We provide a detailed description of the environment settings required to replicate our project.
To reproduce our project, you need to install conda, cuda and cudnn firstly.
Install conda (we use 23.7.4 version)
Install cuda >= 11.8.0 (we use 11.8.0 version)
Install cudnn (we use 8.9.2 for cuda11.x version)
You can create and enter your own conda environment by: """ conda create -n 3dseg python=3.9.0 source activate 3dseg """
Alternatively, you can install the required conda environment from the 'environment.yaml' file by: """ conda create -f environment.yaml """ Notice that you need to replace 'prefix: /home/user/anaconda3/envs/3dseg' with your virtual environment path in the 'environment.yaml',and you can rename the created virtual environment.
Install pytorch, torchvision, torchaudio (it depends on your nvidia driver and your cuda version) We use torch==2.0.0, torchvision==0.15.1, torchaudio==2.0.1 for cuda 11.8. You can install by: """ pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118 """
You need to put the downloaded datasets in the ./datasets path by:
./datasets/
├── FLARE
├── OIMHS
├── SegTHOR
├── ...
The dataset is further divided into training sets, validation sets, and test sets in the following format:
path to the dataset/
├── imagesTr
├── labelsTr
├── imagesVal
├── labelsVal
├── imagesTs
├── original_labelTs
├── shapes.json
Where original_labelTs holds all raw unlabeled data, shapes.json is each sample and its corresponding shape, for example: "train_000.nii.gz": [512, 512, 110]. Currently our framework only supports data in .nii.gz format,
We currently provide run_gpu.sh to quickly train and test the model, and pretrained_test.sh to inference using the pre-trained weights provided in ./pretrained_models.