This Project was conceived during the lab course work "Statistical Signal Processing/Deep Learning" at the University of Stuttgart.

Diabetic Retinopathy Detection

Poster: Deep Learning based Diabetic Retinopathy Detection

Results

1. Input Pipeline
2. Model Architecture
3. Data Augmentation
   • Rotation
   • Horizontal Flip
   • Zoom
   • Balanced Dataset
4. Metrics
   • Confusion Matrix
5. Training and evaluation
6. Hyperparamter Optimization
   • Optimizer
   • Number of epochs
   • Neurons in Dense layer
7. Deep Visualization
   • Grad-CAM
   • Grad-CAM with Guided Backpropagation

Results obtained so far:

Accuracy Test Data Set : 72.75 %

1. Augmented Images

2. Model Architecture

3. Metrics

4. Grad-CAM + Guided Backpropagation Output