How smrutiparhi/satellite-anomaly-detection-using-convolutional-autoencoders works
This repository delivers satellite-image novelty detection through a PyTorch model, a FastAPI service, a Streamlit upload UI, and a React dashboard. The main user path is dashboard upload → API inference → reconstruction/error score and visualizations, with optional threshold override and browser-saved scan summaries. Separate offline stages prepare data, train/calibrate the autoencoder, and evaluate it. The graph preserves the landing and login routes as distinct frontend capabilities. Source excerpts establish most runtime calls; the README documents the HTTP integration, while unsampled evaluation and login internals are represented conservatively.
Main entry points: detect.py Anomaly detector · App.tsx App routing · Dashboard.tsx Scan dashboard · train.py Training and calibration
Components 18
React application
- App routing
frontend/src/App.tsx - Landing experience
frontend/src/pages/Landing.tsx - Landing sections
frontend/src/components/landing - Login
frontend/src/pages/Login.tsx - Scan dashboard
frontend/src/pages/Dashboard.tsx - Image dropzone
frontend/src/components/dashboard/ImageDropzone.tsx - Browser scan history
frontend/src/pages/Dashboard.tsx
Inference interfaces
- FastAPI service
satellite_anomaly_detection/api.py - Streamlit uploader
satellite_anomaly_detection/app.py - Anomaly detector
satellite_anomaly_detection/src/detect.py - Calibrated checkpoint
satellite_anomaly_detection/models/best_autoencoder.pth
Model lifecycle
- Dataset partitions
satellite_anomaly_detection/src/dataset.py - Autoencoder architectures
satellite_anomaly_detection/src/model.py - Training and calibration
satellite_anomaly_detection/src/train.py - Held-out evaluation
satellite_anomaly_detection/src/evaluate.py - Evaluation reporting
satellite_anomaly_detection/src/report.py
Other components
- Visitor
- Dashboard user
Connections 20
- Visitor → App routing: opens app
frontend/src/App.tsx - App routing → Landing experience: routes to
frontend/src/App.tsx - App routing → Login: routes to
frontend/src/App.tsx - App routing → Scan dashboard: protects route
frontend/src/App.tsx - Landing experience → Landing sections: renders
frontend/src/pages/Landing.tsx - Dashboard user → Image dropzone: selects image
frontend/src/components/dashboard/ImageDropzone.tsx - Image dropzone → Scan dashboard: passes file
frontend/src/pages/Dashboard.tsx - Scan dashboard → FastAPI service: requests inference
README.md - FastAPI service → Anomaly detector: invokes prediction
satellite_anomaly_detection/api.py - Streamlit uploader → Anomaly detector: invokes prediction
satellite_anomaly_detection/app.py - Anomaly detector → Autoencoder architectures: reconstructs images
satellite_anomaly_detection/src/detect.py - Anomaly detector → Calibrated checkpoint: loads weights
satellite_anomaly_detection/src/detect.py - Anomaly detector → Dataset partitions: uses transform
satellite_anomaly_detection/src/detect.py - Scan dashboard → Browser scan history: stores summaries
README.md - Training and calibration → Dataset partitions: loads partitions
satellite_anomaly_detection/src/train.py - Training and calibration → Autoencoder architectures: trains architecture
satellite_anomaly_detection/src/train.py - Training and calibration → Calibrated checkpoint: writes calibrated model
satellite_anomaly_detection/src/train.py - Held-out evaluation → Dataset partitions: loads partitions
satellite_anomaly_detection/src/evaluate.py - Held-out evaluation → Anomaly detector: scores images
satellite_anomaly_detection/src/evaluate.py - Evaluation reporting → Anomaly detector: uses predictions
satellite_anomaly_detection/src/report.py