• Laser & Optoelectronics Progress
  • Vol. 59, Issue 16, 1611002 (2022)
Jiatao Liu1, Yaping Zhang1、*, and Yuwei Yang2
Author Affiliations
  • 1School of Information Science and Technology, Yunnan Normal University, Kunming 650500, Yunnan , China
  • 2Nantong Institute of Technology, Nantong 226000, Jiangsu , China
  • show less
    DOI: 10.3788/LOP202259.1611002 Cite this Article Set citation alerts
    Jiatao Liu, Yaping Zhang, Yuwei Yang. Efficient Monocular Image Depth Estimation Based on Transfer Learning[J]. Laser & Optoelectronics Progress, 2022, 59(16): 1611002 Copy Citation Text show less
    Model structure
    Fig. 1. Model structure
    Encoder structure and encoding process
    Fig. 2. Encoder structure and encoding process
    Decoder structure and decoding process
    Fig. 3. Decoder structure and decoding process
    AdaBins module structure and processing process
    Fig. 4. AdaBins module structure and processing process
    Prediction results of different models. (a) Input RGB images; (b) ground truth depth maps; (c) depth map predicted by model in literature [7]; (d) depth map predicted by model in literature [11]; (e) depth map predicted by proposed model
    Fig. 5. Prediction results of different models. (a) Input RGB images; (b) ground truth depth maps; (c) depth map predicted by model in literature [7]; (d) depth map predicted by model in literature [11]; (e) depth map predicted by proposed model
    Examples of proposed model generating a three-dimensional point cloud from a single RGB image
    Fig. 6. Examples of proposed model generating a three-dimensional point cloud from a single RGB image
    Ablation study of different loss functions. (a) Comparison of using Lbins or not; (b)(c)(d) comparison of using Lbins combined with different loss items and weight values
    Fig. 7. Ablation study of different loss functions. (a) Comparison of using Lbins or not; (b)(c)(d) comparison of using Lbins combined with different loss items and weight values
    Ablation study of different encoders
    Fig. 8. Ablation study of different encoders
    Ablation study of different initialization methods
    Fig. 9. Ablation study of different initialization methods
    Ablation study of AdaBins module
    Fig. 10. Ablation study of AdaBins module
    Modelδ1δ2δ3REL↓RMS↓log10Train cost /105
    Proposed model0.9020.9870.9980.1030.3560.0442.53
    Model in literature [50.7690.9500.9880.1580.641
    Model in literature[80.8280.9650.9920.1150.5090.051
    Model in literature[40.8950.9800.9960.1030.3900.04380
    Model in literature[70.9030.9840.9970.1030.3640.04412.67
    Model in literature[110.9040.9880.9980.1100.3570.045691.2
    Table 1. Quantitative comparison with prediction results of other models
    Jiatao Liu, Yaping Zhang, Yuwei Yang. Efficient Monocular Image Depth Estimation Based on Transfer Learning[J]. Laser & Optoelectronics Progress, 2022, 59(16): 1611002
    Download Citation