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 11, 1111004(2025)

A Deep Learning‑Based Method For Binocular Endoscopic Stereoscopic Imaging

Yuanjue Ma1, Jingyi Xu2,3, Jun Yan2,3、**, and Liang Li1,3、*
Author Affiliations
  • 1Department of Engineering Physics, Tsinghua University, Beijing 100084, China
  • 2School of Medicine, Tsinghua University, Beijing 100084, China
  • 3Institute for Precision Medicine, Tsinghua University, Beijing 100084, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (10)
    Equations (0)
    References (31)
    Tools
    Paper Information
    Topics
    Figures & Tables(10)
    Architecture of CroCo v2[25]

    Fig. 1. Architecture of CroCo v2[25]

    Download full size
    Simulation dataset. (a) Human heart model; (b) left image; (c) right image; (d) disparity ground truth of left image

    Fig. 2. Simulation dataset. (a) Human heart model; (b) left image; (c) right image; (d) disparity ground truth of left image

    Download full size
    Augmentation of SCARED training dataset. (a) Left image; (b) right image; (c) disparity ground truth of left image

    Fig. 3. Augmentation of SCARED training dataset. (a) Left image; (b) right image; (c) disparity ground truth of left image

    Download full size
    Test results of SCARED test dataset. (a) Left image; (b) disparity ground truth of left image; (c) output of pre-trained network; (d) output of network trained on the SCARED training dataset without pre-training; (e) output of network trained on the SCARED training dataset with pre-training

    Fig. 4. Test results of SCARED test dataset. (a) Left image; (b) disparity ground truth of left image; (c) output of pre-trained network; (d) output of network trained on the SCARED training dataset without pre-training; (e) output of network trained on the SCARED training dataset with pre-training

    Download full size
    Test results in scenarios with weak textures and varying illumination conditions. The red boxes highlight weak texture regions, the green boxes highlight reflective regions, and the white boxes highlight low-illumination regions. (a1)(a2) Left image; (b1)(b2) disparity ground truth of left image; (c1)(c2) output of the network

    Fig. 5. Test results in scenarios with weak textures and varying illumination conditions. The red boxes highlight weak texture regions, the green boxes highlight reflective regions, and the white boxes highlight low-illumination regions. (a1)(a2) Left image; (b1)(b2) disparity ground truth of left image; (c1)(c2) output of the network

    Download full size
    Test results of binocular endoscopic dataset. The red boxes highlight weak texture regions, the green boxes highlight reflective regions, and the white boxes highlight low-illumination regions. (a) Left image; (b) right image; (c) output of pre-trained network; (d) output of network trained on the SCARED training dataset without pre-training; (e) output of network trained on the SCARED training dataset with pre-training

    Fig. 6. Test results of binocular endoscopic dataset. The red boxes highlight weak texture regions, the green boxes highlight reflective regions, and the white boxes highlight low-illumination regions. (a) Left image; (b) right image; (c) output of pre-trained network; (d) output of network trained on the SCARED training dataset without pre-training; (e) output of network trained on the SCARED training dataset with pre-training

    Download full size

    • Table 1. Experimental environment configuration

      View table

      Table 1. Experimental environment configuration

      ItemModel
      GPUNVIDIA GeForce RTX 4090
      CPUIntel（R） Core（TM） i9-14900K @ 3.20 GHz
      Operating systemWindows 10 Pro for Workstations 64-bit
      CUDACUDA 11.3.1
      Open source frameworkPyTorch 1.11.0
      LanguagePython 3.9.19

    • Table 2. Quantitative evaluation results of the SCARED test dataset

      View table

      Table 2. Quantitative evaluation results of the SCARED test dataset

      Number of simulation dataset /pairsRMAE /pixelRKPE @ K=1.0 /%RKPE @ K=3.0 /%Depth RRMSE /mm
      05.38561.71923.5055.654
      8003.05952.53317.3654.658
      16002.81551.62016.6864.618
      32002.24047.99814.0644.468

    • Table 3. Comparison of the results for different methods based on SCARED test dataset

      View table

      Table 3. Comparison of the results for different methods based on SCARED test dataset

      MethodYearDepth RRMSE /mm
      STTR20215.047
      BDIS20226.403
      Yang20245.585
      MonoDiffusion20245.116
      EndoDAC20244.442
      Ours20254.468

    • Table 4. Quantitative evaluation results of binocular endoscopic dataset

      View table

      Table 4. Quantitative evaluation results of binocular endoscopic dataset

      ImageFig. 6 （c）Fig. 6 （d）Fig. 6 （e）
      Mean ηSSIM0.7630.8350.875
    Tools

    Get Citation

    Copy Citation Text

    Yuanjue Ma, Jingyi Xu, Jun Yan, Liang Li. A Deep Learning‑Based Method For Binocular Endoscopic Stereoscopic Imaging[J]. Acta Optica Sinica, 2025, 45(11): 1111004

    Download Citation

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

    Category: Imaging Systems

    Received: Jan. 14, 2025

    Accepted: Apr. 14, 2025

    Published Online: Jun. 24, 2025

    The Author Email: Jun Yan (yanjun1619@mail.tsinghua.edu.cn), Liang Li (lliang@tsinghua.edu.cn)

    DOI:10.3788/AOS250491

    CSTR:32393.14.AOS250491

    Topics

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