Opto-Electronic Engineering, Volume. 50, Issue 4, 220246(2023)

STransMNet: a stereo matching method with swin transformer fusion

Gaoping Wang1... Xun Li1,2,*, Xuefang Jia1, Zhewen Li1 and Wenjie Wang1 |Show fewer author(s)
Author Affiliations
  • 1School of Electronics and Information, Xi'an Polytechnic University, Xi'an, Shaanxi 710600, China
  • 2Xi'an Polytechnic University Branch of Shaanxi Artificial Intelligence Joint Laboratory, Xi'an, Shaanxi 710600, China
  • show less
    Figures & Tables(11)
    The network structure of STTR-light
    (a) The network structure of STransMNet; (b) The structure of extractor
    Euclidean distance between the pixel features on the left image. (a) There is a feature differentiation loss; (b) No a feature differentiation loss
    Disparity map estimated by different methods on the Sceneflow datasets
    Disparity map estimated by different methods on the KITTI datasets
    • Table 1. Ablation study

      View table
      View in Article

      Table 1. Ablation study

      实验基于Swin Transformer模块相关运算特征差异化损失3 px error / % ↓EPE Occ IOU ↑
      第1组1.680.560.94
      第2组1.360.480.96
      第3组1.400.510.95
      第4组1.030.420.97
    • Table 2. Experimental results of different loss weights

      View table
      View in Article

      Table 2. Experimental results of different loss weights

      Ld1,rLd1,fLrrLbe,fLdiff3 px error /%EPE Occ IOU ↑
      0.20.20.20.20.20.850.410.97
      0.20.20.20.10.30.930.510.84
      0.30.30.10.10.20.890.430.85
      0.30.40.10.10.10.840.390.96
      0.40.30.10.10.10.870.400.91
    • Table 3. Test results Ⅰ of model generalization performance

      View table
      View in Article

      Table 3. Test results Ⅰ of model generalization performance

      模型MPI SintelKITTI
      3 px error /% ↓EPE ↓Occ IOU ↑3 px error /% ↓EPE ↓Occ IOU ↑
      PSMNet[10]6.813.31N/A27.796.56N/A
      AANet[11]5.911.89N/A12.421.99N/A
      STTR-light[19]5.822.950.697.21.560.95
      STTR[19]5.753.010.866.741.500.98
      本文算法5.232.780.846.511.440.97
    • Table 4. Test results Ⅱ of model generalization performance

      View table
      View in Article

      Table 4. Test results Ⅱ of model generalization performance

      模型MiddleburySCARED
      3 px error /% ↓EPE ↓Occ IOU ↑3 px error /% ↓EPE ↓Occ IOU ↑
      PSMNet[10]12.963.05N/AOOMOOMN/A
      AANet[11]12.802.19N/A6.391.36N/A
      STTR-light[19]5.362.050.763.301.190.89
      STTR[19]6.192.330.953.691.570.96
      本文算法6.242.120.963.151.260.94
    • Table 5. Comparative experiments

      View table
      View in Article

      Table 5. Comparative experiments

      模型SceneflowKITTI
      3 px error / % ↓EPE ↓Occ IOU ↑3 px error / % ↓EPE ↓Occ IOU ↑
      PSMNet[10]3.941.11N/A1.250.57N/A
      AANet[11]3.890.82N/A1.930.64N/A
      STTR-light[19]1.840.560.981.680.560.94
      STTR[19]1.430.480.911.120.440.97
      本文算法1.030.420.970.840.390.96
    • Table 6. Comparison of model operation efficiency

      View table
      View in Article

      Table 6. Comparison of model operation efficiency

      模型Params ↓ / MFLOPs ↓ / GMemory ↓ / GRuntime ↓ / s
      PSMNet[10]5.22613.904.080.63
      AANet[11]3.68119.641.630.09
      STTR-light[19]2.33110.210.430.65
      STTR[19]2.51510.931.230.67
      本文算法27.85136.112.900.73
    Tools

    Get Citation

    Copy Citation Text

    Gaoping Wang, Xun Li, Xuefang Jia, Zhewen Li, Wenjie Wang. STransMNet: a stereo matching method with swin transformer fusion[J]. Opto-Electronic Engineering, 2023, 50(4): 220246

    Download Citation

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

    Category: Article

    Received: Oct. 8, 2022

    Accepted: Jan. 19, 2023

    Published Online: Jun. 15, 2023

    The Author Email: Li Xun (lixun@xpu.edu.cn)

    DOI:10.12086/oee.2023.220246

    Topics