1. Implement a CNN to detect diabetic retinopathy (DR) from retina images taken using fundus photography under a variety of imaging conditions
2. Improve generalization performance and reduce overfitting using image transformation and data augmentation techniques
3. Tune hyperparameters of the CNN model

Diabetic retinopathy (DR) is an eye condition that affects blood vessels in the retina. It can cause vision loss and blindness in people who have diabetes. Screening for DR allows earlier and more effective treatment options for millions of people.

The original dataset is hosted by Kaggle as part of a [code competition] hosted by the Asia Pacific Tele-Opthalmology Society (APTOS) in 2019.

The dataset consists of retina images images) and labels.

Our objective will be to build and train a CNN to predict whether or not to refer a patient for DR treatment using the binarized severity of DR in patients:

• no referral if {No DR, mild}
• referral if {moderate, severe, and proliferate DR}