Diabetic Retinopathy Detection

How to run the code

Before running the script, make sure the requirements are met by installing from requirements.txt using
pip install -r requirements.txt

1. Configure the config.gin file in /configs

2. Use one of the following commands to perform training/evaluation/visualization. Make sure training is performed before evaluation or visualization. The experiment name which is used for training should be used for evaluation and visualization.

Note: Information on using checkpoints for evaluation and visualization can be found in the config file.

Train

python main.py --train True --dir_name "Experiment-name"

Evaluate

python main.py --eval True --dir_name "Experiment-name"

Visualize (Only for Diabetic Retinopathy)

python main.py --visualize True --dir_name "Experiment-name"

3. The files related to the experiment can be found under /experiments/Experiment-name

Directory structure

Diabetic-Retinopathy-Detection
|-- configs/
|   |-- config.gin
|-- evaluation/
|   |-- eval.py
|   |-- grad_cam.py
|   |-- guided_backprop.py
|   |-- metrics.py
|   |-- visualization.py
|-- input_pipeline/
|   |-- datasets.py
|   |-- preprocessing.py
|   |-- tf_records.py
|-- models/
|   |-- architectures.py
|   |-- layers.py
|-- utils/
|   |-- utils_misc.py
|   |-- utils_params.py
|-- experiments/
|-- .gitignore
|-- README.md
|-- requirements.txt
|-- datasets
|-- dataset_analysis.ipynb
|-- drbatch.sh
|-- drtune.sh
|-- main.py
|-- train.py
|-- wandb-tune.py

Results

• Hyperparameter Tuning:

• Evaluation Accuracy:

  • Graham Preprocessing: 76.70%
  • Green Channel Preprocessing: 79.61%

• Confusion Matrix:

• Deep Visualization: