Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Acta Optica Sinica, Volume. 45, Issue 8, 0815001(2025)

Three-Dimensional Reconstruction Method of AA-HD Neural Radiation Field Based on PE-FNF and Point Cloud Prior

Tao Song1...2,*, Hongyao Tang1, Bin Xing2, Yichen Yang1, Yao Xiao1, Shengjie Lei1, Jianxu Wang1 and Zourong Long1 |Show fewer author(s)
Author Affiliations
  • 1Chongqing Key Laboratory of Optical Fiber Sensing and Photoelectric Detection, Chongqing University of Technology, Chongqing 400054, China
  • 2Chongqing Industrial Big Data Innovation Center Co., Ltd., Chongqing 400711, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (16)
    Equations (0)
    References (24)
    Tools
    Paper Information
    Topics
    Figures & Tables(16)
    NeRF network structure

    Fig. 1. NeRF network structure

    Download full size
    Overall framework of AA-HD neural radiation field algorithm based on PE-FNF and point cloud prior

    Fig. 2. Overall framework of AA-HD neural radiation field algorithm based on PE-FNF and point cloud prior

    Download full size
    Camera pose and sparse point clouds

    Fig. 3. Camera pose and sparse point clouds

    Download full size
    Lego scene of dense point cloud reconstruction

    Fig. 4. Lego scene of dense point cloud reconstruction

    Download full size
    Non-sensitive area diagram

    Fig. 5. Non-sensitive area diagram

    Download full size
    Screening diagram of position coding frequency number filter

    Fig. 6. Screening diagram of position coding frequency number filter

    Download full size
    Position coding frequency number filter

    Fig. 7. Position coding frequency number filter

    Download full size
    Point cloud prior process

    Fig. 8. Point cloud prior process

    Download full size
    Gated channel transformation structure

    Fig. 9. Gated channel transformation structure

    Download full size
    New view reconstruction effect of fern and Lego scenes under action of each module

    Fig. 10. New view reconstruction effect of fern and Lego scenes under action of each module

    Download full size
    New view reconstruction results of three methods in three scenarios of Realistic Synthetic 360° dataset

    Fig. 11. New view reconstruction results of three methods in three scenarios of Realistic Synthetic 360° dataset

    Download full size
    New view reconstruction results of three methods in five scenes of LLFF dataset

    Fig. 12. New view reconstruction results of three methods in five scenes of LLFF dataset

    Download full size

    • Table 1. Ablation experiments of new view reconstruction in Lego and fern scenes

      View table

      Table 1. Ablation experiments of new view reconstruction in Lego and fern scenes

      SceneAblation experimentPSNR /dBSSIMLPIPS
      LegoA26.12 （1.63%）0.9203 （1.02%）0.0250 （-6.05%）
      B27.33 （2.84%）0.9252 （1.03%）0.0433 （-2.07%）
      C29.16 （0.67%）0.9174 （0.03%）0.0612 （-0.05%）
      D32.68 （5.87%）0.9751 （10.86%）0.0162 （-8.69%）
      fernA27.36 （1.42%）0.9122 （0.01%）0.0371 （-12.43%）
      B28.64 （2.6%）0.9383 （3.01%）0.0731 （-2.07%）
      C26.92 （0.08%）0.8940 （-0.07%）0.0313 （-0.49%）
      D29.64 （5.87%）0.9561 （7.26%）0.0282 （-16.89%）

    • Table 2. Comparison of parameters of different methods on Realistic Synthetic 360° dataset

      View table

      Table 2. Comparison of parameters of different methods on Realistic Synthetic 360° dataset

      SceneNeRFMIP-NeRFOurs
      PSNR /dBSSIMLPIPSPSNR /dBSSIMLPIPSPSNR /dBSSIMLPIPS
      Average32.320.9310.05833.410.9550.04533.540.9590.042
      chair34.260.9570.02135.090.9610.01935.260.9740.017
      drum28.870.9320.06329.420.9640.04330.270.9720.041
      ship33.450.8740.12534.840.9290.08733.780.9160.089
      materials35.240.9380.04635.790.9680.04235.830.9710.038
      hotdog29.760.9520.03731.890.9550.03232.530.9620.027

    • Table 3. Parameter comparison of different methods on LLFF dataset

      View table

      Table 3. Parameter comparison of different methods on LLFF dataset

      SceneNeRFMIP-NeRFOurs
      PSNR /dBSSIMLPIPSPSNR /dBSSIMLPIPSPSNR /dBSSIMLPIPS
      Average27.020.8190.08330.180.8740.06831.440.9080.057
      flower29.380.8310.09732.190.8630.07833.270.9030.071
      fortress29.140.8220.06930.120.8740.06531.170.8970.052
      horn28.360.8490.05831.750.8770.04932.660.9220.044
      leaves22.840.7110.12827.490.8430.08929.420.8810.068
      trex25.360.8790.06529.370.9130.05730.690.9350.052

    • Table 4. Comparison of computational costs of different methods

      View table

      Table 4. Comparison of computational costs of different methods

      DatasetNeRFMIP-NeRFOurs
      PSNR /dBTrain-time /hPSNR /dBTrain-time /hPSNR /dBTrain-time /h
      Realistic Synthetic 360°32.324.2633.414.8633.543.86
      LLFF27.024.8730.185.0231.444.13
    Tools

    Get Citation

    Copy Citation Text

    Tao Song, Hongyao Tang, Bin Xing, Yichen Yang, Yao Xiao, Shengjie Lei, Jianxu Wang, Zourong Long. Three-Dimensional Reconstruction Method of AA-HD Neural Radiation Field Based on PE-FNF and Point Cloud Prior[J]. Acta Optica Sinica, 2025, 45(8): 0815001

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Set citation alerts for article
    Save article for my favorites
    Paper Information

    Category: Machine Vision

    Received: Dec. 23, 2024

    Accepted: Feb. 14, 2025

    Published Online: Apr. 27, 2025

    The Author Email: Song Tao (tsong@cqut.edu.cn)

    DOI:10.3788/AOS241920

    CSTR:32393.14.AOS241920

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology