How rudrasalokhe/dr-binary-classifier-mobilenet works
RetinaScope is a Flask retinal-image screening app. A user uploads a fundus image in the web interface; the browser submits it to the prediction route, which validates and decodes it, preprocesses it, runs CNN inference, and returns class confidences for display. The graph includes the documented Grad-CAM capability, but no implementation or connection for it appears in the supplied source. The README describes additional preprocessing details that differ from the sampled implementation, so the graph follows the code where available.
Main entry points: app.py Flask routes · script.js Upload and results UI · templates Web pages
Components 14
Web interface
- Web pages
templates - Upload and results UI
static/script.js - Confidence display
static/script.js
Prediction backend
- Flask routes
app.py - Upload validation
app.py - Class confidences
app.py
Image analysis
- Image decoding
app.py - Fundus preprocessing
app.py - FOV and illumination
app.py - Contrast and features
app.py - Stack and normalize
app.py - Keras CNN
Interpretability
- Grad-CAM support
Other components
- Image uploader
Connections 14
- Image uploader → Web pages: opens
README.md - Image uploader → Upload and results UI: uploads image
README.md - Flask routes → Web pages: renders pages
app.py - Upload and results UI → Flask routes: posts image
static/script.js - Flask routes → Upload validation: validates upload
app.py - Flask routes → Image decoding: decodes image
app.py - Flask routes → Fundus preprocessing: preprocesses image
app.py - Fundus preprocessing → FOV and illumination: masks and corrects
app.py - Fundus preprocessing → Contrast and features: enhances features
app.py - Fundus preprocessing → Stack and normalize: stacks and resizes
app.py - Flask routes → Keras CNN: runs inference
app.py - Flask routes → Class confidences: builds response
app.py - Upload and results UI → Confidence display: renders confidences
static/script.js - Grad-CAM support → Keras CNN: interprets model