Large Vessel Mask Generation Script

This script generates large vessel masks for eyeflow measures in a dataset or a single measure folder. It also provides an option to revert previously generated masks.

Requirements

Python 3.12

CUDA 12.x (for GPU execution)

cuDNN compatible with CUDA and ONNX Runtime GPU

Python packages (see requirements.txt)

pip install -r requirements.txt

Dataset structure

Excpected dataset structure:

dataset/
├─ YYYY-MM-DD/
│  ├─ measure1/
│  ├─ optional_subfolder/
│  │  ├─ measure2/
│  └─ ...
├─ YYYY-MM-DD/
│  ├─ measure1/
│  └─ ...
└─ ...

Script Usage

1. Single Folder Mode (default)

Run the script on a single measure folder:

python script.py path/to/measure_folder

Revert masks:

python script.py path/to/measure_folder --revert

2. Dataset Mode

Run the script on all measure folders in a dataset:

python script.py --dataset path/to/dataset

Revert masks for the entire dataset:

python script.py --dataset path/to/dataset --revert

Result

For each folder, new masks are created as forced Masks : eyeflow/forceMaskArtery.png and eyeflow/forceMaskVein.png. During the next eyeflow processing, if the paramater ForcedMasks in the input_EF_params.json equals 1 or -1, they will be chosen as masks. If forced Masks already exist, they will be renamed eyeflow/forceMaskArtery_full.png and eyeflow/forceMaskVein_full.png

When reverting, original forced masks will retrieve their names, and large masks will be renamed eyeflow/forceMaskArtery_large.png and eyeflow/forceMaskVein_large.png

Notes

The script automatically downloads the ONNX model for optic disc detection if it is not present.

By default, the script runs in single-folder mode if a positional folder argument is given.

Do not specify both --folder and --dataset at the same time; the script will throw an error.

The --revert option removes previously generated masks instead of generating new ones.