• Acta Optica Sinica
  • Vol. 42, Issue 1, 0112005 (2022)
Jiangping Zhu1、2, Ruike Wang1, Zhijuan Duan1、2, Yijie Huang1, Guohuan He1, and Pei Zhou1、2、*
Author Affiliations
  • 1College of Computer Science, Sichuan University, Chengdu, Sichuan 610065, China
  • 2National Key Laboratory of Fundamental Science on Synthetic Vision, Sichuan University, Chengdu, Sichuan 610065, China
  • show less
    DOI: 10.3788/AOS202242.0112005 Cite this Article Set citation alerts
    Jiangping Zhu, Ruike Wang, Zhijuan Duan, Yijie Huang, Guohuan He, Pei Zhou. Three-Dimensional Face Modeling Based on Multi-Scale Attention Phase Unwrapping[J]. Acta Optica Sinica, 2022, 42(1): 0112005 Copy Citation Text show less
    Network architecture of MSAPUNet
    Fig. 1. Network architecture of MSAPUNet
    Calculation process in spatial attention module
    Fig. 2. Calculation process in spatial attention module
    Loss function curve
    Fig. 3. Loss function curve
    Acquisition system
    Fig. 4. Acquisition system
    Schematic diagram of dataset construction. (a) FACE dataset; (b) MASK dataset
    Fig. 5. Schematic diagram of dataset construction. (a) FACE dataset; (b) MASK dataset
    Phase map, point cloud model and error of point cloud model obtained by different methods in undersampling experiment. (a)--(e) Phase map; (f)--(j) point cloud model; (k)--(n) error of point cloud model
    Fig. 6. Phase map, point cloud model and error of point cloud model obtained by different methods in undersampling experiment. (a)--(e) Phase map; (f)--(j) point cloud model; (k)--(n) error of point cloud model
    Error maps obtained by different methods in undersampling experiment. (a) MSAPUNet; (b) U-Net; (c) QG algorithm; (d) BC algorithm
    Fig. 7. Error maps obtained by different methods in undersampling experiment. (a) MSAPUNet; (b) U-Net; (c) QG algorithm; (d) BC algorithm
    Phase maps, point cloud models and errors of point cloud model obtained by different methods in phase discontinuity experiment. (a)--(e) Phase map; (f)--(j) point cloud model; (k)--(n) error of point cloud model
    Fig. 8. Phase maps, point cloud models and errors of point cloud model obtained by different methods in phase discontinuity experiment. (a)--(e) Phase map; (f)--(j) point cloud model; (k)--(n) error of point cloud model
    Error maps obtained by different methods in phase discontinuity experiment. (a) MSAPUNet; (b) U-Net; (c) QG algorithm; (d) BC algorithm
    Fig. 9. Error maps obtained by different methods in phase discontinuity experiment. (a) MSAPUNet; (b) U-Net; (c) QG algorithm; (d) BC algorithm
    Experimental results of dynamic target. (a) Texture map; (b) wrapped phase; (c) phase generated by TPU algorithm; (d) phase generated by U-Net; (e) phase generated by MSAPUNet
    Fig. 10. Experimental results of dynamic target. (a) Texture map; (b) wrapped phase; (c) phase generated by TPU algorithm; (d) phase generated by U-Net; (e) phase generated by MSAPUNet
    The 60th column phase in dynamic target experiment
    Fig. 11. The 60th column phase in dynamic target experiment
    DatasetQGBCU-NetMSAPUNet
    MRMSE /radMSSIMMRMSE /radMSSIMMRMSE /radMSSIMMRMSE /radMSSIM
    FACE0.20480.75130.18880.73530.05040.97950.03870.9850
    MASK0.11120.77920.10310.80100.05510.96870.02730.9793
    Table 1. RMSE and SSIM of different algorithms in FACE dataset and MASK dataset
    MethodQGBCU-NetMSAPUNet
    Time>10 s>2 s30 ms40 ms
    Table 2. Efficiency of different methods
    IndexQGalgorithmBCalgorithmU-NetMSAPUNet
    MRMSE /rad0.33920.26880.01630.0135
    MSSIM0.69000.72760.99520.9976
    Table 3. RMSE and SSIM of different methods in undersampling experiment
    IndexQGalgorithmBCalgorithmU-NetMSAPUNet
    Averagedistance0.78160.70970.19560.0006
    Standarddeviation0.55690.59770.60300.0174
    Table 4. Errors of point cloud model obtained by different methods in undersampling experimentunit: mm
    IndexQGalgorithmBCalgorithmU-NetMSAPUNet
    MRMSE /rad0.13640.11800.05430.0291
    MSSIM0.71920.78880.96500.9754
    Table 5. RMSE and SSIM of different methods in phase discontinuity experiment
    IndexQGalgorithmBCalgorithmU-NetMSAPUNet
    Averagedistance1.77252.04200.10830.0065
    Standarddeviation1.10661.40770.63760.0915
    Table 6. Errors of point cloud model obtained by different methods in undersampling experimentunit: mm
    Jiangping Zhu, Ruike Wang, Zhijuan Duan, Yijie Huang, Guohuan He, Pei Zhou. Three-Dimensional Face Modeling Based on Multi-Scale Attention Phase Unwrapping[J]. Acta Optica Sinica, 2022, 42(1): 0112005
    Download Citation