by Zixu Zhao,Yueming Jin, Xiaojie Gao, Qi Dou, Pheng-Ann Heng.
- The Pytorch implementation for our MICCAI 2020 paper 'Learning Motion Flows for Semi-supervised Instrument Segmentation from Surgical Robotic Videos'.
- The Code contains two parts: motion learning (flow prediction and flow compensation) and semi-supervised segmentation.
-
We use the dataset from MICCAI 2017 Endoscopic vision challenge.
-
Please follow the data preprocessing steps in this repository.
-
The structure of data folder is arranged as follows:
(root folder)
├── data
| ├── cropped_train
| | ├── instrument_dataset_1
| | | ├── images
| | | ├── binary_masks
| | | ├── parts_masks
| | | ├── instrument_masks
| | ├── instrument_dataset_2
| | | ├── images
| | | ├── binary_masks
| | | ├── parts_masks
| | | ├── instrument_masks
| | ├── ......
- Check dependencies:
- Python 3.6
- pytorch 0.4.1+
- pytorch-ignite 0.2.0+
- tensorboardX
- albumentations
- opencv-python
- cupy (please check your CUDA version before install)
- tqdm
- Flow prediction & Flow compensation (./motion_learning/):
- To train the flow prediction model, run
$ python train_mp.py. - To train the flow compensation model, run
$ python train_mc.py. - Arguments for model training in
train.share in default settings. - To propagate frame-label pairs, run
$ bash propagate.sh.
- Semi-supervised segmentation (./segmentation/):
$ bash train.sh
Note: You may try other models from /Models/plane_model.py
If this repository is useful for your research, please cite:
@inproceedings{zhao2020learning,
title={Learning Motion Flows for Semi-supervised Instrument Segmentation from Robotic Surgical Video},
author={Zhao, Zixu and Jin, Yueming and Gao, Xiaojie and Dou, Qi and Heng, Pheng-Ann},
booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
pages={679--689},
year={2020},
organization={Springer}
}
For further question about the code or paper, please contact 'zxzhao@cse.cuhk.edu.hk'