Evaluation of Robustness of Disc/Cup Segmentation in Different Fundus Imaging Conditions
This repository contains the code and resources for the ARVO abstract from OPHAI Lab (supervised by Dr. Mohammad Eslami and Dr. Tobias Elze). Instructions for reproducing results and using this repository are below. All files for training, evaluating, and plotting are located in the parent directory. The code for all models we used are in /models. The code for data loading, metrics, and loss functions can be found in /utils.
Setup
Install all required packages using pip install -r requirements.txt. The code was tested using Python 3.7.9.
Training a model
To use one of the models in the /models directory, pass in the filename of the file with the code for that model along with the other required arguments to main_train.py.
usage: main_train.py [-h] --model-name
{cenet,deeplabv3plus,unetpp}
--name-csv-train NAME_CSV_TRAIN --data-dir DATA_DIR
--path-save PATH_SAVE --img-size IMG_SIZE
[--batch-size BATCH_SIZE] [--binary BINARY]
Train model.
optional arguments:
-h, --help show this help message and exit
--model-name {cenet,deeplabv3plus,unetpp}
Name of model to train.
--name-csv-train NAME_CSV_TRAIN
Name of the CSV file with training dataset
information.
--data-dir DATA_DIR Path to the folder with the CSV files and image
subfolders.
--path-save PATH_SAVE
Path to the folder where model will be saved.
--img-size IMG_SIZE Size to which the images should be reshaped (one
number, i.e. 256 or 512).
--batch-size BATCH_SIZE
Batch size for the model during training.
--binary BINARY Whether the segmentation masks are binary (True) or
multi-class (False).
Example usage: python main_train.py --model-name unetpp --name-csv-train train.csv --data-dir /path/to/data --path-save /path/to/save/model --img-size 256 --batch-size 8 --binary True
Evaluating a model
To save the predictions of a saved model on a testing dataset and calculate the metrics (code in metrics.py), use main_test.py.
usage: main_test.py [-h]
[--model-name {cenet,deeplabv3plus,unetpp}]
--name-csv-test NAME_CSV_TEST --data-dir DATA_DIR
--path-model PATH_MODEL --img-size IMG_SIZE
--path-save-results PATH_SAVE_RESULTS
[--save-masks SAVE_MASKS] [--binary BINARY]
Test model.
optional arguments:
-h, --help show this help message and exit
--model-name {cenet,deeplabv3plus,unetpp}
Name of model to train.
--name-csv-test NAME_CSV_TEST
Name of the CSV file with testing dataset information.
--data-dir DATA_DIR Path to the folder with the CSV files and image
subfolders.
--path-model PATH_MODEL
Path to the saved model.
--img-size IMG_SIZE Size to which the images should be reshaped (one
number, i.e. 256 or 512).
--path-save-results PATH_SAVE_RESULTS
Path to the folder where results and predictions will
be saved.
--save-masks SAVE_MASKS
Whether you want to save all predicted masks for test
set or not.
--binary BINARY Whether the segmentation masks are binary (True) or
multi-class (False).
Example usage: python main_test.py --model-name unetpp --name-csv-test test.csv --data-dir /path/to/data --path-model /path/to/saved/model --img-size 256 --path-save-results /path/to/save/results --save-masks True --binary True
Figures and statistics
To recreate the figures in the abstract, run main_create_summary_results.py and then run main_plots.py. To obtain the data for the metrics table reported in the abstract, run main_stats.py. You may need to replace the paths to the results files and the desired summary files in the code.
Example usage:
python main_create_summary_results.py
python main_plots.py
python main_stats.py