UNet for retina vessel segmentation.
We run the code on Ubuntu 18.04 LTS with a GTX 1080ti GPU.
Python (3.7.3) | Tensorflow (2.0.0) | CUDA (10.0) | cuDNN (7.6.0)
You can download the datasets from here. And then move the files into the ./data/ folder.
Because of my RAM limited, I firstly transform the train.mat file to .tfrecord files.
python preprocess.py
Then you can train the model initially.
CUDA_VISIBLE_DEVICES=0 python main.py -e train -c ./config/config_0.ini
When you have trained successfully, you can evaluate the model.
CUDA_VISIBLE_DEVICES=0 python main.py -e test -c ./config/config_0.ini
| Metric | Value |
|---|---|
| Area under ROC curve | 0.9791076715589634 |
| Area under PR curve | 0.9096035722442897 |
For threshold: 0.5
| Metric | Value |
|---|---|
| Jaccard similarity score | 0.6899763444446503 |
| F1 score (F-measure) | 0.8165514821704634 |
| Accuracy | 0.9557572337407614 |
| Precision | 0.8645987922457612 |
| Recall | 0.7735631845636364 |