (overall accuracy, mean F1-score, and per-class improvements) together with the magnitude of gains over the strongest baseline. The results section will be expanded to include error bars from repeated runs with different random seeds, the requested ablation tables, and paired statistical tests (e.g., McNemar or Wilcoxon) to establish significance of the observed differences. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical claims with no derivations or self-referential fitting

full rationale

The paper introduces a neural architecture (two-stream attention fusion, residual block, joint loss) and two new datasets, then reports classification accuracies against SOTA baselines. No equations, first-principles derivations, fitted parameters renamed as predictions, or load-bearing self-citations appear in the provided text. All claims rest on external experimental comparison rather than any reduction of outputs to inputs by construction. This matches the default expectation of a non-circular empirical ML paper.

Axiom & Free-Parameter Ledger

3 free parameters · 2 axioms · 0 invented entities

Only the abstract is available, so the ledger is necessarily incomplete; typical deep-learning papers of this type contain dozens of fitted hyperparameters whose values are not visible here.

free parameters (3)
  • attention head count and dimension
    Standard in transformer-style blocks; values chosen to fit the spectral feature size.
  • joint loss weighting coefficients
    Weights balancing classification, geometric, and spectral terms are fitted or hand-tuned on the training split.
  • multikernel sizes in point convolution
    Kernel radii and number of kernels are design choices fitted to the point density of the airborne data.
axioms (2)
  • domain assumption Training and test splits are i.i.d. samples from the same underlying distribution
    Implicit in all supervised learning claims on the two datasets.
  • ad hoc to paper Attention mechanisms can meaningfully separate inter-class spectral similarity without additional regularization
    Core modeling assumption required for the two-stream fusion to succeed.

pith-pipeline@v0.9.1-grok · 5771 in / 1435 out tokens · 16597 ms · 2026-06-27T17:18:17.481747+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

44 extracted references · 12 canonical work pages

  1. [1]

    An adaptive 3D reconstruction method for asymmetric dual -angle multispectral stereo imaging syst em on UAV platform,

    C. Wang, X. Li, and Y. Gu, “An adaptive 3D reconstruction method for asymmetric dual -angle multispectral stereo imaging syst em on UAV platform,” Sci. China Inf. Sci., vol. 67, no. 8, Art. no. 182305, 2024, doi: 10.1007/s11432-024-4056-8

    work page doi:10.1007/s11432-024-4056-8 2024

  2. [2]

    A Robust Multispectral Point Cloud Generation Method Based on 3 -D Reconstruction from Multispectral Images,

    C. Wang, Y. Gu, and X. Li, “A Robust Multispectral Point Cloud Generation Method Based on 3 -D Reconstruction from Multispectral Images,” IEEE Trans. Geosci. Remote Sens. , vol. 61, pp. 1 -12, Art. no. 5407612, 2023, doi: 10.1109/TGRS.2023.3326153

    work page doi:10.1109/tgrs.2023.3326153 2023

  3. [3]

    Multimodal fusion of UAV -based computer vision and plant water content dynamics for high -throughput soybean maturity classification,

    Y. Li, T. Li, Y. Zhao, and X. Zhang, “Multimodal fusion of UAV -based computer vision and plant water content dynamics for high -throughput soybean maturity classification,” Crop and Environment, 2025

    2025

  4. [4]

    Advancing soybean biomass estimation through multi-source UAV data fusion and machine learning algorithms,

    H. Da, Y. Li, L. Xu, and X. Wang, “Advancing soybean biomass estimation through multi-source UAV data fusion and machine learning algorithms,” Smart Agric. Technol., vol. 10, p. 100778, 2025

    2025

  5. [5]

    A high- resolution canopy height model of the earth,

    N. Lang, W. Jetz , K. Schindler, and M. Wegner, “A high -resolution canopy height model of the Earth,” Nat Ecol Evol, vol. 7, pp. 1778-1789, 2023, doi: 10.1038/s41559-023-02206-6

    work page doi:10.1038/s41559-023-02206-6 2023

  6. [6]

    Feasibility of underwater true color three -dimensional i maging using hyperspectral LiDAR,

    Y. Wang, H. Pan, S. Qiu, and X. Liu, “Feasibility of underwater true color three -dimensional i maging using hyperspectral LiDAR,” Opt. Lasers Eng, vol. 194, p. 109158, 2025

    2025

  7. [7]

    A Normalized Spatial– Spectral Supervoxel Segmentation Method for Multispectral Point Cloud Data,

    L. Chen, L. Gu, Y. Gu, X. Li, X. Zhang, and B. Liu, “A Normalized Spatial-Spectral Supervoxel Segmentation Method for Multispectral Point Cloud Data,” IEEE Trans. Geosci. Remote Sens., vol. 61, pp. 1-11, Art. no. 5704311, 2023, doi: 10.1109/TGRS.2023.3313734

    work page doi:10.1109/tgrs.2023.3313734 2023

  8. [8]

    Masking Graph Cross - Convolution Network for Multispectral Point Cloud Classification,

    Q. Wang, Y. Gu, X. Li, and L. Chen, “Masking Graph Cross - Convolution Network for Multispectral Point Cloud Classification,” IEEE Trans. Geosci. Remote Sens. , vol. 63, pp. 1 -15, Art. no. 5701815, 2025, doi: 10.1109/TGRS.2025.3545783

    work page doi:10.1109/tgrs.2025.3545783 2025

  9. [9]

    A feature selection method for multimodal multispectral LiDAR sensing,

    Y. Han, D. Salido -Monzu, J. A. Butt, and K. Schindler, “A feature selection method for multimodal multispectral LiDAR sensing,” ISPRS J. Photogramm. Remote Sens., vol. 212, pp. 42-57, 2024

    2024

  10. [10]

    Supercontinuum-based hyperspectral LiDAR for precision laser scanning,

    P. Ray , D. Salido -Monzú , S. L. Camenzind, and M. Wegner, “Supercontinuum-based hyperspectral LiDAR for precision laser scanning,” Opt. Express, vol. 31, no. 20, pp. 33486-33499, 2023

    2023

  11. [11]

    An Intensity -Independent Stereo Registration Method of Push -Broom Hyperspectral Scanner and LiDAR on UAV Platforms,

    Y. Gu, C. Wang, and X. Li, “An Intensity -Independent Stereo Registration Method of Push-Broom Hyperspectral Scanner and LiDAR on UAV Platforms,” IEEE Trans. Geosci. Remote Sens. , vol. 60, pp. 1 - 14, Art. no. 5540014, 2022, doi: 10.1109/TGRS.2022.3211202

    work page doi:10.1109/tgrs.2022.3211202 2022

  12. [12]

    LPRNet: A Self -Supervised Registration Network for L iDAR and Photogrammetric Point Clouds,

    C. Wang, Y. Gu, and X. Li, “LPRNet: A Self -Supervised Registration Network for L iDAR and Photogrammetric Point Clouds,” IEEE Trans. Geosci. Remote Sens. , vol. 63, pp. 1 -12, Art. no. 4404012, 2025, doi: 10.1109/TGRS.2025.3541639

    work page doi:10.1109/tgrs.2025.3541639 2025

  13. [13]

    A maximum entropy -based optimal neighbor selection for multispectral airborne LiDAR point cloud classification,

    G. Jiang, W. Y. Yan, and D. D. Lichti, “A maximum entropy -based optimal neighbor selection for multispectral airborne LiDAR point cloud classification,” IEEE Trans. Geosci. Remote Sens. , vol. 61, pp. 1 -18, 2023

    2023

  14. [14]

    Unsupervised occluded target detection based on spherical shell with multispectral point clouds,

    L. Chen, Y. Gu, and X. Li, “Unsupervised Occluded Target Detection Based on Spherical Shell with Multispectral Point Clouds,” IEEE Trans. Geosci. Remote Sens. , vol. 63, pp. 1 -13, Art. no. 4413413, 2025, doi: 10.1109/TGRS.2025.3585524

    work page doi:10.1109/tgrs.2025.3585524 2025

  15. [15]

    Pillar-Voxel Fusion Network for 3D Object Detection in Airborne Hyperspectral Point Clouds,

    Y. Jiang, Y. Gu, and X. Li, “Pillar-Voxel Fusion Network for 3D Object Detection in Airborne Hyperspectral Point Clouds,” Sci. China Inf. Sci. , vol. 69, no. 1, Art. no. 112301, 2026, doi: 10.1007/s11432-024-4458-0

    work page doi:10.1007/s11432-024-4458-0 2026

  16. [16]

    Development of a multispectral fluorescence LiDAR for point cloud segmentation of plants,

    K. Zheng, H. Lin, X. Hong, and Y. Wang, “Development of a multispectral fluorescence LiDAR for point cloud segmentation of plants,” Opt. Express, vol. 31, no. 11, pp. 18613-18629, 2023

    2023

  17. [17]

    CapViT: Cross-Context Capsule Vision Transformers for Land Cover Classification with Airborne Multispectral LiDAR Data,

    Y. Yu, et al, “CapViT: Cross-Context Capsule Vision Transformers for Land Cover Classification with Airborne Multispectral LiDAR Data,” Int. J. Appl. Earth Obs. Geoinf., vol. 111, p. 102837, 2022

    2022

  18. [18]

    Spatial -Spectral Feature Fusion and Spectral Recon struction of Multispectral LiDAR Point Clouds by Attention Mechanism,

    G. Zhou, H. Qi, S. Shi, and Y. Gu, “Spatial -Spectral Feature Fusion and Spectral Recon struction of Multispectral LiDAR Point Clouds by Attention Mechanism,” Remote Sens., vol. 17, no. 14, p. 2411, 2025

    2025

  19. [19]

    Land-Cover Classification of Multispectral LiDAR Data Using CNN with Optimized Hyper-Parameters,

    S. Pan, et al, “Land-Cover Classification of Multispectral LiDAR Data Using CNN with Optimized Hyper-Parameters,” ISPRS J. Photogramm. Remote Sens., vol. 166, pp. 241-254, 2020

    2020

  20. [20]

    3D -UMamba: 3D U -Net with State Space Model for Semantic Segmentation of Multi -Source LiDAR Point Clouds,

    D. Lu, L. Xu, J. Zhou, and Y. Li, “3D -UMamba: 3D U -Net with State Space Model for Semantic Segmentation of Multi -Source LiDAR Point Clouds,” Int. J. Appl. Earth Obs. Geoinf., vol. 136, p. 104401, 2025

    2025

  21. [21]

    An Enhanced Classification Method Based on Adaptive Multi-Scale Fusion for Long -Tailed Multispectral Point Clouds,

    T. Liu, et al, “An Enhanced Classification Method Based on Adaptive Multi-Scale Fusion for Long -Tailed Multispectral Point Clouds,” Sci. China Inf. Sci. , vol. 68, no. 8, Art. no. 182302, 2025, doi: 10.1007/s11432-024-4324-6

    work page doi:10.1007/s11432-024-4324-6 2025

  22. [22]

    Multispectral Point Cloud Classification Network Based on Multilateral Attention,

    B. Hu, X. Li, and T. Liu, “Multispectral Point Cloud Classification Network Based on Multilateral Attention,” in Proc. 13th WHISPERS, 2023, pp. 1-5

    2023

  23. [23]

    SCSQ -Net: A Shared Kernel Point Convolution Semantic Query Network for Weakly Supervised Classification of Multispec tral LiDAR Point Clouds,

    K. Chen, H. Guan, Y. Yu, and X. Li, “SCSQ -Net: A Shared Kernel Point Convolution Semantic Query Network for Weakly Supervised Classification of Multispec tral LiDAR Point Clouds,” IEEE Trans. Geosci. Remote Sens., 2024

    2024

  24. [24]

    Spherical Frustum Sparse Convolution Network for LiDAR Point Cloud Semantic Segmentation,

    Y. Zheng, G. Wang, J. Liu, and H. Zhang, “Spherical Frustum Sparse Convolution Network for LiDAR Point Cloud Semantic Segmentation,” Adv. Neural Inf. Process. Syst., vol. 37, pp. 121827-121858, 2024

    2024

  25. [25]

    PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation,

    C. R. Qi, H. Su, K. Mo, and L. J. Guibas, “PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2017, pp. 652-660

    2017

  26. [26]

    PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space,

    C. R. Qi, L. Yi, H. Su, and L. J. Guibas, “PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space,” Adv. Neural Inf. Process. Syst., vol. 30, 2017

    2017

  27. [27]

    Beyond Single Receptive Field: A Receptive Field Fusion-and-Stratification Network for Airborne Laser Scanning Point Cloud Classification,

    Y. Mao, K. Chen, W. Diao, and X. Li, “Beyond Single Receptive Field: A Receptive Field Fusion-and-Stratification Network for Airborne Laser Scanning Point Cloud Classification,” ISPRS J. Photogramm. Remote Sens., vol. 188, pp. 45-61, 2022

    2022

  28. [28]

    Point Transformer,

    H. Zhao, L. Jiang, J. Jia, and S. Wu, “Point Transformer,” in Proc. IEEE/CVF Int. Conf. Comput. Vis., 2021, pp. 16259-16268

    2021

  29. [29]

    Point Transformer V2: Grouped Vector Attention and Partition-Based Pooling,

    X. Wu, Y. Lao, L. Jiang, and P. S. Wang, “Point Transformer V2: Grouped Vector Attention and Partition-Based Pooling,” Adv. Neural Inf. Process. Syst., vol. 35, pp. 33330-33342, 2022

    2022

  30. [30]

    Point Transformer V3: Simpler Faster Stronger,

    X. Wu, L. Jiang, P. S. Wang, and S. Wu, “Point Transformer V3: Simpler Faster Stronger,” in Proc. IE EE/CVF Conf. Comput. Vis. Pattern Recognit., 2024, pp. 4840-4851

    2024

  31. [31]

    PointConv: Deep Convolutional Networks on 3D Point Clouds,

    W. Wu, Z. Qi, and L. Fuxin, “PointConv: Deep Convolutional Networks on 3D Point Clouds,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., 2019, pp. 9621-9630

    2019

  32. [32]

    RandLA-Net: Efficient Semantic Segmentation of Large -Scale Point Clouds,

    Q. Hu, B. Yang, L. Xie, and S. Rosa, “RandLA-Net: Efficient Semantic Segmentation of Large -Scale Point Clouds,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., 2020, pp. 11108-11117

    2020

  33. [33]

    PointNeXt: Revisiting PointNet++ with Improved Train ing and Scaling Strategies,

    G. Qian, Y. Li, H. Peng, and X. Li, “PointNeXt: Revisiting PointNet++ with Improved Train ing and Scaling Strategies,” Adv. Neural Inf. Process. Syst., vol. 35, pp. 23192-23204, 2022

    2022

  34. [34]

    Unsupervised Domain Adaptation for Cross -Scene Multispectral Point Cloud Classification,

    Q. Wang, M. Wang, J. Huang, and X. Li, “Unsupervised Domain Adaptation for Cross -Scene Multispectral Point Cloud Classification,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-15, 2024

    2024

  35. [35]

    Point Tree Transformer for Point Cloud Registration,

    M. Wang, et al, “Point Tree Transformer for Point Cloud Registration,” IEEE Trans. Circuits Syst. Video Technol., 2025

    2025

  36. [36]

    Point Cloud Mamba: Point Cloud Learning via State Space Model,

    T. Zhang, H. Yuan, L. Qi, and Y. Gu, “Point Cloud Mamba: Point Cloud Learning via State Space Model,” in Proc. AAAI Conf. Artif. Intell., vol. 39, no. 10, pp. 10121-10130, 2025

    2025

  37. [37]

    Quantity-Quality Enhanced Self-Training Network for Weakly Supervised Point Cloud Semantic Segmentation,

    J. Deng, J. Lu, and T. Zhang, “Quantity-Quality Enhanced Self-Training Network for Weakly Supervised Point Cloud Semantic Segmentation,” IEEE Trans. Pattern Anal. Mach. Intell., 2025

    2025

  38. [38]

    Multilateral Cascading Network for Semantic Segmentation of Large -Scale Outdoor Point Clouds,

    H. Gong, H. Wang, and D. Wang, “Multilateral Cascading Network for Semantic Segmentation of Large -Scale Outdoor Point Clouds,” IEEE Geosci. Remote Sens. Lett., vol. 22, pp. 1-5, Art. no. 6501005, 2025, doi: 10.1109/LGRS.2025.3547913

    work page doi:10.1109/lgrs.2025.3547913 2025

  39. [39]

    GeoSegNet: Point Cloud Semantic Segmentation via Geometric Encoder -Decoder Modeling,

    C. Chen, Y. Wang, H. Chen , and X. Li, “GeoSegNet: Point Cloud Semantic Segmentation via Geometric Encoder -Decoder Modeling,” The Visual Comput., vol. 40, no. 8, pp. 5107-5121, 2024

    2024

  40. [40]

    Understanding Center Loss -Based Network for Image Retrieval with Few Tr aining Data,

    P. Ghosh, and L. S. Davis, “Understanding Center Loss -Based Network for Image Retrieval with Few Tr aining Data,” in Proc. Eur. Conf. Comput. Vis. (ECCV) Workshops., 2018, pp. 0-0

    2018

  41. [41]

    An End -to-End Framework for Joint Denoising and Classification of Hyperspectral Images,

    X. Li, M. Ding, Y. Gu, and A. Pižurica, “An End -to-End Framework for Joint Denoising and Classification of Hyperspectral Images,” IEEE Trans. Neural Netw. Learn. Syst. , vol. 3 4, no. 7, pp. 3269 -3283, Jul. 2023, doi: 10.1109/TNNLS.2023.3264587

    work page doi:10.1109/tnnls.2023.3264587 2023

  42. [42]

    Semantic Segmentation for Real Point Cloud Scenes via Bilateral Augmentation and Adaptive Fusion,

    S. Qiu, at al, “Semantic Segmentation for Real Point Cloud Scenes via Bilateral Augmentation and Adaptive Fusion,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., 2021, pp. 1757-1767

    2021

  43. [43]

    CGA -Net: Category Guided Aggregation for Point Cloud Semantic Segmentation,

    T. Lu, L. Wang, and G. Wu, “CGA -Net: Category Guided Aggregation for Point Cloud Semantic Segmentation,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., 2021, pp. 11693-11702

    2021

  44. [44]

    Human Vision -Based 3D Point Cloud Semantic Segmentation of Large-Scale Outdoor Scenes,

    S. Yoo, et al , “Human Vision -Based 3D Point Cloud Semantic Segmentation of Large-Scale Outdoor Scenes,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., 2023, pp. 6577-6586. 11 > REPLACE THIS LINE WITH YOUR MANUSCRIPT ID NUMBER (DOUBLE -CLICK HERE TO EDIT) < Xian Li (Member, IEEE) received the Ph.D. degree in instrument science and technology fro...

    2023