• Laser & Optoelectronics Progress
  • Vol. 58, Issue 14, 1410010 (2021)
Yuzheng Zhu, Yaping Zhang*, and Qiaosheng Feng
Author Affiliations
  • School of Information Science and Technology, Yunnan Normal University, Kunming, Yunnan 650500, China
  • show less
    DOI: 10.3788/LOP202158.1410010 Cite this Article Set citation alerts
    Yuzheng Zhu, Yaping Zhang, Qiaosheng Feng. Colorful 3D Reconstruction from Single Image Based on Deep Learning[J]. Laser & Optoelectronics Progress, 2021, 58(14): 1410010 Copy Citation Text show less
    Model structure
    Fig. 1. Model structure
    Reconstruction network of 3D shape
    Fig. 2. Reconstruction network of 3D shape
    Encoder structure added to CBAM module
    Fig. 3. Encoder structure added to CBAM module
    Color reconstruction network
    Fig. 4. Color reconstruction network
    SoftRas rendering structure
    Fig. 5. SoftRas rendering structure
    Comparison of predictions of different 3D models. (a) Original images; (b) 3D-R2N2; (c) PSGN; (d) OccNet; (e) SoftRas_O; (f) designed model
    Fig. 6. Comparison of predictions of different 3D models. (a) Original images; (b) 3D-R2N2; (c) PSGN; (d) OccNet; (e) SoftRas_O; (f) designed model
    Reconstruction results of different color 3D models. (a1) (a2) Original images; (b1) (b2) results of reconstruction
    Fig. 7. Reconstruction results of different color 3D models. (a1) (a2) Original images; (b1) (b2) results of reconstruction
    Attention mechanismIOUSSIMMSE
    SoftRas_O0.5760.5480.175
    CBAM_S0.5890.5790.161
    SENet0.5930.5630.186
    CBAM0.5930.5780.163
    Table 1. Comparison of indicators of different attention mechanism
    CategoryIOU
    SoftRas_OCBAM 1-2CBAM 3-4CBAM 1-4
    Airplane0.6030.6460.6480.648
    Bench0.4440.4610.4560.454
    Cabinet0.6160.6290.6130.633
    Car0.6720.6650.6690.675
    Chair0.4810.4920.4780.493
    Display0.5560.5710.5540.563
    Lamp0.4440.4630.4600.459
    Loudspeaker0.6150.6310.6350.636
    Rifle0.6430.6770.6830.674
    Sofa0.6190.6180.6280.630
    Table0.4530.4870.4760.470
    Telephone0.7410.7440.7760.766
    Vessel0.5890.6110.6100.612
    Mean0.5755080.591920.591230.59330
    Table 2. Comparison of IOU indicators of different CBAMs and SoftRas_O
    CategoryIOU
    3D-R2N2OccNetSoftRas_OOur model
    Airplane0.4260.5470.6030.648
    Bench0.3730.4520.4440.454
    Cabinet0.6760.7320.6160.633
    Car0.6610.7310.6720.675
    Chair0.4390.5020.4810.493
    Display0.4400.4790.5560.563
    Lamp0.2810.3700.4440.459
    Loudspeaker0.6110.6530.6150.636
    Rifle0.3750.4580.6430.674
    Sofa0.6260.6710.6190.630
    Table0.4200.5060.4530.470
    Telephone0.6110.7090.7410.766
    Vessel0.4820.5210.5890.612
    Mean0.4930.5640.5710.593
    Table 3. Comparison of IOU indicators of different models
    CategorySSIMMSE
    SoftRas_OOur modelSoftRas_OOur model
    Airplane0.6270.6100.0840.090
    Bench0.4750.4930.1860.183
    Cabinet0.5650.5870.1880.182
    Car0.5840.5920.1970.184
    Chair0.4420.4860.2400.222
    Display0.4530.4880.2640.250
    Lamp0.5370.7490.1560.068
    Loudspeaker0.5690.5910.2180.219
    Rifle0.5920.5920.0810.070
    Sofa0.4890.4940.2300.228
    Table0.5370.5480.1980.197
    Telephone0.6540.6730.1360.133
    Vessel0.6000.6170.0960.098
    Mean0.5480.5780.1750.163
    Table 4. Comparison results of SSIM and MSE between designed model and SoftRas_O
    Yuzheng Zhu, Yaping Zhang, Qiaosheng Feng. Colorful 3D Reconstruction from Single Image Based on Deep Learning[J]. Laser & Optoelectronics Progress, 2021, 58(14): 1410010
    Download Citation