Automatic Tool Annotation for CATARACT Surgery: Masters Work

All information about tasks, goals, data and etc. see on competition page https://cataracts.grand-challenge.org/

Local deployment

$ git clone https://github.com/evgeniivas/Masters-Work.git
$ cd Masters-Work
$ pip3 install -r requirements.txt

Data

Download and place CATARACTS competition data to the learning/data/ folder, as a result in data folder you will have 3 subfolders: train, test, train_labels.

Launch scripts

All scripts for making classification models and data processing can be run from /learning directory (see params to each script with --help flag) :

1. Run /preparation/extractor.py script to prepare data for learning, set the params;
2. To learn models run learning.py script, set the params;
   • To run training on cluster (distributed systems): set --distributed parameter;
3. Make predictions to train\test datasets: run predict.py script, set the params;
4. Make baseline model (ExtraTreesClassifier trained using HOG descriptors) and predict on test dataset:
   run baseline.py script, set the params;

Convert to ONNX format:

From /learning folder run transfer bash script:

$ ./model/transfer.sh

As a result in each model folder appear file: model.onnx