Self-Supervised Depth Estimation in Laparoscopic Image using 3D Geometric Consistency (MICCAI 2022)

By Baoru Huang, Jian-Qing Zheng, Anh Nguyen, Chi Xu, Ioannis Gkouzionis, Kunal Vyas, David Tuch, Stamatia Giannarou, Daniel S. Elson

Contents

1. Requirements
2. Training&Testing
3. Notes

Requirements

1. Python version 3.8.5
2. Pytorch version: pytorch==1.6.0 torchvision==0.8.2 torchaudio==0.7.0 cudatoolkit=10.2.89
3. pytorch3d version 0.3.0
4. Cuda version 10.2

Training & Testing

1. We train M3Depth on SCARED

   • We need to extract images from the videos and rectify the stereo image pairs with the given camera parameters (endoscope_calibration.yaml).
   • Only keyframes were together with accurate depth maps.
   • Sort the absolute image root and save them in $splits/Endovis_origin/train_files.txt.
   • Change the data directory to the folder of data.

2. Train M3Depth:

   • cd $M3Depth
   • python main.py --mode train

3. Test M3Depth:

   • cd $M3Depth
   • python main.py --mode test

4. Results

Citing

If you find our paper useful in your research, please consider citing:

    @inproceedings{huang2022self,
      title={Self-supervised Depth Estimation in Laparoscopic Image Using 3D Geometric Consistency},
      author={Huang, Baoru and Zheng, Jian-Qing and Nguyen, Anh and Xu, Chi and Gkouzionis, Ioannis and Vyas, Kunal and Tuch, David and Giannarou, Stamatia and Elson, Daniel S},
      booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
      pages={13--22},
      year={2022},
      organization={Springer}
    }

License

MIT License

Acknowledgement

1. This repo used a lot of source code from monodepth and monodepth2

2. This work was supported by the UK National Institute for Health Research (NIHR) Invention for Innovation Award NIHR200035, the Cancer Research UK Imperial Centre, the Royal Society (UF140290) and the NIHR Imperial Biomedical Research Centre.