Sanro Health - Diabetic Retinopathy Classifier

Installation Steps

git clone https://github.com/davidwmcdevitt/SanroDR
%cd SanroDR
!pip install -r requirements.txt

Data Preparation

unzip [TRAINING_DATA.ZIP] -d data
unzip [TEST_DATA.ZIP] -d test
unzip [MESSIDOR2.ZIP] -d test

Replicate Training

python train.py [OPTIONS]

--data_dir : Specifies the directory where the training data is located. Default is data.
--num_epochs : Sets the number of training epochs. Default is 25.
--oversample : Enables oversampling of minority classes to balance the dataset.
--class_weights : Includes class weights to handle class imbalance during training.
--force_balance : Forces the dataset to have a 50/50 split between 'No DR' cases and 'any DR' cases, helping balance the training data.
--state_dict : Path to a saved model state dictionary for continuing training from a previously saved state.

Evaluate Model

python eval.py [OPTIONS]

--eval_set : Specifies evaluation dataset: ['kaggle' or 'messidor2']
--model_dict : Path to the model state dictionary file. 
--data_dir : Directory where the evaluation data is stored.