A CNN classification model that achieves 74% accuracy on APTOS 2019 Blindness Detection. The data used are from https://www.kaggle.com/c/aptos2019-blindness-detection. It contains large images in high resolution. Clinicians rated the presence of diabetic retinopathy in each image between 0 and 4, according to the following scale:
- No DR
- Mild
- Moderate
- Severe
- Proliferative DR
Images are decreased to 50x50 in size, since using the large high-resolution images would have slowed down the training significant.