/3DUNet-Pytorch

3DUNet implemented with pytorch

Primary LanguagePython

3DUNet implemented with pytorch

Introduction

The repository is a 3DUNet implemented with pytorch, referring to this project. I have redesigned the code structure and used the model to perform liver and tumor segmentation on the lits2017 dataset.
paper: 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation

Requirements:

pytorch >= 1.1.0
torchvision
SimpleITK
Tensorboard
Scipy

Quickly Start

1) LITS2017 dataset preprocessing:

  1. Download dataset from google drive: Liver Tumor Segmentation Challenge.
    Or from my share: https://pan.baidu.com/s/1WgP2Ttxn_CV-yRT4UyqHWw Extraction code：hfl8
  2. Then you need decompress the data set and put the volume data and segmentation labels into different local folders, such as ./dataset/data and ./dataset/label
  3. Finally, you need to change the root path of the volume data and segmentation labels in preprocess/preprocess_LiTS.py, such as:
    row_dataset_path = './dataset/'  # path of origin dataset
    fixed_dataset_path = './fixed/'  # path of fixed(preprocessed) dataset
  1. Run python preprocess/preprocess_LiTS.py
    If nothing goes wrong, you can see the following files in the dir ./fixed
│  test_name_list.txt
│  train_name_list.txt
│  val_name_list.txt
│
├─data
│      volume-0.nii
│      volume-1.nii
│      volume-10.nii
│      ...
└─label
        segmentation-0.nii
        segmentation-1.nii
        segmentation-10.nii
        ...

2) Training 3DUNet

  1. Firstly, you should change the some parameters in config.py,especially, please set --dataset_path to ./fixed
    All parameters are commented in the file config.py.
  2. Secondely,run python train.py --save model_name
  3. Besides, you can observe the dice and loss during the training process in the browser through tensorboard --logdir ./output/model_name.

In addition, during the training process you will find that loading train data is time-consuming, you can use train_faster.py to train model. train_faster.py calls ./dataset/dataset2_lits.py, which will crop multiple training samples from an input sample to form a batch for quickly training.

3) Testing

run test.py (Please pay attention to path of trained model)

You can also read my Chinese introduction about this 3DUNet project here.
If you have any suggestions or questions, welcome to open an issue to communicate with me.